fix(mcp): fix tool-docs generator and regenerate artifact

Fixes all reviewer-blocking and minor issues from the initial review of
the feat/404-tool-docs-artifact implementation.

BLOCKING fixes:
- Replace the naïve Annotated[...] regex with a bracket-aware character
  scan (_unwrap_annotated) so types like Dict[str, Any] | None and
  List[str] | None are extracted correctly instead of being truncated or
  emitting internal validator details (BeforeValidator, AliasChoices, etc.)
- Escape unescaped | characters in table cells (_escape_table_cell) so
  union types like list[str] | str | None render correctly in GitHub
  Flavoured Markdown tables instead of creating spurious extra columns

Minor fixes:
- Fix _extract_examples_section to detect the next top-level section
  (Raises:, Returns:, etc.) and stop there; previously the Raises: block
  was fenced inside the Python code example
- Add Description column data: _format_args_section now returns a parsed
  dict of arg descriptions (_parse_args_block) which populates the table
  Description column; fix indent-detection so the first arg line is not
  stripped of its leading spaces by .strip()
- Add deduplication: drop the docstring first line when normalised it is
  identical to the decorator description
- Demote ### sub-headers in docstring bodies to #### so they don't
  pollute GitHub's outline sidebar at the same level as tool headings
- Fix TOC anchor collision: the `search` tool anchor would resolve to the
  same #search as the ## Search group header; _tool_anchor detects
  collisions and appends -tool; an HTML <a id> anchor is injected in
  render_tool_section so the link resolves correctly
- Add comment explaining why ui_sdk.py is excluded from TOOL_FILES
- Format script with ruff (was failing CI ruff format --check gate)

Co-Authored-By: Claude <noreply@anthropic.com>
Signed-off-by: Drew Cain <groksrc@gmail.com>

.agents/plugins/marketplace.json

@@ -0,0 +1,20 @@
+{
+  "name": "basic-memory-local",
+  "interface": {
+    "displayName": "Basic Memory Local"
+  },
+  "plugins": [
+    {
+      "name": "codex",
+      "source": {
+        "source": "local",
+        "path": "./plugins/codex"
+      },
+      "policy": {
+        "installation": "AVAILABLE",
+        "authentication": "ON_INSTALL"
+      },
+      "category": "Developer Tools"
+    }
+  ]
+}

.agents/skills/adversarial-review/SKILL.md

@@ -0,0 +1,159 @@
+---
+name: adversarial-review
+description: Cross-vendor adversarial code review of the current branch. Two different model families (Claude + Codex/GPT) review the diff independently, then try to refute each other's findings; survivors are reported by confidence. Runs from either Claude Code or Codex. Use when the user asks for an adversarial review, a cross-model / second-opinion review, or wants high-confidence findings before merging. Report-only — never auto-applies fixes.
+license: MIT
+---
+
+# Adversarial code review
+
+Two reviewers from **different model families** — **Claude** and **Codex/GPT** — review the
+same diff independently, then each tries to **refute** the other's findings. A finding's
+confidence comes from whether it survives that cross-examination. This kills the two failure
+modes of solo LLM review: self-ratification (a model won't critique its own work) and
+confident false positives.
+
+## You are the orchestrator — and one of the two reviewers
+
+This skill runs from **either** Claude Code **or** Codex. First, **identify which model
+family you are** (Claude or Codex/GPT). Then:
+
+- **You** are reviewer #1. You review **natively**, in this session, using your own tools.
+- **The other family** is reviewer #2. You invoke it as a **subprocess CLI** for an
+  independent pass: a fresh process, no shared context — that independence is the point.
+
+The CLI for "the other model":
+
+| If you are… | Invoke the other via… |
+|-------------|------------------------|
+| **Claude**  | `codex exec` (GPT)     |
+| **Codex**   | `claude -p` (Claude)   |
+
+Everything else in the flow is symmetric. Resolve the `prompts/` and `schemas/` paths
+below relative to **this skill's own directory** (where this SKILL.md lives).
+
+## Inputs
+
+Two independent, optional inputs:
+
+- `BASE` — the ref to diff against. Default `main`.
+- `SCOPE` — a pathspec to narrow the review (e.g. `src/basic_memory`). Default: none (whole diff).
+
+These are separate: a ref and a pathspec are not interchangeable. Build the **canonical diff
+command** once in preflight and reuse it everywhere below — never re-spell the diff inline
+(the scattered, inconsistent spelling is what broke earlier). Build it as an **argv array**,
+not a string, so a `$SCOPE` containing spaces or glob characters survives intact:
+
+```bash
+BASE="${BASE:-main}"
+DIFF=(git diff "$BASE...HEAD")          # argv array — never a scalar string
+[ -n "$SCOPE" ] && DIFF+=(-- "$SCOPE")  # pathspec stays one argument even with spaces
+DIFF_STR=$(printf '%q ' "${DIFF[@]}")   # shell-quoted rendering, for embedding in a prompt
+```
+
+To **run** it, use `"${DIFF[@]}"` (quoted, no word-splitting). To **embed** it as text inside
+a subprocess prompt, use `$DIFF_STR`.
+
+## Preflight
+
+0. Set `SKILL_DIR` to the directory this SKILL.md lives in. Canonical location is
+   `.agents/skills/adversarial-review` (the shared agent-skills store); Claude Code reaches it
+   via the `.claude/skills/adversarial-review` symlink, Codex via its own skills path. The
+   `prompts/` and `schemas/` subdirs are siblings of this file in every case.
+1. Confirm the *other* model's CLI is on PATH (`codex` if you're Claude, `claude` if you're
+   Codex). If it's missing, tell the user the panel falls back to single-model (which loses
+   the cross-vendor benefit) and ask whether to proceed or stop.
+2. Run `"${DIFF[@]}"`. If it prints nothing, report "nothing to review against $BASE"
+   (mention `$SCOPE` if set) and stop.
+3. `RUN=$(mktemp -d)` — scratch dir for the other model's output. Transient, never committed.
+   No persisted artifacts, no state file.
+
+## Phase 0 — Deterministic gates (before the models)
+
+Models are statistically blind to negation ("never do X"). Enforce mechanical house rules
+with tools, not prompts, and treat hits as high-confidence facts (reported separately from
+model findings):
+
+- `just lint` and `just typecheck` if the diff touches `src/`.
+- Grep the diff for catchable house-rule violations: `getattr(.*,.*,` defaults, bare
+  `except:` / `except Exception: pass`, function-scope imports.
+
+## Phase 1 — Independent review (you + the other model, concurrently)
+
+Both reviewers get the same brief: `prompts/review.md` + the repo's `CLAUDE.md` house rules,
+reviewing the diff from `"${DIFF[@]}"`. Both emit findings matching `schemas/findings.schema.json`.
+
+**Your native pass:** review as yourself, following `prompts/review.md`. Hold your findings
+as that JSON shape.
+
+**The other model's pass** — run, from the repo root, the row that matches you:
+
+Always redirect `codex` stdin from `/dev/null` — if stdin is a pipe (e.g. the call gets
+backgrounded), `codex exec` blocks "Reading additional input from stdin..." and fails.
+
+```bash
+# You are Claude → run Codex:
+codex exec -s read-only \
+  --output-schema "$SKILL_DIR/schemas/findings.schema.json" \
+  -o "$RUN/other_findings.json" \
+  "$(cat "$SKILL_DIR/prompts/review.md")
+
+Review the diff: $DIFF_STR" </dev/null
+
+# You are Codex → run Claude (read-only via plan mode; parse the JSON block it returns):
+claude -p --permission-mode plan --output-format json \
+  "$(cat "$SKILL_DIR/prompts/review.md")
+
+Review the diff: $DIFF_STR
+Return ONLY a JSON object matching this schema:
+$(cat "$SKILL_DIR/schemas/findings.schema.json")" </dev/null > "$RUN/other_raw.json"
+# claude --output-format json output shape varies by CLI version: it may be a JSON ARRAY
+# of event objects, OR a single result object. Normalize before reading: if it's an array,
+# take the element with type=='result'; otherwise use the object as-is. Then read its
+# .result string, strip the ```json fence if present, and parse that.
+# (Verified empirically: the CLI in this environment emits the array form.)
+```
+
+> Runtime note for Codex orchestrating: `claude -p` needs network access, which Codex's
+> default sandbox blocks. Run it from a Codex session whose project is trusted with network
+> allowed (or approve the `claude` call when prompted). Keep Codex's own sandbox on — do not
+> bypass it just to reach the network.
+
+Tag each finding with its origin (`claude` / `codex`).
+
+## Phase 2 — Cross-refute
+
+Each model tries to refute the *other's* findings, per `prompts/refute.md`
+(verdicts match `schemas/verdicts.schema.json`).
+
+- **You** refute the other model's findings natively.
+- **The other model** refutes *your* findings — invoke it again the same way (swap
+  `prompts/review.md` for `prompts/refute.md`, append your findings JSON **and `$DIFF_STR`**
+  so it judges against the right base and scope, and for Codex use
+  `--output-schema "$SKILL_DIR/schemas/verdicts.schema.json"`).
+
+Match verdicts to findings by `id`.
+
+## Phase 3 — Synthesize and report (no auto-fix)
+
+Merge, dedupe (same file + overlapping lines + same root cause = one finding), assign
+confidence from provenance:
+
+- **High** — both models raised it independently, OR one raised it and the other upheld it.
+- **Medium** — one raised it; the other could not refute it but did not independently find it.
+- **Low / contested** — one raised it and the other **refuted** it. Keep it, show both sides,
+  let the human judge. Never silently drop a contested finding.
+- Deterministic-gate hits are reported as facts, separate from the model panel.
+
+Rank by `severity × confidence`. Present a compact table: `severity | confidence | file:line
+| claim | found-by / upheld-or-refuted-by`. Expand the high-confidence ones with `why` and
+any suggested fix.
+
+End by asking which findings, if any, to fix. **Do not edit code until the user picks.**
+Convergence between the models is not correctness — your job is to surface a ranked,
+cross-examined list, not to declare the branch clean.
+
+## Deliberately NOT done
+
+- No loop-until-both-agree (models converge by going silent, not by being right).
+- No persisted artifacts / state machine — the scratch dir is thrown away.
+- No auto-applying fixes.

.agents/skills/adversarial-review/prompts/refute.md

@@ -0,0 +1,22 @@
+# Refute the other reviewer
+
+A different reviewer (a different model family) produced the findings below against the
+same diff under review (the exact `git diff` command is provided with the findings). Your
+job is to try to **refute** each one.
+
+Default to skepticism: assume a finding is wrong until the code proves it right. A finding
+that survives a genuine attempt to disprove it is worth far more than one nobody checked.
+
+For each finding, read the actual code it points at and return a verdict:
+
+- **refuted** — the claim is wrong, the code does not do what the finding says, the case
+  cannot occur, or it is pure style with no correctness impact. Cite the specific code or
+  fact that disproves it.
+- **upheld** — you tried to refute it and could not; the finding is real.
+- **partial** — the underlying issue is real but the finding mis-states the severity or
+  scope. Explain, and set `corrected_severity` if the severity should change.
+
+Do not be agreeable for its own sake, and do not refute for its own sake. Follow the code.
+
+Return ONLY the structured verdicts object conforming to the provided schema. Every
+verdict's `id` must match the `id` of the finding it judges.

.agents/skills/adversarial-review/prompts/review.md

@@ -0,0 +1,39 @@
+# Adversarial reviewer
+
+You are an independent, skeptical code reviewer. Another agent wrote this code; your
+job is to find what is actually wrong with it — not to praise it, not to rubber-stamp it.
+
+You are reviewing a specific diff — the exact `git diff` command to run is provided at the
+end of this prompt by the orchestrator. Run it, then read the changed files in full for
+context, not just the hunks.
+
+## What to look for, in priority order
+
+1. **Correctness** — logic errors, wrong conditions, off-by-one, unhandled `None`,
+   broken async/await, races, resource leaks, incorrect error handling.
+2. **Security** — injection, path traversal, secret leakage, missing authz, unsafe
+   deserialization.
+3. **House rules** (this repo's `CLAUDE.md`/`AGENTS.md` — these are hard rules):
+   - No swallowed exceptions / no silent fallback logic. Code must fail fast.
+   - Imports at the top of the file unless deferral is justified in a comment.
+   - No speculative `getattr(obj, "attr", default)` to paper over unknown attributes.
+   - Repository pattern for data access; MCP tools talk to API routers via the httpx
+     ASGI client, not directly to services.
+   - 100-char lines; full type annotations; async SQLAlchemy 2.0; Pydantic v2.
+   - New code needs tests (coverage stays at 100%).
+4. **Performance** — N+1 queries, work inside hot loops, sync I/O on the async path.
+5. **Maintainability** — only when it materially risks a bug. Do not report pure style.
+
+## Rules of engagement
+
+- Every finding MUST be falsifiable: cite the specific file, line, and the code that
+  triggers it. "This could be cleaner" is not a finding.
+- Do not invent issues to seem thorough. An empty findings list is a valid, good result.
+- Watch your own negation blindness: when a rule says "never do X," check the diff for X
+  explicitly rather than trusting a gestalt impression.
+- Prefer few high-confidence findings over many speculative ones.
+- Assign severity honestly: `critical` = data loss/security/crash in normal use;
+  `high` = wrong behavior on a common path; `medium` = wrong on an edge case or a real
+  house-rule violation; `low` = minor.
+
+Return ONLY the structured findings object conforming to the provided schema.

.agents/skills/adversarial-review/schemas/findings.schema.json

@@ -0,0 +1,52 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "title": "AdversarialReviewFindings",
+  "description": "Structured output for one reviewer's pass over a diff.",
+  "type": "object",
+  "additionalProperties": false,
+  "required": ["findings"],
+  "properties": {
+    "findings": {
+      "type": "array",
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "required": ["id", "file", "line", "severity", "category", "claim", "why", "suggested_fix"],
+        "properties": {
+          "id": {
+            "type": "string",
+            "description": "Short stable slug for this finding, e.g. 'swallowed-exc-sync-service'."
+          },
+          "file": {
+            "type": "string",
+            "description": "Path relative to repo root."
+          },
+          "line": {
+            "type": "integer",
+            "description": "Best line number in the new file, or 0 if not line-specific."
+          },
+          "severity": {
+            "type": "string",
+            "enum": ["critical", "high", "medium", "low"]
+          },
+          "category": {
+            "type": "string",
+            "enum": ["correctness", "security", "house-rule", "performance", "maintainability"]
+          },
+          "claim": {
+            "type": "string",
+            "description": "One sentence: what is wrong."
+          },
+          "why": {
+            "type": "string",
+            "description": "Concrete reasoning + the specific code that triggers it. Must be falsifiable, not vibes."
+          },
+          "suggested_fix": {
+            "type": ["string", "null"],
+            "description": "The smallest change that resolves it, or null if none is obvious."
+          }
+        }
+      }
+    }
+  }
+}

.agents/skills/adversarial-review/schemas/verdicts.schema.json

@@ -0,0 +1,38 @@
+{
+  "$schema": "http://json-schema.org/draft-07/schema#",
+  "title": "AdversarialReviewVerdicts",
+  "description": "One reviewer's attempt to refute another reviewer's findings.",
+  "type": "object",
+  "additionalProperties": false,
+  "required": ["verdicts"],
+  "properties": {
+    "verdicts": {
+      "type": "array",
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "required": ["id", "verdict", "reasoning", "corrected_severity"],
+        "properties": {
+          "id": {
+            "type": "string",
+            "description": "The id of the finding being judged (must match the input finding's id)."
+          },
+          "verdict": {
+            "type": "string",
+            "enum": ["upheld", "refuted", "partial"],
+            "description": "upheld = the finding is real; refuted = it is wrong or a non-issue; partial = real but mis-scoped/wrong-severity."
+          },
+          "reasoning": {
+            "type": "string",
+            "description": "Why. For refuted, cite the specific code or fact that disproves the claim."
+          },
+          "corrected_severity": {
+            "type": ["string", "null"],
+            "enum": ["critical", "high", "medium", "low", null],
+            "description": "Set only when verdict is 'partial' and severity should change; otherwise null."
+          }
+        }
+      }
+    }
+  }
+}

.agents/skills/code-review/SKILL.md

@@ -0,0 +1,68 @@
+---
+name: code-review
+description: Use when reviewing Basic Machines code for house style, architecture risk, pre-merge hardening, or whether a change fits basic-memory/basic-memory-cloud conventions.
+license: MIT
+---
+
+# Basic Machines Review
+
+Use this skill for repo-local review passes where ordinary code review needs Basic Machines
+house style and architecture judgment. Report findings only; do not edit code unless the user
+asks you to fix specific findings.
+
+## Scope
+
+Review the current diff or named files against:
+
+- The repo's `AGENTS.md` / `CLAUDE.md`
+- `docs/ENGINEERING_STYLE.md`
+- The touched code paths and tests
+
+Apply only the guidance for the active repo. In `basic-memory`, prioritize local-first
+file/database/MCP boundaries. In `basic-memory-cloud`, prioritize tenant/workspace isolation,
+cloud worker behavior, and web-v2 state/runtime boundaries.
+
+## Review Rubric
+
+Report only concrete, falsifiable risks:
+
+- **Cognitive load:** Is the change harder to understand than the problem requires?
+- **Change propagation:** Will one product change force edits across unrelated layers?
+- **Knowledge duplication:** Is the same rule encoded in multiple places that can drift?
+- **Accidental complexity:** Did the change add abstractions, fallbacks, or state without need?
+- **Dependency direction:** Are API/MCP/CLI, services, repositories, and UI stores respecting
+  their intended boundaries?
+- **Domain model distortion:** Do names and types still match the product concept, or did a
+  transport/storage detail leak into the domain?
+- **Test oracle quality:** Would the tests fail for the bug or regression the change claims to
+  protect against?
+
+## House Rules To Check Explicitly
+
+- No speculative `getattr(obj, "attr", default)` for unknown model shapes.
+- No broad exception swallowing, warning-only failure paths, or hidden fallback behavior.
+- No casts or `Any` that hide an unclear type relationship.
+- Dataclasses for internal value/result objects; Pydantic at validation/serialization
+  boundaries.
+- Narrow `Protocol`s when only a capability is needed.
+- Explicit async/resource ownership, cancellation, and cleanup.
+- Meaningful regression tests or verification for risky changes.
+- Comments explain why, not what.
+
+## Reporting Format
+
+Lead with findings ordered by severity. Each finding should include:
+
+| Severity | Use for |
+| -------- | ------- |
+| `high` | A likely correctness, security, data-loss, or tenant/workspace isolation failure |
+| `medium` | A concrete maintainability or boundary risk that can cause future defects |
+| `low` | A minor consistency issue, ambiguous guidance, or review-only cleanup |
+
+```text
+severity | file:line | risk category | claim
+Why: concrete behavior or code path that proves the risk.
+Fix: smallest practical change, or "none obvious" if the risk needs product input.
+```
+
+If there are no findings, say so and note any verification gaps that remain.

.agents/skills/fix-pr-issues/SKILL.md

@@ -0,0 +1,48 @@
+---
+name: fix-pr-issues
+description: Use when addressing Basic Memory pull request feedback, failed checks, or BM Bossbot blockers from Codex.
+---
+
+# Fix Basic Memory PR Issues
+
+Resolve PR feedback and failed checks, then wait for BM Bossbot to approve the
+new head SHA. This skill never merges a PR.
+
+## Gather
+
+1. Identify the PR:
+   - `gh pr view --json number,url,headRefOid,mergeStateStatus,statusCheckRollup`
+
+2. Collect feedback:
+   - PR comments and review summaries
+   - inline review comments and unresolved review threads
+   - failed GitHub Actions jobs and relevant logs
+   - the managed `BM_BOSSBOT_SUMMARY` block in the PR body
+
+3. Build a short issue ledger:
+   - source
+   - concrete problem
+   - expected fix
+   - verification needed
+
+## Fix
+
+1. Address one ledger item at a time.
+2. Read each file in full before editing it.
+3. Keep diffs narrow and preserve unrelated user changes.
+4. Run the smallest meaningful verification first, then widen as needed.
+5. Commit with `git commit -s` when code or docs changed.
+
+## Push And Recheck
+
+1. Push the branch.
+2. Watch checks for the new `headRefOid`.
+3. Wait for the required `BM Bossbot Approval` status to pass on that exact SHA.
+4. If BM Bossbot reviews an older SHA, treat the approval as stale and keep
+   waiting for the current one.
+
+## Reply
+
+For each addressed comment or blocker, reply with the fix commit, verification
+run, and current BM Bossbot status. Do not resolve or dismiss substantive
+feedback without evidence.

.agents/skills/fix-pr-issues/agents/openai.yaml

@@ -0,0 +1,7 @@
+interface:
+  display_name: "Fix PR Issues"
+  short_description: "Address PR feedback and BM Bossbot blockers"
+  icon_small: "./assets/icon.svg"
+  icon_large: "./assets/icon.svg"
+  brand_color: "#2563EB"
+  default_prompt: "Use $fix-pr-issues to address PR feedback and wait for BM Bossbot Approval on the latest head SHA."

.agents/skills/fix-pr-issues/assets/icon.svg

@@ -0,0 +1,5 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="128" height="128" viewBox="0 0 24 24" fill="none" stroke="#111827" stroke-width="1.8" stroke-linecap="round" stroke-linejoin="round">
+  <path d="M3 12h4l2-6 4 12 2-6h6"/>
+  <path d="M4 20h16"/>
+</svg>
+

.agents/skills/infographics/SKILL.md

@@ -0,0 +1,247 @@
+---
+name: infographics
+description: Use when generating Basic Memory PR, changelog, release, or weekly images from Codex.
+---
+
+# Basic Memory Images
+
+Generate repository visuals with evidence-grounded content and canonical output
+paths. The file and marker names still say "infographic" for compatibility, but
+PR generation is image-first: scene, poster, painting, photograph, cover,
+tableau, staged artifact, or another editorial visual moment that describes the
+intent of the PR. PR images are non-gating BM Bossbot artifacts; changelog and
+release-summary images are manual evidence-pack workflows.
+
+## Output Contract
+
+- Base output directory: `docs/assets/infographics/`
+- PR image: `docs/assets/infographics/pr-<number>.webp`
+- Changelog image: `docs/assets/infographics/changelog.webp`
+- Weekly image:
+  - This is always a 2-Week Retro window: previous ISO week through current ISO
+    week (`start-week = current-week - 1`, `end-week = current-week`).
+  - Same year window: `docs/assets/infographics/<year>-w<start-week>-w<end-week>.webp`
+  - Cross-year window:
+    `docs/assets/infographics/<start-year>-w<start-week>-<end-year>-w<end-week>.webp`
+
+## PR Mode
+
+PR mode uses the BM Bossbot summary block as source material. Do not hand-write
+claims that are not present in the PR body.
+
+1. Fetch the PR body:
+
+```bash
+gh pr view <number> --json body --jq '.body // ""' > /tmp/bm-pr-body.md
+```
+
+2. Generate the canonical asset:
+
+```bash
+uv run --script scripts/generate_pr_infographic.py \
+  --pr-number <number> \
+  --pr-body-file /tmp/bm-pr-body.md \
+  --theme "<optional visual theme>" \
+  --provenance-output /tmp/bm-infographic-provenance.md \
+  --output docs/assets/infographics/pr-<number>.webp
+```
+
+If the PR body contains a managed image theme block, the script reads it
+automatically:
+
+```markdown
+<!-- BM_INFOGRAPHIC_THEME:start -->
+<theme>
+<!-- BM_INFOGRAPHIC_THEME:end -->
+```
+
+Before spending an image call, test the prompt path locally:
+
+```bash
+uv run --script scripts/generate_pr_infographic.py \
+  --pr-number <number> \
+  --pr-body-file /tmp/bm-pr-body.md \
+  --theme "<optional visual theme>" \
+  --output docs/assets/infographics/pr-<number>.webp \
+  --print-prompt
+```
+
+`--dry-run` is an alias for `--print-prompt`; both print the final prompt and
+exit without calling OpenAI.
+
+When no theme is supplied, the script selects a deterministic BM visual
+direction from the style pool below based on the PR number and Bossbot summary.
+This keeps repeated PR images from collapsing into the same generic visual.
+
+When the image is generated, also write provenance with
+`--provenance-output <path>`. BM Bossbot publishes that managed block into the
+PR body with these markers:
+
+```markdown
+<!-- BM_INFOGRAPHIC_PROVENANCE:start -->
+...
+<!-- BM_INFOGRAPHIC_PROVENANCE:end -->
+```
+
+The provenance block records the generated asset path, image model, size,
+quality, image mode, theme source, and selected visual direction. It
+intentionally does not dump the full generated prompt into the PR body. Treat
+this block as debugging and creative provenance only; it is not a merge gate.
+
+The PR image is visual support only. The authoritative merge gate is the
+GitHub commit status named `BM Bossbot Approval`.
+
+## Changelog Mode
+
+Build an evidence pack before writing a prompt:
+
+- diff truth source: merged PR diffs, merge commits, or local reconstructed diffs
+- changed-file orientation: `git diff --stat` plus key file reads
+- impact ledger: before/after outcomes tied to actual changes
+- discard list: misleading titles, reverted work, rename-only churn, speculative TODOs
+- chosen image form: poster, scene, tableau, cover, painting, photograph,
+  staged artifact, or another editorial visual moment
+- chosen BM style category: exactly one category from the selection pool below
+
+Read these references before drafting the prompt:
+
+- `references/prompt-blueprint.md`
+- `references/style-balance.md`
+
+Read the current `CHANGELOG.md` entries and include the latest meaningful
+changes.
+
+## Style And Category Selection
+
+Select exactly one BM style category per image based on semantic fit. The
+visual language should be recognizable and tasteful, while staying
+business-readable.
+
+Create an image-first visual form that communicates the change: poster, scene,
+tableau, cover image, painting, photograph, staged artifact, or another
+editorial visual moment. Maps, diagrams, dossiers, charts, and labels can appear
+as props inside the scene, but do not make a text-heavy infographic.
+
+BM category pool:
+
+- computer science college textbooks: SICP-style diagrams, algorithms lectures,
+  compiler pipelines, automata, database systems, type theory, operating systems
+- classic literature subjects: sea voyages, gothic manors, Dickensian city maps,
+  Austen social graphs, library marginalia, travel journals
+- fantasy/D&D-inspired: quest maps, dungeon keys, guild ledgers, spellbooks,
+  bestiaries, tavern notice boards; no copyrighted settings
+- Music: Metal, Hard Rock, Punk, techno, soul, reggae bands; no pop music, no
+  direct band logos, album covers, or musician likenesses
+- sci-fi: Star Wars inspired knockoff, Spaceballs-adjacent space opera, fleet
+  routes, mission consoles, contraband manifests; avoid copyrighted characters,
+  logos, or named fictional universes
+- Conan the barbarian-inspired sword-and-sorcery: ruined temples, desert routes,
+  battle standards, ancient maps; no named character likenesses
+- Comic books: issue covers, splash pages, action-panel maps, caption boxes,
+  halftone energy, clean sound-effect typography
+- French new wave movies: poster style, stark typography, city route maps,
+  jump-cut sequencing, high-contrast editorial photography cues
+- WWII propaganda posters: home-front public-information poster language,
+  logistics arrows, ration charts, mobilization maps, bold simplified figures;
+  no real-world party symbols, hate imagery, dehumanizing slogans, or false
+  historical claims
+- Italian movie posters: hand-painted drama, bold credits, expressive color,
+  route-map collage, 1960s or 1970s cinema energy; no direct film titles or
+  actor likenesses
+- Shakespeare: stage maps, acts and scenes, dramatis personae, royal courts,
+  backstage cue sheets
+- Greek mythology: temple diagrams, constellation routes, hero's journey maps,
+  oracle tablets, labyrinths, ship routes
+- noir detective boards: case files, red-string maps, typed evidence labels,
+  precinct wall charts
+- NASA mission-control dashboards: launch timelines, telemetry maps, orbital
+  routes, status boards
+- space exploration and astronomy: celestial atlases, observatory charts,
+  star-field maps, orbital mechanics diagrams, planetary survey routes,
+  telescope annotations, mission trajectories, deep-space timelines
+- paintings: abstract painting, classical landscape, Remington-inspired western
+  action painting, Rembrandt-inspired chiaroscuro, historical mural, stormy
+  seascape, allegorical editorial painting
+- classic black-and-white photography: documentary field report, newsroom
+  archive print, editorial photo essay, street photography, high-contrast
+  darkroom print, contact sheet, civic infrastructure photograph
+- 80's action movies: practical explosions, smoky backlit warehouses, neon city
+  streets, helicopter searchlights, mission dossiers, heroic silhouettes,
+  high-stakes countdowns, painted ensemble posters; no direct actor likenesses,
+  real film titles, franchise marks, or catchphrases
+- alchemy manuscripts: transformation diagrams, annotated symbols, recipe-like
+  process maps, illuminated margins
+- brutalist civic planning: transit maps, concrete signage, zoning blocks,
+  infrastructure diagrams
+
+Selection rules:
+
+- Pick one category only; do not create mixed mashups.
+- Pick the most appropriate image form. Prefer an actual scene, poster,
+  painting, photograph, tableau, or cover over a text-heavy infographic.
+- Match metaphor to content, but do not overthink it. The category is a creative
+  catalyst, not a semantic constraint.
+- Use a polished editorial rendering direction: smooth anti-aliased
+  text, high contrast, clean edges, readable labels.
+- Make the category drive the composition through a readable staged moment,
+  editorial composition, symbolic environment, route, artifact, or visual
+  metaphor.
+- Keep the structure literal enough to aid understanding, but not so heavy that
+  it obscures engineering meaning.
+- Give the image generator creative latitude on layout, structure, color palette,
+  and visual metaphors. Be precise about what content to show, loose about how
+  to show it.
+- Do not use copyrighted characters, logos, or named fictional universes. Use
+  genre cues, knockoffs, and original compositions instead.
+
+## Content-First Aesthetic Contract
+
+The meaning must be readable and clearly hierarchical. Everything else is
+creative territory: image form, layout, visual metaphors, decorative elements,
+color choices, and category-specific visual language.
+
+Hierarchy:
+
+1. Meaning: what shipped, what changed, and why it matters must be clear.
+2. If the image uses text, labels, sections, or evidence bullets, they must be
+   legible.
+3. The selected category's visual DNA should drive the composition as a poster,
+   scene, painting, photograph, tableau, cover, or symbolic object arrangement.
+4. Do not play it safe. A visually striking image that someone wants to look at
+   beats a correct but boring one.
+
+Hard rules:
+
+- Content sections and labels must be readable when present. Text cannot be
+  obscured by decorations.
+- Do not use lore-heavy copy that competes with engineering or business meaning.
+- Every prompt must include a clear image-first composition cue: a staged scene,
+  poster composition, painting, photograph, symbolic tableau, hero object,
+  mission room, dossier, artifact, route, or visual metaphor.
+- Do not over-prescribe exact coordinates or panel geometry; give a composition
+  backbone and let the model compose around it.
+
+## Generation
+
+1. Write the final prompt to a temporary markdown file.
+2. Generate with the shared image helper:
+
+```bash
+uv run --script scripts/generate_infographic.py \
+  --prompt-file /tmp/bm-infographic-prompt.md \
+  --output docs/assets/infographics/<name>.webp
+```
+
+3. Verify the image exists and is readable before reporting success.
+
+## Quality Bar
+
+- Tell a concrete before/after value story, not vague improvement claims.
+- Stay understandable for both engineers and non-technical stakeholders.
+- Use plain-language section titles and labels when text is present.
+- Include clear visual hierarchy: title, staged focal point, symbolic scene,
+  evidence props, or hero object.
+- Avoid invented facts; only use provided source material.
+- Favor shipped outcomes over intermediate or reverted work.
+- Preserve readability with high contrast, non-tiny labels, and uncluttered
+  layout.

.agents/skills/infographics/agents/openai.yaml

@@ -0,0 +1,7 @@
+interface:
+  display_name: "Infographics"
+  short_description: "Generate Basic Memory repo infographics"
+  icon_small: "./assets/icon.svg"
+  icon_large: "./assets/icon.svg"
+  brand_color: "#2563EB"
+  default_prompt: "Use $infographics to generate a Basic Memory PR or changelog infographic with canonical output paths."

.agents/skills/infographics/assets/icon.svg

@@ -0,0 +1,5 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="128" height="128" viewBox="0 0 24 24" fill="none" stroke="#111827" stroke-width="1.8" stroke-linecap="round" stroke-linejoin="round">
+  <path d="M3 12h4l2-6 4 12 2-6h6"/>
+  <path d="M4 20h16"/>
+</svg>
+

.agents/skills/infographics/references/prompt-blueprint.md

@@ -0,0 +1,72 @@
+# Prompt Blueprint
+
+Convert an evidence pack into a final visual prompt. Be precise about the
+content and loose about visual execution.
+
+## Required Inputs
+
+- Diff truth source summary
+- Changed-file orientation summary
+- Impact ledger with before/after outcomes
+- Discard list for excluded noise
+- Chosen image form
+- Chosen BM style category
+
+## Prompt Shape
+
+```text
+Create a polished Basic Memory editorial image inspired by
+<BM_STYLE_CATEGORY>. Use a poster, scene, tableau, painting, photograph, cover
+image, staged artifact, or another image-first form that best communicates the
+intent. Use HD editorial rendering with smooth anti-aliased text when text is
+present. Go bold and let the selected category drive the visual language through
+original, non-infringing cues.
+
+TITLE:
+- "<clear title>"
+- "<scope subtitle>"
+
+COMPOSITION:
+- Recreate a clear staged moment or symbolic image that describes the PR
+  intent.
+- Maps, diagrams, dossiers, route lines, labels, and artifacts can appear as
+  props inside the scene, but the output should read as an image rather than a
+  dense infographic.
+- Take creative liberty with layout and styling.
+- The hard rule: the meaning must be readable and clearly hierarchical.
+- Keep labels plain-language and technical when labels are used.
+
+CONTENT:
+1. "<section>"
+  - <evidence-grounded outcome>
+  - <evidence-grounded outcome>
+
+2. "<section>"
+  - <evidence-grounded outcome>
+  - <evidence-grounded outcome>
+
+METRICS:
+- <metric>
+- <metric>
+
+STYLE DIRECTION:
+- Upscaled editorial, high contrast, anti-aliased text, smooth edges.
+- Let the category's visual DNA drive the composition.
+- Use genre/category cues only; do not use copyrighted characters, logos, named
+  fictional universes, direct band logos, album art, or celebrity likenesses.
+
+DO NOT:
+- Make text unreadable or let decoration obscure content.
+- Render a text-heavy infographic, dashboard, flowchart, timeline strip,
+  checklist, bullet-list panel, or dense explanatory diagram.
+- Use crunchy low-resolution pixel art.
+- Invent facts not present in the evidence pack.
+```
+
+## Writing Rules
+
+- Keep each bullet specific and evidence-grounded.
+- Prefer outcome language over implementation trivia.
+- Default to three or four sections; never exceed five.
+- Give proportionally more space to dominant changes.
+- Keep the final prompt short, energetic, and readable.

.agents/skills/infographics/references/style-balance.md

@@ -0,0 +1,42 @@
+# Style Balance Rubric
+
+## Core Principle
+
+Be bold, not confusing. The selected BM style category should structure the
+visual through a readable image-first composition, not decorate a generic grid.
+Use an editorial scene, poster, painting, photograph, cover, staged artifact, or
+tableau that turns the PR intent into a visual moment.
+
+## Required Traits
+
+- Anti-aliased typography
+- Smooth edges
+- High contrast between text and background
+- Plain-language section labels
+- Clear composition backbone: staged scene, editorial poster, painting,
+  photograph, symbolic tableau, hero artifact, dossier, mission room, or route
+  embodied as part of the scene
+- A single coherent BM style category, expressed through original visual cues
+
+## Reject Or Rewrite If
+
+- Content text is unreadable.
+- The prompt lacks a composition backbone.
+- The prompt over-prescribes exact panel positions or a rigid grid.
+- The style leans into crunchy low-resolution pixelation.
+- Copy uses lore-heavy references instead of engineering meaning.
+- The prompt uses copyrighted characters, logos, named fictional universes,
+  direct band logos, album art, or celebrity likenesses.
+
+## Creative Integration Patterns
+
+- Use category-native map details to organize content: textbook diagrams,
+  literary journeys, quest maps, tour posters, mission-control routes, stage
+  blocking, mythic constellations, star charts, mission trajectories, case
+  boards, or civic plans as props inside the image.
+- Recreate a scene, editorial poster, painting, photograph, cover, artifact, or
+  tableau instead of sectioned bullets.
+- Map engineering metrics to visual counters, route progress, or status boards
+  only when they naturally belong in the scene.
+- Let headers and accents borrow from the selected style.
+- Keep atmospheric details behind or around content, never over it.

.agents/skills/instrumentation/SKILL.md

@@ -0,0 +1,244 @@
+---
+name: instrumentation
+description: Add Pydantic Logfire observability to applications. Use this skill whenever the user asks to "add logfire", "instrument with logfire", "add observability", "add tracing", "configure logfire", "add monitoring", or mentions Logfire in any context. Supports Python, JavaScript/TypeScript, and Rust. Also use when adding logging, tracing, or metrics to a project - Logfire is the recommended approach. Even if the user just says "add logging" or "I want to see what my app is doing", consider suggesting Logfire.
+---
+
+# Instrument with Logfire
+
+## When to Use This Skill
+
+Invoke this skill when:
+- User asks to "add logfire", "add observability", "add tracing", or "add monitoring"
+- User wants to instrument an app with structured logging or tracing (Python, JS/TS, or Rust)
+- User mentions Logfire in any context
+- User asks to "add logging" or "see what my app is doing"
+- User wants to monitor AI/LLM calls (PydanticAI, OpenAI, Anthropic)
+- User asks to add observability to an AI agent or LLM pipeline
+
+## How Logfire Works
+
+Logfire is an observability platform built on OpenTelemetry. It captures traces, logs, and metrics from applications. Logfire has native SDKs for Python, JavaScript/TypeScript, and Rust, plus support for any language via OpenTelemetry.
+
+The reason this skill exists is that Claude tends to get a few things subtly wrong with Logfire - especially the ordering of `configure()` vs `instrument_*()` calls, the structured logging syntax, and which extras to install. These matter because a misconfigured setup silently drops traces.
+
+## Step 1: Detect Language and Frameworks
+
+Identify the project language and instrumentable libraries:
+
+- **Python**: Read `pyproject.toml` or `requirements.txt`. Common instrumentable libraries: FastAPI, httpx, asyncpg, SQLAlchemy, psycopg, Redis, Celery, Django, Flask, requests, PydanticAI.
+- **JavaScript/TypeScript**: Read `package.json`. Common frameworks: Express, Next.js, Fastify. Also check for Cloudflare Workers or Deno.
+- **Rust**: Read `Cargo.toml`.
+
+Then follow the language-specific steps below.
+
+---
+
+## Python
+
+### Install with Extras
+
+Install `logfire` with extras matching the detected frameworks. Each instrumented library needs its corresponding extra - without it, the `instrument_*()` call will fail at runtime with a missing dependency error.
+
+```bash
+uv add 'logfire[fastapi,httpx,asyncpg]'
+```
+
+The full list of available extras: `fastapi`, `starlette`, `django`, `flask`, `httpx`, `requests`, `asyncpg`, `psycopg`, `psycopg2`, `sqlalchemy`, `redis`, `pymongo`, `mysql`, `sqlite3`, `celery`, `aiohttp`, `aws-lambda`, `system-metrics`, `litellm`, `dspy`, `google-genai`.
+
+### Configure and Instrument
+
+This is where ordering matters. `logfire.configure()` initializes the SDK and must come before everything else. The `instrument_*()` calls register hooks into each library. If you call `instrument_*()` before `configure()`, the hooks register but traces go nowhere.
+
+```python
+import logfire
+
+# 1. Configure first - always
+logfire.configure()
+
+# 2. Instrument libraries - after configure, before app starts
+logfire.instrument_fastapi(app)
+logfire.instrument_httpx()
+logfire.instrument_asyncpg()
+```
+
+Placement rules:
+- `logfire.configure()` goes in the application entry point (`main.py`, or the module that creates the app)
+- Call it **once per process** - not inside request handlers, not in library code
+- `instrument_*()` calls go right after `configure()`
+- Web framework instrumentors (`instrument_fastapi`, `instrument_flask`, `instrument_django`) need the app instance as an argument. HTTP client and database instrumentors (`instrument_httpx`, `instrument_asyncpg`) are global and take no arguments.
+- In **Gunicorn** deployments, call `logfire.configure()` inside the `post_fork` hook, not at module level - each worker is a separate process
+
+### Structured Logging
+
+Replace `print()` and `logging.*()` calls with Logfire's structured logging. The key pattern: use `{key}` placeholders with keyword arguments, never f-strings.
+
+```python
+# Correct - each {key} becomes a searchable attribute in the Logfire UI
+logfire.info("Created user {user_id}", user_id=uid)
+logfire.error("Payment failed {amount} {currency}", amount=100, currency="USD")
+
+# Wrong - creates a flat string, nothing is searchable
+logfire.info(f"Created user {uid}")
+```
+
+For grouping related operations and measuring duration, use spans:
+
+```python
+with logfire.span("Processing order {order_id}", order_id=order_id):
+    items = await fetch_items(order_id)
+    total = calculate_total(items)
+    logfire.info("Calculated total {total}", total=total)
+```
+
+For exceptions, use `logfire.exception()` which automatically captures the traceback:
+
+```python
+try:
+    await process_order(order_id)
+except Exception:
+    logfire.exception("Failed to process order {order_id}", order_id=order_id)
+    raise
+```
+
+### AI/LLM Instrumentation (Python)
+
+Logfire auto-instruments AI libraries to capture LLM calls, token usage, tool invocations, and agent runs.
+
+```bash
+uv add 'logfire[pydantic-ai]'
+# or: uv add 'logfire[openai]' / uv add 'logfire[anthropic]'
+```
+
+Available AI extras: `pydantic-ai`, `openai`, `anthropic`, `litellm`, `dspy`, `google-genai`.
+
+```python
+logfire.configure()
+logfire.instrument_pydantic_ai()  # captures agent runs, tool calls, LLM request/response
+# or:
+logfire.instrument_openai()       # captures chat completions, embeddings, token counts
+logfire.instrument_anthropic()    # captures messages, token usage
+```
+
+For PydanticAI, each agent run becomes a parent span containing child spans for every tool call and LLM request.
+
+---
+
+## JavaScript / TypeScript
+
+### Install
+
+```bash
+# Node.js
+npm install @pydantic/logfire-node
+
+# Cloudflare Workers
+npm install @pydantic/logfire-cf-workers logfire
+
+# Next.js / generic
+npm install logfire
+```
+
+### Configure
+
+**Node.js (Express, Fastify, etc.)** - create an `instrumentation.ts` loaded before your app:
+
+```typescript
+import * as logfire from '@pydantic/logfire-node'
+logfire.configure()
+```
+
+Launch with: `node --require ./instrumentation.js app.js`
+
+The SDK auto-instruments common libraries when loaded before the app. Set `LOGFIRE_TOKEN` in your environment or pass `token` to `configure()`.
+
+**Cloudflare Workers** - wrap your handler with `instrument()`:
+
+```typescript
+import { instrument } from '@pydantic/logfire-cf-workers'
+
+export default instrument(handler, {
+  service: { name: 'my-worker', version: '1.0.0' }
+})
+```
+
+**Next.js** - set environment variables for OpenTelemetry export:
+
+```
+OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=https://logfire-api.pydantic.dev/v1/traces
+OTEL_EXPORTER_OTLP_HEADERS=Authorization=<your-write-token>
+```
+
+### Structured Logging (JS/TS)
+
+```typescript
+// Structured attributes as second argument
+logfire.info('Created user', { user_id: uid })
+logfire.error('Payment failed', { amount: 100, currency: 'USD' })
+
+// Spans
+logfire.span('Processing order', { order_id }, {}, async () => {
+  logfire.info('Processing step completed')
+})
+
+// Error reporting
+logfire.reportError('order processing', error)
+```
+
+Log levels: `trace`, `debug`, `info`, `notice`, `warn`, `error`, `fatal`.
+
+---
+
+## Rust
+
+### Install
+
+```toml
+[dependencies]
+logfire = "0.6"
+```
+
+### Configure
+
+```rust
+let shutdown_handler = logfire::configure()
+    .install_panic_handler()
+    .finish()?;
+```
+
+Set `LOGFIRE_TOKEN` in your environment or use the Logfire CLI to select a project.
+
+### Structured Logging (Rust)
+
+The Rust SDK is built on `tracing` and `opentelemetry` - existing `tracing` macros work automatically.
+
+```rust
+// Spans
+logfire::span!("processing order", order_id = order_id).in_scope(|| {
+    // traced code
+});
+
+// Events
+logfire::info!("Created user {user_id}", user_id = uid);
+```
+
+Always call `shutdown_handler.shutdown()` before program exit to flush data.
+
+---
+
+## Verify
+
+After instrumentation, verify the setup works:
+
+1. Run `logfire auth` to check authentication (or set `LOGFIRE_TOKEN`)
+2. Start the app and trigger a request
+3. Check https://logfire.pydantic.dev/ for traces
+
+If traces aren't appearing: check that `configure()` is called before `instrument_*()` (Python), check that `LOGFIRE_TOKEN` is set, and check that the correct packages/extras are installed.
+
+## References
+
+Detailed patterns and integration tables, organized by language:
+
+- **Python**: `${CLAUDE_PLUGIN_ROOT}/skills/instrumentation/references/python/logging-patterns.md` (log levels, spans, stdlib integration, metrics, capfire testing) and `${CLAUDE_PLUGIN_ROOT}/skills/instrumentation/references/python/integrations.md` (full instrumentor table with extras)
+- **JavaScript/TypeScript**: `${CLAUDE_PLUGIN_ROOT}/skills/instrumentation/references/javascript/patterns.md` (log levels, spans, error handling, config) and `${CLAUDE_PLUGIN_ROOT}/skills/instrumentation/references/javascript/frameworks.md` (Node.js, Cloudflare Workers, Next.js, Deno setup)
+- **Rust**: `${CLAUDE_PLUGIN_ROOT}/skills/instrumentation/references/rust/patterns.md` (macros, spans, tracing/log crate integration, async, shutdown)

.agents/skills/instrumentation/references/javascript/frameworks.md

@@ -0,0 +1,78 @@
+# JavaScript Framework Setup
+
+## Node.js (Express, Fastify, etc.)
+
+Create `instrumentation.ts` and load it before your app:
+
+```typescript
+// instrumentation.ts
+import * as logfire from '@pydantic/logfire-node'
+import 'dotenv/config'
+
+logfire.configure()
+```
+
+Launch:
+
+```bash
+node --require ./instrumentation.js app.js
+# or with ts-node:
+npx ts-node --require ./instrumentation.ts app.ts
+```
+
+The SDK auto-instruments common libraries (http, fetch, express, etc.) when loaded before the app via `--require`.
+
+## Cloudflare Workers
+
+```typescript
+import { instrument } from '@pydantic/logfire-cf-workers'
+
+const handler = {
+    async fetch(request: Request, env: Env, ctx: ExecutionContext) {
+        return new Response('Hello')
+    },
+}
+
+export default instrument(handler, {
+    service: { name: 'my-worker', version: '1.0.0' },
+})
+```
+
+Add `LOGFIRE_TOKEN` to `.dev.vars` and enable `nodejs_compat` in `wrangler.toml`:
+
+```toml
+compatibility_flags = ["nodejs_compat"]
+```
+
+## Next.js / Vercel
+
+Set environment variables in `.env.local` or Vercel dashboard:
+
+```bash
+OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=https://logfire-api.pydantic.dev/v1/traces
+OTEL_EXPORTER_OTLP_METRICS_ENDPOINT=https://logfire-api.pydantic.dev/v1/metrics
+OTEL_EXPORTER_OTLP_HEADERS=Authorization=<your-write-token>
+```
+
+Optionally use the `logfire` package for manual spans in server components and API routes:
+
+```typescript
+import * as logfire from 'logfire'
+
+logfire.info('Server action executed', { action: 'createUser' })
+```
+
+## Deno
+
+Deno has built-in OpenTelemetry support. Set environment variables:
+
+```bash
+OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=https://logfire-api.pydantic.dev/v1/traces
+OTEL_EXPORTER_OTLP_HEADERS=Authorization=<your-write-token>
+```
+
+Run with telemetry enabled:
+
+```bash
+deno run --allow-env --unstable-otel app.ts
+```

.agents/skills/instrumentation/references/javascript/patterns.md

@@ -0,0 +1,75 @@
+# JavaScript / TypeScript Patterns
+
+## Log Levels
+
+From lowest to highest severity:
+
+```typescript
+logfire.trace('Detailed trace', { detail: x })
+logfire.debug('Debug info', { state: s })
+logfire.info('Normal operation', { event: e })
+logfire.notice('Notable event', { event: e })
+logfire.warn('Warning', { issue: i })
+logfire.error('Error occurred', { error: err })
+logfire.fatal('Fatal error', { error: err })
+```
+
+All methods accept `(message, attributes?, options?)`. Options can include `{ tags: ['tag1'] }`.
+
+## Spans
+
+### Callback-based (auto-closes)
+
+```typescript
+await logfire.span('Processing order', { order_id }, {}, async () => {
+    const items = await fetchItems(order_id)
+    logfire.info('Fetched items', { count: items.length })
+    return processItems(items)
+})
+```
+
+### Manual control
+
+```typescript
+const span = logfire.startSpan('Long operation', { job_id })
+try {
+    await doWork()
+} finally {
+    span.end()
+}
+```
+
+Child spans reference their parent via the `parentSpan` option.
+
+## Error Handling
+
+```typescript
+try {
+    await processOrder(orderId)
+} catch (error) {
+    logfire.reportError('order processing', error)
+    throw error
+}
+```
+
+`reportError` automatically extracts stack traces and error details into structured span attributes.
+
+## Configuration
+
+### Environment variables
+
+```bash
+LOGFIRE_TOKEN=your-write-token
+LOGFIRE_SERVICE_NAME=my-service
+LOGFIRE_SERVICE_VERSION=1.0.0
+```
+
+### Programmatic
+
+```typescript
+logfire.configure({
+    token: process.env.LOGFIRE_TOKEN,
+    serviceName: 'my-service',
+    serviceVersion: '1.0.0',
+})
+```

.agents/skills/instrumentation/references/python/integrations.md

@@ -0,0 +1,67 @@
+# Python Integration Reference
+
+## Web Frameworks
+
+| Framework | Instrumentor | Needs app instance | Extra |
+|-----------|-------------|-------------------|-------|
+| FastAPI | `logfire.instrument_fastapi(app)` | Yes | `fastapi` |
+| Django | `logfire.instrument_django(app)` | Yes | `django` |
+| Flask | `logfire.instrument_flask(app)` | Yes | `flask` |
+| Starlette | `logfire.instrument_starlette(app)` | Yes | `starlette` |
+| AIOHTTP | `logfire.instrument_aiohttp_client()` | No | `aiohttp` |
+
+## HTTP Clients
+
+| Library | Instrumentor | Extra |
+|---------|-------------|-------|
+| httpx | `logfire.instrument_httpx()` | `httpx` |
+| requests | `logfire.instrument_requests()` | `requests` |
+
+## Databases
+
+| Library | Instrumentor | Extra |
+|---------|-------------|-------|
+| asyncpg | `logfire.instrument_asyncpg()` | `asyncpg` |
+| psycopg | `logfire.instrument_psycopg()` | `psycopg` |
+| psycopg2 | `logfire.instrument_psycopg2()` | `psycopg2` |
+| SQLAlchemy | `logfire.instrument_sqlalchemy()` | `sqlalchemy` |
+| PyMongo | `logfire.instrument_pymongo()` | `pymongo` |
+| MySQL | `logfire.instrument_mysql()` | `mysql` |
+| SQLite3 | `logfire.instrument_sqlite3()` | `sqlite3` |
+| Redis | `logfire.instrument_redis()` | `redis` |
+
+## AI/LLM Frameworks
+
+| Framework | Instrumentor | Extra |
+|-----------|-------------|-------|
+| PydanticAI | `logfire.instrument_pydantic_ai()` | `pydantic-ai` |
+| OpenAI | `logfire.instrument_openai()` | `openai` |
+| Anthropic | `logfire.instrument_anthropic()` | `anthropic` |
+| LiteLLM | `logfire.instrument_litellm()` | `litellm` |
+| DSPy | `logfire.instrument_dspy()` | `dspy` |
+| Google GenAI | `logfire.instrument_google_genai()` | `google-genai` |
+
+## Task Queues
+
+| Framework | Instrumentor | Extra |
+|-----------|-------------|-------|
+| Celery | `logfire.instrument_celery()` | `celery` |
+
+## Other
+
+| Feature | Instrumentor | Extra |
+|---------|-------------|-------|
+| System Metrics | `logfire.instrument_system_metrics()` | `system-metrics` |
+| Pydantic Models | `logfire.instrument_pydantic()` | - (built-in) |
+| AWS Lambda | handler wrapper | `aws-lambda` |
+
+## Gunicorn Configuration
+
+```python
+# gunicorn.conf.py
+import logfire
+
+def post_fork(server, worker):
+    logfire.configure()
+    logfire.instrument_fastapi(app)
+```

.agents/skills/instrumentation/references/python/logging-patterns.md

@@ -0,0 +1,101 @@
+# Python Logging Patterns
+
+## Log Levels
+
+From lowest to highest severity:
+
+```python
+logfire.trace("Detailed trace {detail}", detail=x)
+logfire.debug("Debug info {state}", state=s)
+logfire.info("Normal operation {event}", event=e)
+logfire.notice("Notable event {event}", event=e)
+logfire.warn("Warning {issue}", issue=i)
+logfire.error("Error occurred {error}", error=err)
+logfire.fatal("Fatal error {error}", error=err)
+```
+
+## Nested Spans
+
+Spans nest to create a tree visible in the Logfire UI. Use them to show the structure of an operation, not just that it happened:
+
+```python
+with logfire.span("HTTP request {method} {url}", method="POST", url=url):
+    with logfire.span("Serialize payload"):
+        payload = model.model_dump_json()
+    with logfire.span("Send request"):
+        response = await client.post(url, content=payload)
+    logfire.info("Response {status}", status=response.status_code)
+```
+
+## Standard Library Logging Integration
+
+For projects that already use Python's `logging` module, route existing log calls through Logfire rather than rewriting them all:
+
+```python
+from logging import basicConfig
+import logfire
+
+logfire.configure()
+basicConfig(handlers=[logfire.LogfireLoggingHandler()])
+```
+
+Or with `dictConfig`:
+
+```python
+from logging.config import dictConfig
+import logfire
+
+logfire.configure()
+dictConfig({
+    'version': 1,
+    'handlers': {
+        'logfire': {'class': 'logfire.LogfireLoggingHandler'},
+    },
+    'root': {'handlers': ['logfire']},
+})
+```
+
+## Suppressing Noisy Libraries
+
+Some libraries emit excessive debug logs. Silence them at the `logging` level:
+
+```python
+import logging
+
+logging.getLogger('httpcore').setLevel(logging.WARNING)
+logging.getLogger('httpx').setLevel(logging.WARNING)
+```
+
+## Custom Metrics
+
+For dashboards and alerting, create metrics:
+
+```python
+counter = logfire.metric_counter("orders_processed", unit="1")
+counter.add(1, {"status": "success"})
+
+histogram = logfire.metric_histogram("request_duration", unit="s")
+histogram.record(0.123, {"endpoint": "/api/users"})
+
+gauge = logfire.metric_gauge("active_connections")
+gauge.set(42)
+```
+
+## Testing with capfire
+
+Use the `capfire` pytest fixture to assert on emitted spans without sending data to production:
+
+```python
+from logfire.testing import CaptureLogfire
+
+def test_order_processing(capfire: CaptureLogfire) -> None:
+    process_order(order_id=123)
+
+    spans = capfire.exporter.exported_spans_as_dict()
+    assert any(
+        span['attributes'].get('order_id') == 123
+        for span in spans
+    )
+```
+
+Configure logfire with `send_to_logfire=False` in test fixtures to prevent production data leakage.

.agents/skills/instrumentation/references/rust/patterns.md

@@ -0,0 +1,106 @@
+# Rust Patterns
+
+## Core Macros
+
+The Rust SDK is built on `tracing` and `opentelemetry`. All `tracing` macros work automatically with Logfire.
+
+### Events (log points)
+
+```rust
+logfire::trace!("Detailed trace {detail}", detail = x);
+logfire::debug!("Debug info {state}", state = s);
+logfire::info!("Normal operation {event}", event = e);
+logfire::warn!("Warning {issue}", issue = i);
+logfire::error!("Error occurred {err}", err = e);
+```
+
+### Spans
+
+```rust
+// Scoped - span closes when closure completes
+logfire::span!("Processing order {order_id}", order_id = id).in_scope(|| {
+    let items = fetch_items(id);
+    logfire::info!("Fetched {count} items", count = items.len());
+    process_items(items)
+});
+
+// Guard-based - span closes when guard is dropped
+let _guard = logfire::span!("Long operation {job_id}", job_id = id).entered();
+do_work();
+// span ends when _guard goes out of scope
+```
+
+## Configuration
+
+```rust
+use logfire;
+
+fn main() -> Result<(), Box<dyn std::error::Error>> {
+    let shutdown_handler = logfire::configure()
+        .install_panic_handler()  // captures panics as error spans
+        .finish()?;
+
+    // application code...
+
+    shutdown_handler.shutdown()?;  // flush all pending spans
+    Ok(())
+}
+```
+
+Set `LOGFIRE_TOKEN` in your environment or use the Logfire CLI (`logfire auth`).
+
+## Tracing Crate Compatibility
+
+Any library using `tracing` macros automatically sends data through Logfire:
+
+```rust
+use tracing;
+
+tracing::info!("This also appears in Logfire");
+
+#[tracing::instrument]
+fn my_function(param: &str) {
+    // automatically creates a span with param as an attribute
+}
+```
+
+## Log Crate Integration
+
+The `log` crate is automatically captured and forwarded to Logfire. Libraries using `log::info!()`, `log::error!()`, etc. will appear in your Logfire dashboard without any additional configuration.
+
+## Async Spans
+
+```rust
+use tracing::Instrument;
+
+async fn process_order(order_id: u64) {
+    let span = logfire::span!("process order {order_id}", order_id = order_id);
+    async {
+        fetch_items(order_id).await;
+        logfire::info!("Order processed");
+    }
+    .instrument(span)
+    .await;
+}
+```
+
+## Shutdown
+
+Always call `shutdown()` before program exit to flush pending data:
+
+```rust
+// In main()
+let shutdown_handler = logfire::configure().finish()?;
+
+// ... app runs ...
+
+// Before exit
+shutdown_handler.shutdown()?;
+```
+
+For web servers using `tokio`, handle shutdown via signal:
+
+```rust
+tokio::signal::ctrl_c().await?;
+shutdown_handler.shutdown()?;
+```

.agents/skills/pr-create/SKILL.md

@@ -0,0 +1,114 @@
+---
+name: pr-create
+description: Use when creating or updating a Basic Memory pull request from Codex with BM Bossbot merge-gate monitoring.
+---
+
+# Create A Basic Memory PR
+
+Create or update a pull request for the current branch, then wait for BM
+Bossbot to approve the latest head SHA. This skill never merges a PR.
+
+## Inputs
+
+- Optional `<theme>`: free-form visual direction for the non-gating PR
+  image. Example: `$pr-create "Italian movie poster"`.
+- Treat `<theme>` as style guidance only. It must not affect PR readiness,
+  BM Bossbot review, status checks, or merge behavior.
+
+## How To Use
+
+Ask Codex to use the skill from a feature branch:
+
+```text
+$pr-create
+$pr-create "Italian movie poster"
+$pr-create "80's action movies"
+```
+
+Use the plain form when you only want the PR workflow. Pass a theme when you
+want the non-gating image to lean toward a particular visual direction. The
+theme can be specific ("Rembrandt-inspired approval scene") or broad ("let the
+model choose from BM categories").
+
+## What Happens
+
+1. Codex checks the branch, local verification, GitHub auth, commit sign-offs,
+   and semantic PR title shape.
+2. Codex pushes the branch, creates or reuses the PR, and adds the optional
+   `BM_INFOGRAPHIC_THEME` block when a theme was supplied.
+3. BM Bossbot runs from trusted base code, reviews sanitized PR metadata and
+   diff context, and sets the required `BM Bossbot Approval` status for the
+   exact head SHA.
+4. If approval succeeds, BM Bossbot may publish a non-gating image block and a
+   provenance block:
+
+   ```markdown
+   <!-- BM_INFOGRAPHIC_PROVENANCE:start -->
+   ...
+   <!-- BM_INFOGRAPHIC_PROVENANCE:end -->
+   ```
+
+   The provenance records the image mode, theme source, selected visual
+   direction, and image settings. It is for review/debugging context only.
+5. Codex reports the PR URL, head SHA, checks watched, verification run, and BM
+   Bossbot verdict.
+
+The skill never merges, never enables auto-merge, and never treats the image or
+provenance block as a gate. The only required merge signal is the
+`BM Bossbot Approval` status on the current PR head SHA.
+
+## Preflight
+
+1. Confirm the repo and branch:
+   - `git status --short --branch`
+   - stop if detached or on `main`
+   - keep unrelated user changes intact
+
+2. Confirm GitHub access:
+   - `gh auth status`
+   - `gh repo view --json nameWithOwner,defaultBranchRef,url`
+
+3. Check PR readiness:
+   - commits are signed off with `git commit -s`
+   - title uses the repo semantic format
+   - local verification appropriate to the change has run
+
+## Create Or Reuse
+
+1. Push the branch:
+   - `git push -u origin HEAD`
+
+2. Check for an existing PR:
+   - `gh pr view --json number,url,headRefOid,mergeStateStatus,statusCheckRollup`
+
+3. If no PR exists, create one:
+   - `gh pr create --fill`
+   - adjust the title if it does not satisfy the semantic PR title workflow
+
+4. If `<theme>` is provided, add or update this managed block in the PR body:
+
+```markdown
+<!-- BM_INFOGRAPHIC_THEME:start -->
+<theme>
+<!-- BM_INFOGRAPHIC_THEME:end -->
+```
+
+Keep the rest of the PR body intact. The theme is non-gating image guidance
+only.
+
+5. Do not merge. Do not enable auto-merge.
+
+## Watch The Gate
+
+1. Trigger or wait for `.github/workflows/bm-bossbot.yml`.
+2. Watch the required commit status named `BM Bossbot Approval`.
+3. Treat approval as valid only when it is green for the current `headRefOid`.
+4. If the branch changes after approval, wait for BM Bossbot to review the new
+   head SHA.
+5. If BM Bossbot fails or requests changes, use `$fix-pr-issues`.
+
+## Report
+
+Return the PR URL, current head SHA, checks watched, verification run, and the
+BM Bossbot verdict. Include the image `<theme>` if one was supplied. Be
+explicit when any check is still pending.

.agents/skills/pr-create/agents/openai.yaml

@@ -0,0 +1,7 @@
+interface:
+  display_name: "PR Create"
+  short_description: "Create PRs and wait for BM Bossbot"
+  icon_small: "./assets/icon.svg"
+  icon_large: "./assets/icon.svg"
+  brand_color: "#2563EB"
+  default_prompt: "Use $pr-create to create or update this Basic Memory PR and wait for BM Bossbot Approval."

.agents/skills/pr-create/assets/icon.svg

@@ -0,0 +1,5 @@
+<svg xmlns="http://www.w3.org/2000/svg" width="128" height="128" viewBox="0 0 24 24" fill="none" stroke="#111827" stroke-width="1.8" stroke-linecap="round" stroke-linejoin="round">
+  <path d="M3 12h4l2-6 4 12 2-6h6"/>
+  <path d="M4 20h16"/>
+</svg>
+

.claude-plugin/marketplace.json

@@ -0,0 +1,29 @@
+{
+  "name": "basicmachines-co",
+  "owner": {
+    "name": "Basic Machines",
+    "email": "hello@basicmachines.co"
+  },
+  "metadata": {
+    "description": "Official Basic Memory plugins from the canonical basic-memory repository",
+    "version": "0.22.0"
+  },
+  "plugins": [
+    {
+      "name": "basic-memory",
+      "source": "./plugins/claude-code",
+      "description": "The bridge between Claude's working memory and Basic Memory's durable knowledge graph \u2014 session briefings, pre-compaction checkpoints, and capture reflexes",
+      "version": "0.22.0",
+      "author": {
+        "name": "Basic Machines"
+      },
+      "keywords": [
+        "memory",
+        "knowledge",
+        "mcp",
+        "specs",
+        "context"
+      ]
+    }
+  ]
+}

.claude/commands/release/beta.md

@@ -30,7 +30,8 @@ The justfile target handles:
 - ✅ Beta version format validation (supports b1, b2, rc1, etc.)
 - ✅ Git status and branch checks
 - ✅ Quality checks (`just check` - lint, format, type-check, tests)
-- ✅ Version update in `src/basic_memory/__init__.py`
+- ✅ Version update across all consolidated manifests via `just set-version` (Python
+  package + Claude Code plugin/marketplaces + Codex plugin + Hermes + OpenClaw)
 - ✅ Automatic commit with proper message
 - ✅ Tag creation and pushing to GitHub
 - ✅ Beta release workflow trigger
@@ -90,6 +91,6 @@ Monitor release: https://github.com/basicmachines-co/basic-memory/actions
 - Beta releases are pre-releases for testing new features
 - Automatically published to PyPI with pre-release flag
 - Uses the automated justfile target for consistency
-- Version is automatically updated in `__init__.py`
+- Version is automatically updated across all consolidated manifests via `just set-version`
 - Ideal for validating changes before stable release
-- Supports both beta (b1, b2) and release candidate (rc1, rc2) versions
+- Supports both beta (b1, b2) and release candidate (rc1, rc2) versions

.claude/commands/release/release.md

@@ -42,7 +42,8 @@ The justfile target handles:
 - ✅ Version format validation
 - ✅ Git status and branch checks
 - ✅ Quality checks (`just check` - lint, format, type-check, tests)
-- ✅ Version update in `src/basic_memory/__init__.py`
+- ✅ Version update across all consolidated manifests via `just set-version` (Python
+  package + Claude Code plugin/marketplaces + Codex plugin + Hermes + OpenClaw)
 - ✅ Automatic commit with proper message
 - ✅ Tag creation and pushing to GitHub
 - ✅ Release workflow trigger (automatic on tag push)
@@ -191,9 +192,16 @@ Users can now upgrade:
 - This creates production releases used by end users
 - Must pass all quality gates before proceeding
 - Uses the automated justfile target for consistency
-- Version is automatically updated in `__init__.py` and `server.json`
+- Version is automatically updated across **all** consolidated manifests via
+  `just set-version <version>` (which calls `scripts/update_versions.py`): the
+  Python package (`__init__.py`, `server.json`) **and** the plugin/agent artifacts
+  (Claude Code `plugin.json` + root/local marketplaces, Codex `plugin.json`,
+  Hermes `plugin.yaml` + `__init__.py`, OpenClaw `package.json`). To bump only
+  the plugin/agent artifacts
+  out of band, use `just set-packages-version <version>` (preview with
+  `just set-packages-version-dry-run <version>`).
 - Triggers automated GitHub release with changelog
 - Package is published to PyPI for `pip` and `uv` users
 - Homebrew formula is automatically updated for stable releases
 - MCP Registry is updated manually via `mcp-publisher publish`
-- Supports multiple installation methods (uv, pip, Homebrew)
+- Supports multiple installation methods (uv, pip, Homebrew)

.claude/settings.json

@@ -1,3 +1,21 @@
 {
-  "enabledPlugins": {}
+  "$schema": "https://json.schemastore.org/claude-code-settings.json",
+  "env": {
+    "CLAUDE_BASH_MAINTAIN_PROJECT_WORKING_DIR": "1",
+    "CLAUDE_CODE_DISABLE_FEEDBACK_SURVEY": "1",
+    "CLAUDE_CODE_NO_FLICKER": "1",
+    "CLAUDE_CODE_DISABLE_ADAPTIVE_THINKING": "1"
+  },
+  "permissions": {
+    "allow": [
+      "Bash(just fast-check)",
+      "Bash(just check)",
+      "Bash(just fix)",
+      "Bash(just typecheck)",
+      "Bash(just lint)",
+      "Bash(just test)"
+    ],
+    "deny": []
+  },
+  "enableAllProjectMcpServers": true
 }

.claude/skills/adversarial-review

@@ -0,0 +1 @@
+../../.agents/skills/adversarial-review

.claude/skills/basic-machines-review

@@ -0,0 +1 @@
+../../.agents/skills/basic-machines-review

.claude/skills/instrumentation

@@ -0,0 +1 @@
+../../.agents/skills/instrumentation

.github/basic-memory/SOUL.md

@@ -0,0 +1,25 @@
+# Auto BM Soul
+
+Write project updates for humans who will return later trying to understand what happened.
+
+## Voice
+
+- Clear, direct, warm, and technically honest.
+- Prefer concrete observations over generic praise.
+- It is okay to say when code is messy, risky, clever, boring, or satisfying.
+- Keep personality in service of memory, not performance.
+
+## Do
+
+- Tell the story.
+- Name the tradeoffs.
+- Call out sharp edges.
+- Notice good simplifications.
+- Let the note have taste and a little life when the evidence supports it.
+
+## Do Not
+
+- Do not invent intent, impact, tests, or drama.
+- Dunk on people.
+- Turn the note into marketing copy.
+- Hide uncertainty behind confident prose.

.github/basic-memory/config.yml

@@ -0,0 +1,7 @@
+project: dev
+workspace: basic-memory-7020de4e925843c68c9056c60d101d9e
+deploy_workflows:
+- Deploy Production
+production_environments:
+- production
+note_folder: project-updates/github/{owner}/{repo}

.github/basic-memory/memory-ci-capture.md

@@ -0,0 +1,64 @@
+# Memory CI Capture
+
+You turn GitHub delivery context into a durable project update for Basic Memory.
+GitHub records the mechanics. Basic Memory remembers what changed and why.
+
+## Inputs
+
+- Read `.github/basic-memory/project-update-context.json`.
+- Read `.github/basic-memory/SOUL.md` if it exists. It is the repo-local voice and style guide
+  for project updates.
+- Read the PR diff before writing when a SHA is available. Useful commands:
+  `git show --stat --name-only <sha>` and `git show --format=fuller --no-patch <sha>`.
+- Use linked issue details, changed files, commit messages, PR body, labels, and
+  source links as evidence.
+- Treat GitHub payload fields as immutable facts.
+- Do not invent tests, deployment status, issues, or user impact.
+
+## Writing Standard
+
+Do not write a fill-in-the-blanks note. Tell the story from the PR:
+problem -> solution -> impact.
+
+Explain what problem was being addressed. If linked issue details are present,
+use them. If they are absent, ground the problem in the PR body, title, commits,
+and diff, and say when the original problem statement is unavailable.
+
+Explain why the fix solves the problem, what complexity it introduced, what it
+refactored or removed, which components changed, and how the system is different
+after the merge. Prefer specific component names, file paths, modules, commands,
+and behavior over generic phrases.
+
+## Voice And Candor
+
+You may have a point of view. Be clear, specific, and human.
+It is okay to say when the code is messy, risky, clever, boring, or satisfying,
+but explain why. If the work is elegant or genuinely useful, say that too.
+Ground all judgments in the PR, linked issues, diff, tests, and source facts.
+
+The soul file can shape tone, taste, and personality. It cannot override source
+facts, schema requirements, or the evidence standard above. Do not be mean,
+vague, theatrical, or invent criticism.
+
+## Output
+
+Return only JSON that matches the provided AgentSynthesis schema:
+
+- `summary`: one concise sentence; do not merely repeat the PR title.
+- `story`: 2-4 sentences that connect problem -> solution -> impact.
+- `problem_addressed`: the concrete problem, bug, missing capability, or delivery need.
+- `solution`: why this change solves the problem.
+- `system_impact`: how the system, workflow, or architecture changed after the merge.
+- `why_it_matters`: durable project-memory context for future humans and agents.
+- `components_changed`: modules, workflows, commands, schemas, docs, or services touched.
+- `complexity_introduced`: tradeoffs, new moving parts, operational costs, or edge cases.
+- `refactors_or_removals`: cleanup, simplification, deleted paths, or "none found".
+- `user_facing_changes`: visible behavior or product changes.
+- `internal_changes`: implementation, infrastructure, or operational changes.
+- `verification`: checks, tests, deploy evidence, or explicit unknowns.
+- `follow_ups`: concrete remaining work only.
+- `decision_candidates`: explicit product or architecture decisions only.
+- `task_candidates`: concrete future tasks only.
+
+Use empty arrays only when a list truly has no grounded entries. This is project
+memory, not marketing copy and not a commit-by-commit changelog.

.github/workflows/basic-memory.yml

@@ -0,0 +1,75 @@
+name: Basic Memory Project Updates
+
+"on":
+  pull_request:
+    types: [closed]
+  workflow_run:
+    workflows: ["Deploy Production"]
+    types: [completed]
+
+jobs:
+  project-update:
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      pull-requests: read
+      issues: read
+      actions: read
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v6
+
+      - name: Set up Python
+        uses: actions/setup-python@v6
+        with:
+          python-version: "3.12"
+
+      - name: Install Basic Memory from checkout
+        run: |
+          python -m pip install --upgrade pip
+          pip install -e .
+
+      - name: Collect project update context
+        id: collect
+        env:
+          GITHUB_TOKEN: ${{ github.token }}
+        run: |
+          bm ci collect \
+            --config .github/basic-memory/config.yml \
+            --output .github/basic-memory/project-update-context.json
+
+      - name: Stop when event is not eligible
+        if: steps.collect.outputs.eligible != 'true'
+        run: |
+          echo "Auto BM skipped: ${{ steps.collect.outputs.skip_reason }}"
+
+      - name: Write Codex output schema
+        if: steps.collect.outputs.eligible == 'true'
+        run: |
+          bm ci agent-schema --output "${{ runner.temp }}/agent-synthesis.schema.json"
+
+      - name: Synthesize project update with Codex
+        if: steps.collect.outputs.eligible == 'true'
+        uses: openai/codex-action@v1
+        with:
+          openai-api-key: ${{ secrets.OPENAI_API_KEY }}
+          prompt-file: .github/basic-memory/memory-ci-capture.md
+          output-file: ${{ runner.temp }}/agent-synthesis.json
+          output-schema-file: ${{ runner.temp }}/agent-synthesis.schema.json
+          sandbox: read-only
+          safety-strategy: drop-sudo
+
+      - name: Publish project update
+        if: steps.collect.outputs.eligible == 'true'
+        env:
+          BASIC_MEMORY_CLOUD_API_KEY: ${{ secrets.BASIC_MEMORY_API_KEY }}
+          BASIC_MEMORY_CI_CLOUD_HOST: ${{ vars.BASIC_MEMORY_CLOUD_HOST }}
+        run: |
+          if [ -n "$BASIC_MEMORY_CI_CLOUD_HOST" ]; then
+            export BASIC_MEMORY_CLOUD_HOST="$BASIC_MEMORY_CI_CLOUD_HOST"
+          fi
+          bm ci publish \
+            --cloud \
+            --config .github/basic-memory/config.yml \
+            --context .github/basic-memory/project-update-context.json \
+            --synthesis "${{ runner.temp }}/agent-synthesis.json"

.github/workflows/claude-code-review.yml

@@ -1,23 +1,18 @@
 name: Claude Code Review
 
-on:
-  pull_request:
-    types: [opened, synchronize]
-    # Optional: Only run on specific file changes
-    # paths:
-    #   - "src/**/*.ts"
-    #   - "src/**/*.tsx"
-    #   - "src/**/*.js"
-    #   - "src/**/*.jsx"
+"on":
+  workflow_dispatch:
+    inputs:
+      pr_number:
+        description: Pull request number to review manually
+        required: true
+
+env:
+  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
 
 jobs:
   claude-review:
-    # Only run for organization members and collaborators
-    if: |
-      github.event.pull_request.author_association == 'OWNER' ||
-      github.event.pull_request.author_association == 'MEMBER' ||
-      github.event.pull_request.author_association == 'COLLABORATOR'
-
+    if: inputs.pr_number != ''
     runs-on: ubuntu-latest
     permissions:
       contents: read
@@ -27,7 +22,7 @@ jobs:
 
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
         with:
           fetch-depth: 1
 
@@ -40,7 +35,14 @@ jobs:
           track_progress: true  # Enable visual progress tracking
           allowed_bots: '*'
           prompt: |
-            Review this Basic Memory PR against our team checklist:
+            Review Basic Memory PR #${{ inputs.pr_number }} as an advisory manual review.
+
+            Use `gh pr view ${{ inputs.pr_number }}` and related `gh pr`/`gh api`
+            commands to inspect the pull request. Do not merge the PR and do not
+            treat this advisory review as the required merge gate. BM Bossbot owns
+            the required `BM Bossbot Approval` status.
+
+            Review the PR against our team checklist:
 
             ## Code Quality & Standards
             - [ ] Follows Basic Memory's coding conventions in CLAUDE.md

.github/workflows/claude-issue-triage.yml

@@ -4,6 +4,9 @@ on:
   issues:
     types: [opened]
 
+env:
+  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
+
 jobs:
   triage:
     runs-on: ubuntu-latest
@@ -12,7 +15,7 @@ jobs:
       id-token: write
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
         with:
           fetch-depth: 1
 
@@ -68,4 +71,4 @@ jobs:
 
             Read the issue carefully and provide helpful triage with appropriate labels.
 
-          claude_args: '--allowed-tools "Bash(gh issue:*),Bash(gh search:*),Read"'
+          claude_args: '--allowed-tools "Bash(gh issue:*),Bash(gh search:*),Read"'

.github/workflows/claude.yml

@@ -12,6 +12,9 @@ on:
   pull_request_target:
     types: [opened, synchronize]
 
+env:
+  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
+
 jobs:
   claude:
     if: |
@@ -41,7 +44,7 @@ jobs:
       actions: read # Required for Claude to read CI results on PRs
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
         with:
           # For pull_request_target, checkout the PR head to review the actual changes
           ref: ${{ github.event_name == 'pull_request_target' && github.event.pull_request.head.sha || github.sha }}
@@ -65,4 +68,3 @@ jobs:
           # See https://github.com/anthropics/claude-code-action/blob/main/docs/usage.md
           # or https://docs.claude.com/en/docs/claude-code/sdk#command-line for available options
           # claude_args: '--model claude-opus-4-1-20250805 --allowed-tools Bash(gh pr:*)'
-

.github/workflows/consolidated-packages.yml

@@ -0,0 +1,110 @@
+name: Consolidated Packages
+
+on:
+  pull_request:
+    paths:
+      - ".claude-plugin/**"
+      - "plugins/**"
+      - "skills/**"
+      - "integrations/**"
+      - "scripts/update_versions.py"
+      - "scripts/validate_*.py"
+      - "justfile"
+      - ".github/workflows/consolidated-packages.yml"
+  push:
+    branches:
+      - main
+    paths:
+      - ".claude-plugin/**"
+      - "plugins/**"
+      - "skills/**"
+      - "integrations/**"
+      - "scripts/update_versions.py"
+      - "scripts/validate_*.py"
+      - "justfile"
+      - ".github/workflows/consolidated-packages.yml"
+
+permissions:
+  contents: read
+
+jobs:
+  claude-code:
+    name: Claude Code marketplace
+    permissions:
+      contents: read
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: extractions/setup-just@v4
+
+      - name: Validate manifests, agent, and bundled skills
+        run: just --justfile plugins/claude-code/justfile --working-directory plugins/claude-code ci-check
+
+      - name: Verify shared version dry run
+        run: just release-dry-run v0.99.0
+
+  skills:
+    name: Shared skills
+    permissions:
+      contents: read
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: extractions/setup-just@v4
+
+      - name: Validate SKILL.md source
+        run: just --justfile skills/justfile --working-directory skills check
+
+  hermes:
+    name: Hermes unit tests
+    permissions:
+      contents: read
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: integrations/hermes
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: extractions/setup-just@v4
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v3
+
+      - name: Set up Python
+        run: uv python install 3.12
+
+      - name: Validate manifest and run unit tests
+        run: just check
+
+  openclaw:
+    name: OpenClaw package
+    permissions:
+      contents: read
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: integrations/openclaw
+    steps:
+      - uses: actions/checkout@v4
+
+      - uses: extractions/setup-just@v4
+
+      - name: Setup Bun
+        uses: oven-sh/setup-bun@v2
+        with:
+          bun-version: "1.3.8"
+
+      - name: Setup Node
+        uses: actions/setup-node@v4
+        with:
+          node-version: "24"
+          registry-url: "https://registry.npmjs.org"
+
+      - name: Install dependencies
+        run: bun install --frozen-lockfile
+
+      - name: Release readiness
+        run: just release-check

.github/workflows/dev-release.yml

@@ -5,6 +5,9 @@ on:
     branches: [main]
   workflow_dispatch:  # Allow manual triggering
 
+env:
+  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
+
 jobs:
   dev-release:
     runs-on: ubuntu-latest
@@ -13,12 +16,12 @@ jobs:
       contents: write
 
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
         with:
           fetch-depth: 0
 
       - name: Set up Python
-        uses: actions/setup-python@v5
+        uses: actions/setup-python@v6
         with:
           python-version: "3.12"
 
@@ -50,4 +53,4 @@ jobs:
         uses: pypa/gh-action-pypi-publish@release/v1
         with:
           password: ${{ secrets.PYPI_TOKEN }}
-          skip-existing: true  # Don't fail if version already exists
+          skip-existing: true  # Don't fail if version already exists

.github/workflows/docker.yml

@@ -9,27 +9,27 @@ on:
 env:
   REGISTRY: ghcr.io
   IMAGE_NAME: basicmachines-co/basic-memory
+  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
 
 jobs:
   docker:
-    runs-on: ubuntu-latest
+    runs-on: depot-ubuntu-24.04
     permissions:
       contents: read
+      id-token: write
       packages: write
 
     steps:
       - name: Checkout repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@v6
         with:
           fetch-depth: 0
 
-      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
-        with:
-          platforms: linux/amd64,linux/arm64
+      - name: Set up Depot
+        uses: depot/setup-action@v1
 
       - name: Log in to GitHub Container Registry
-        uses: docker/login-action@v3
+        uses: docker/login-action@v4
         with:
           registry: ${{ env.REGISTRY }}
           username: ${{ github.actor }}
@@ -37,7 +37,7 @@ jobs:
 
       - name: Extract metadata
         id: meta
-        uses: docker/metadata-action@v5
+        uses: docker/metadata-action@v6
         with:
           images: ${{ env.REGISTRY }}/${{ env.IMAGE_NAME }}
           tags: |
@@ -48,14 +48,12 @@ jobs:
             type=raw,value=latest,enable={{is_default_branch}}
 
       - name: Build and push Docker image
-        uses: docker/build-push-action@v5
+        uses: depot/build-push-action@v1
         with:
+          project: ${{ vars.DEPOT_BASIC_MEMORY_PROJECT_ID || vars.DEPOT_PROJECT_ID }}
           context: .
           file: ./Dockerfile
           platforms: linux/amd64,linux/arm64
           push: true
           tags: ${{ steps.meta.outputs.tags }}
           labels: ${{ steps.meta.outputs.labels }}
-          cache-from: type=gha
-          cache-to: type=gha,mode=max
-

.github/workflows/pr-title.yml

@@ -7,11 +7,14 @@ on:
       - edited
       - synchronize
 
+env:
+  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
+
 jobs:
   main:
     runs-on: ubuntu-latest
     steps:
-      - uses: amannn/action-semantic-pull-request@v5
+      - uses: amannn/action-semantic-pull-request@v6
         env:
           GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
         with:
@@ -35,7 +38,11 @@ jobs:
             mcp
             sync
             ui
+            ci
             deps
             installer
+            plugins
+            skills
+            integrations
           # Allow breaking changes (needs "!" after type/scope)
-          requireScopeForBreakingChange: true
+          requireScopeForBreakingChange: true

.github/workflows/release.yml

@@ -5,6 +5,9 @@ on:
     tags:
       - 'v*'  # Trigger on version tags like v1.0.0, v0.13.0, etc.
 
+env:
+  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
+
 jobs:
   release:
     runs-on: ubuntu-latest
@@ -13,12 +16,12 @@ jobs:
       contents: write
 
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
         with:
           fetch-depth: 0
 
       - name: Set up Python
-        uses: actions/setup-python@v5
+        uses: actions/setup-python@v6
         with:
           python-version: "3.12"
 
@@ -39,7 +42,7 @@ jobs:
           echo "Build completed successfully"
 
       - name: Create GitHub Release
-        uses: softprops/action-gh-release@v2
+        uses: softprops/action-gh-release@v3
         with:
           files: |
             dist/*.whl
@@ -53,6 +56,46 @@ jobs:
         with:
           password: ${{ secrets.PYPI_TOKEN }}
 
+  openclaw:
+    name: Publish OpenClaw npm Package
+    needs: release
+    # npm publishes only for stable product tags. Pre-release Python tags use
+    # versions like 0.21.3b1, which are not valid npm pre-release semver.
+    if: ${{ !contains(github.ref_name, 'dev') && !contains(github.ref_name, 'b') && !contains(github.ref_name, 'rc') }}
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      id-token: write
+    defaults:
+      run:
+        working-directory: integrations/openclaw
+    steps:
+      - uses: actions/checkout@v6
+
+      - uses: extractions/setup-just@v4
+
+      - name: Setup Bun
+        uses: oven-sh/setup-bun@v2
+        with:
+          bun-version: "1.3.8"
+
+      - name: Setup Node
+        uses: actions/setup-node@v4
+        with:
+          node-version: "24"
+          registry-url: "https://registry.npmjs.org"
+
+      - name: Install dependencies
+        run: bun install --frozen-lockfile
+
+      - name: Release readiness
+        run: just release-check
+
+      - name: Publish to npm
+        run: npm publish --access public --provenance
+        env:
+          NODE_AUTH_TOKEN: ${{ secrets.NPM_TOKEN }}
+
   homebrew:
     name: Update Homebrew Formula
     needs: release
@@ -60,26 +103,63 @@ jobs:
     # Only run for stable releases (not dev, beta, or rc versions)
     if: ${{ !contains(github.ref_name, 'dev') && !contains(github.ref_name, 'b') && !contains(github.ref_name, 'rc') }}
     permissions:
-      contents: write
-      actions: read
+      contents: read
     steps:
+      # Inline bump replaces mislav/bump-homebrew-formula-action@v4.x.
+      # The action does a HEAD request to api.github.com /repos/.../tarball/<ref>
+      # with the bearer token and expects a 302 redirect. GitHub now returns
+      # 303 on that endpoint when authenticated, which the action treats as a
+      # fatal error. Re-implementing the bump as plain git+sed keeps the same
+      # contract (update url + sha256, commit, push) with no third-party action.
       - name: Update Homebrew formula
-        uses: mislav/bump-homebrew-formula-action@v3
-        with:
-          # Formula name in homebrew-basic-memory repo
-          formula-name: basic-memory
-          # The tap repository
-          homebrew-tap: basicmachines-co/homebrew-basic-memory
-          # Base branch of the tap repository
-          base-branch: main
-          # Download URL will be automatically constructed from the tag
-          download-url: https://github.com/basicmachines-co/basic-memory/archive/refs/tags/${{ github.ref_name }}.tar.gz
-          # Commit message for the formula update
-          commit-message: |
-            {{formulaName}} {{version}}
-
-            Created by https://github.com/basicmachines-co/basic-memory/actions/runs/${{ github.run_id }}
         env:
-          # Personal Access Token with repo scope for homebrew-basic-memory repo
-          COMMITTER_TOKEN: ${{ secrets.HOMEBREW_TOKEN }}
+          HOMEBREW_TOKEN: ${{ secrets.HOMEBREW_TOKEN }}
+          REF: ${{ github.ref_name }}
+          REPO: ${{ github.repository }}
+          RUN_URL: https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }}
+        run: |
+          set -euo pipefail
 
+          VERSION="${REF#v}"
+          ARCHIVE_URL="https://github.com/${REPO}/archive/refs/tags/${REF}.tar.gz"
+
+          echo "::group::Compute tarball sha256"
+          SHA256="$(curl --fail --silent --location "$ARCHIVE_URL" | sha256sum | awk '{print $1}')"
+          test -n "$SHA256"
+          echo "sha256: $SHA256"
+          echo "::endgroup::"
+
+          echo "::group::Clone tap"
+          git clone \
+            --depth 1 \
+            "https://x-access-token:${HOMEBREW_TOKEN}@github.com/basicmachines-co/homebrew-basic-memory.git" \
+            tap
+          cd tap
+          git config user.name "github-actions[bot]"
+          git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
+          echo "::endgroup::"
+
+          echo "::group::Patch Formula/basic-memory.rb"
+          # Pipe-delimited sed because the URL contains slashes. The Formula
+          # only has one `url` and one `sha256` directive, so a first-match
+          # replacement is unambiguous. POSIX character classes ([[:space:]])
+          # keep this portable across BSD and GNU sed.
+          sed -i -E \
+            -e "s|^([[:space:]]*url[[:space:]]+)\"[^\"]+\"|\1\"${ARCHIVE_URL}\"|" \
+            -e "s|^([[:space:]]*sha256[[:space:]]+)\"[^\"]+\"|\1\"${SHA256}\"|" \
+            Formula/basic-memory.rb
+          git --no-pager diff Formula/basic-memory.rb
+          echo "::endgroup::"
+
+          if git diff --quiet Formula/basic-memory.rb; then
+            echo "Formula already at ${REF}; nothing to do."
+            exit 0
+          fi
+
+          echo "::group::Commit & push"
+          git add Formula/basic-memory.rb
+          git commit -m "basic-memory ${VERSION}
+
+          Created by ${RUN_URL}"
+          git push origin HEAD:main
+          echo "::endgroup::"

.github/workflows/test.yml

@@ -5,24 +5,62 @@ concurrency:
   cancel-in-progress: true
 
 on:
+  # Trigger: PR branch pushes already publish commit statuses that show up on the PR.
+  # Why: running the full matrix on both push and pull_request doubles CI time for the
+  #      exact same branch head commit.
+  # Outcome: each branch push runs the test suite once, including PR updates.
   push:
-    branches: [ "main" ]
-  pull_request:
-    branches: [ "main" ]
+
+env:
+  FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: "true"
+  # Branch builds (PRs arrive as push events — this workflow has no
+  # pull_request trigger) select only impacted tests from the cached testmon
+  # baseline (branch cache falling back to main's full-run recording). Pushes
+  # to main run the full suite with --testmon-noselect to refresh the baseline.
+  BASIC_MEMORY_TESTMON_FLAGS: ${{ github.ref_name == 'main' && '--testmon-noselect' || '--testmon --testmon-forceselect' }}
 
 jobs:
+  changes:
+    # Docs/workflow-only changes skip the entire test matrix while the workflow
+    # still concludes successfully, so the BM Bossbot gate (workflow_run on
+    # Tests success) keeps firing and the PR stays mergeable.
+    name: Detect code changes
+    runs-on: ubuntu-latest
+    outputs:
+      code: ${{ steps.filter.outputs.code }}
+    steps:
+      - uses: actions/checkout@v6
+      - id: filter
+        uses: dorny/paths-filter@v3
+        with:
+          # Tests only runs on push events; for branch pushes compare against
+          # main (merge-base), for main pushes dorny diffs the push range.
+          base: main
+          filters: |
+            code:
+              - 'src/**'
+              - 'tests/**'
+              - 'test-int/**'
+              - 'alembic/**'
+              - 'pyproject.toml'
+              - 'uv.lock'
+              - 'justfile'
+              - '.github/workflows/test.yml'
+
   static-checks:
+    needs: changes
+    if: needs.changes.outputs.code == 'true'
     name: Static Checks (Python 3.12)
     timeout-minutes: 20
     runs-on: ubuntu-latest
 
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
         with:
           submodules: true
 
       - name: Set up Python 3.12
-        uses: actions/setup-python@v4
+        uses: actions/setup-python@v6
         with:
           python-version: "3.12"
           cache: "pip"
@@ -31,7 +69,7 @@ jobs:
         run: |
           pip install uv
 
-      - uses: extractions/setup-just@v3
+      - uses: extractions/setup-just@v4
 
       - name: Create virtual env
         run: |
@@ -50,9 +88,10 @@ jobs:
           just lint
 
   test-sqlite-unit:
+    needs: changes
+    if: needs.changes.outputs.code == 'true'
     name: Test SQLite Unit (${{ matrix.os }}, Python ${{ matrix.python-version }})
-    timeout-minutes: 30
-    needs: [static-checks]
+    timeout-minutes: 45
     strategy:
       fail-fast: false
       matrix:
@@ -61,6 +100,8 @@ jobs:
             python-version: "3.12"
           - os: ubuntu-latest
             python-version: "3.13"
+          # Python 3.14 unit tests are the longest full-suite slice; keep this
+          # one on GitHub-hosted runners after Depot terminated it mid-suite.
           - os: ubuntu-latest
             python-version: "3.14"
           - os: windows-latest
@@ -68,12 +109,12 @@ jobs:
     runs-on: ${{ matrix.os }}
 
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
         with:
           submodules: true
 
       - name: Set up Python ${{ matrix.python-version }}
-        uses: actions/setup-python@v4
+        uses: actions/setup-python@v6
         with:
           python-version: ${{ matrix.python-version }}
           cache: 'pip'
@@ -82,7 +123,20 @@ jobs:
         run: |
           pip install uv
 
-      - uses: extractions/setup-just@v3
+      - uses: extractions/setup-just@v4
+
+      - name: Cache pytest-testmon results
+        uses: actions/cache@v4
+        with:
+          path: |
+            .testmondata
+            .testmondata-shm
+            .testmondata-wal
+          key: ${{ runner.os }}-testmon-sqlite-unit-py${{ matrix.python-version }}-${{ github.ref_name }}-${{ github.run_id }}
+          restore-keys: |
+            ${{ runner.os }}-testmon-sqlite-unit-py${{ matrix.python-version }}-${{ github.ref_name }}-
+            ${{ runner.os }}-testmon-sqlite-unit-py${{ matrix.python-version }}-main-
+            ${{ runner.os }}-testmon-sqlite-unit-py${{ matrix.python-version }}-
 
       - name: Create virtual env
         run: |
@@ -97,9 +151,10 @@ jobs:
           just test-unit-sqlite
 
   test-sqlite-integration:
+    needs: changes
+    if: needs.changes.outputs.code == 'true'
     name: Test SQLite Integration (${{ matrix.os }}, Python ${{ matrix.python-version }})
     timeout-minutes: 45
-    needs: [static-checks]
     strategy:
       fail-fast: false
       matrix:
@@ -115,12 +170,12 @@ jobs:
     runs-on: ${{ matrix.os }}
 
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
         with:
           submodules: true
 
       - name: Set up Python ${{ matrix.python-version }}
-        uses: actions/setup-python@v4
+        uses: actions/setup-python@v6
         with:
           python-version: ${{ matrix.python-version }}
           cache: 'pip'
@@ -129,7 +184,20 @@ jobs:
         run: |
           pip install uv
 
-      - uses: extractions/setup-just@v3
+      - uses: extractions/setup-just@v4
+
+      - name: Cache pytest-testmon results
+        uses: actions/cache@v4
+        with:
+          path: |
+            .testmondata
+            .testmondata-shm
+            .testmondata-wal
+          key: ${{ runner.os }}-testmon-sqlite-integration-py${{ matrix.python-version }}-${{ github.ref_name }}-${{ github.run_id }}
+          restore-keys: |
+            ${{ runner.os }}-testmon-sqlite-integration-py${{ matrix.python-version }}-${{ github.ref_name }}-
+            ${{ runner.os }}-testmon-sqlite-integration-py${{ matrix.python-version }}-main-
+            ${{ runner.os }}-testmon-sqlite-integration-py${{ matrix.python-version }}-
 
       - name: Create virtual env
         run: |
@@ -144,27 +212,44 @@ jobs:
           just test-int-sqlite
 
   test-postgres-unit:
-    name: Test Postgres Unit (Python ${{ matrix.python-version }})
-    timeout-minutes: 30
-    needs: [static-checks]
+    needs: changes
+    if: needs.changes.outputs.code == 'true'
+    name: Test Postgres Unit (Python ${{ matrix.python-version }}, shard ${{ matrix.group }}/3)
+    timeout-minutes: 60
     strategy:
       fail-fast: false
       matrix:
-        include:
-          - python-version: "3.12"
-          - python-version: "3.13"
-          - python-version: "3.14"
+        # Shard the largest suite across parallel jobs: each shard is a full job
+        # with its own Postgres service running 1/3 of the collection.
+        # Postgres runs on the latest Python only — the SQLite matrix carries
+        # Python-version coverage; Postgres carries backend coverage.
+        group: [1, 2, 3]
+        python-version: ["3.14"]
     runs-on: ubuntu-latest
-
-    # Note: No services section needed - testcontainers handles Postgres in Docker
+    services:
+      postgres:
+        image: pgvector/pgvector:pg16
+        env:
+          POSTGRES_USER: basic_memory_user
+          POSTGRES_PASSWORD: dev_password
+          POSTGRES_DB: basic_memory_test
+        ports:
+          - 5432:5432
+        options: >-
+          --health-cmd "pg_isready -U basic_memory_user -d basic_memory_test"
+          --health-interval 10s
+          --health-timeout 5s
+          --health-retries 5
+    env:
+      BASIC_MEMORY_TEST_POSTGRES_URL: postgresql://basic_memory_user:dev_password@127.0.0.1:5432/basic_memory_test
 
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
         with:
           submodules: true
 
       - name: Set up Python ${{ matrix.python-version }}
-        uses: actions/setup-python@v4
+        uses: actions/setup-python@v6
         with:
           python-version: ${{ matrix.python-version }}
           cache: 'pip'
@@ -173,7 +258,21 @@ jobs:
         run: |
           pip install uv
 
-      - uses: extractions/setup-just@v3
+      - uses: extractions/setup-just@v4
+
+      - name: Cache pytest-testmon results
+        uses: actions/cache@v4
+        with:
+          path: |
+            .testmondata
+            .testmondata-shm
+            .testmondata-wal
+          key: ${{ runner.os }}-testmon-postgres-unit-py${{ matrix.python-version }}-g${{ matrix.group }}-${{ github.ref_name }}-${{ github.run_id }}
+          restore-keys: |
+            ${{ runner.os }}-testmon-postgres-unit-py${{ matrix.python-version }}-g${{ matrix.group }}-${{ github.ref_name }}-
+            ${{ runner.os }}-testmon-postgres-unit-py${{ matrix.python-version }}-g${{ matrix.group }}-main-
+            ${{ runner.os }}-testmon-postgres-unit-py${{ matrix.python-version }}-main-
+            ${{ runner.os }}-testmon-postgres-unit-py${{ matrix.python-version }}-
 
       - name: Create virtual env
         run: |
@@ -185,30 +284,44 @@ jobs:
 
       - name: Run tests
         run: |
-          just test-unit-postgres
+          BASIC_MEMORY_PYTEST_SPLIT_FLAGS="--splits 3 --group ${{ matrix.group }}" just test-unit-postgres
 
   test-postgres-integration:
+    needs: changes
+    if: needs.changes.outputs.code == 'true'
     name: Test Postgres Integration (Python ${{ matrix.python-version }})
     timeout-minutes: 45
-    needs: [static-checks]
     strategy:
       fail-fast: false
       matrix:
-        include:
-          - python-version: "3.12"
-          - python-version: "3.13"
-          - python-version: "3.14"
+        # Latest Python only: SQLite carries version coverage, Postgres carries
+        # backend coverage.
+        python-version: ["3.14"]
     runs-on: ubuntu-latest
-
-    # Note: No services section needed - testcontainers handles Postgres in Docker
+    services:
+      postgres:
+        image: pgvector/pgvector:pg16
+        env:
+          POSTGRES_USER: basic_memory_user
+          POSTGRES_PASSWORD: dev_password
+          POSTGRES_DB: basic_memory_test
+        ports:
+          - 5432:5432
+        options: >-
+          --health-cmd "pg_isready -U basic_memory_user -d basic_memory_test"
+          --health-interval 10s
+          --health-timeout 5s
+          --health-retries 5
+    env:
+      BASIC_MEMORY_TEST_POSTGRES_URL: postgresql://basic_memory_user:dev_password@127.0.0.1:5432/basic_memory_test
 
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
         with:
           submodules: true
 
       - name: Set up Python ${{ matrix.python-version }}
-        uses: actions/setup-python@v4
+        uses: actions/setup-python@v6
         with:
           python-version: ${{ matrix.python-version }}
           cache: 'pip'
@@ -217,7 +330,20 @@ jobs:
         run: |
           pip install uv
 
-      - uses: extractions/setup-just@v3
+      - uses: extractions/setup-just@v4
+
+      - name: Cache pytest-testmon results
+        uses: actions/cache@v4
+        with:
+          path: |
+            .testmondata
+            .testmondata-shm
+            .testmondata-wal
+          key: ${{ runner.os }}-testmon-postgres-integration-py${{ matrix.python-version }}-${{ github.ref_name }}-${{ github.run_id }}
+          restore-keys: |
+            ${{ runner.os }}-testmon-postgres-integration-py${{ matrix.python-version }}-${{ github.ref_name }}-
+            ${{ runner.os }}-testmon-postgres-integration-py${{ matrix.python-version }}-main-
+            ${{ runner.os }}-testmon-postgres-integration-py${{ matrix.python-version }}-
 
       - name: Create virtual env
         run: |
@@ -232,18 +358,19 @@ jobs:
           just test-int-postgres
 
   test-semantic:
+    needs: changes
+    if: needs.changes.outputs.code == 'true'
     name: Test Semantic (Python 3.12)
     timeout-minutes: 45
-    needs: [static-checks]
     runs-on: ubuntu-latest
 
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@v6
         with:
           submodules: true
 
       - name: Set up Python 3.12
-        uses: actions/setup-python@v4
+        uses: actions/setup-python@v6
         with:
           python-version: "3.12"
           cache: "pip"
@@ -252,7 +379,20 @@ jobs:
         run: |
           pip install uv
 
-      - uses: extractions/setup-just@v3
+      - uses: extractions/setup-just@v4
+
+      - name: Cache pytest-testmon results
+        uses: actions/cache@v4
+        with:
+          path: |
+            .testmondata
+            .testmondata-shm
+            .testmondata-wal
+          key: ${{ runner.os }}-testmon-semantic-py3.12-${{ github.ref_name }}-${{ github.run_id }}
+          restore-keys: |
+            ${{ runner.os }}-testmon-semantic-py3.12-${{ github.ref_name }}-
+            ${{ runner.os }}-testmon-semantic-py3.12-main-
+            ${{ runner.os }}-testmon-semantic-py3.12-
 
       - name: Create virtual env
         run: |

.gitignore

@@ -49,12 +49,19 @@ ENV/
 /docs/.obsidian/
 /examples/.obsidian/
 /examples/.basic-memory/
-
+/docs/assets
 
 # claude action
 claude-output
 **/.claude/settings.local.json
 .mcp.json
+!/plugins/codex/.mcp.json
 .mcpregistry_*
 /.testmondata
 .benchmarks/
+
+# Consolidated package build artifacts
+/integrations/openclaw/node_modules/
+/integrations/openclaw/dist/
+/integrations/openclaw/skills/
+/integrations/openclaw/*.tgz

AGENTS.md

@@ -22,16 +22,22 @@ See the [README.md](README.md) file for a project overview.
 - Run unit tests (Postgres): `just test-unit-postgres`
 - Run integration tests (SQLite): `just test-int-sqlite`
 - Run integration tests (Postgres): `just test-int-postgres`
-- Run impacted tests: `just testmon` (pytest-testmon)
+- Run impacted tests: `just testmon` (pytest-testmon; only tests affected by changed code)
 - Run MCP smoke test: `just test-smoke`
-- Fast local loop: `just fast-check`
+- Fast local loop: `just fast-check` (default iteration flow)
 - Local consistency check: `just doctor`
+- Run all consolidated agent package checks: `just package-check`
+- Run Claude Code plugin checks: `just package-check-claude-code`
+- Run shared skills checks: `just package-check-skills`
+- Run Hermes plugin checks: `just package-check-hermes`
+- Run OpenClaw plugin checks: `just package-check-openclaw`
+- Run host-native agent harness checks: `just agent-harness-check`
 - Generate HTML coverage: `just coverage`
 - Single test: `pytest tests/path/to/test_file.py::test_function_name`
 - Run benchmarks: `pytest test-int/test_sync_performance_benchmark.py -v -m "benchmark and not slow"`
 - Lint: `just lint` or `ruff check . --fix`
-- Type check: `just typecheck` or `uv run pyright`
-- Type check (supplemental): `just typecheck-ty` or `uv run ty check src/`
+- Type check: `just typecheck` or `uv run ty check src tests test-int`
+- Type check (pyright): `just typecheck-pyright` or `uv run pyright`
 - Format: `just format` or `uv run ruff format .`
 - Run all code checks: `just check` (runs lint, format, typecheck, test)
 - Create db migration: `just migration "Your migration message"`
@@ -48,12 +54,37 @@ See the [README.md](README.md) file for a project overview.
 ### Code/Test/Verify Loop (fast path)
 
 1) **Code:** make changes.
-2) **Test:** `just fast-check` (lint/format/typecheck + impacted tests + MCP smoke).
+2) **Test:** `just fast-check` (lint/format/typecheck + pytest-testmon impacted tests for changed code).
 3) **Verify:** `just doctor` (end-to-end file ↔ DB loop in a temp project).
-4) **Full gate (when needed):** `just test` or `just check` for SQLite + Postgres.
+4) **Package verify:** `just package-check` when changes touch `plugins/`, `skills/`, `integrations/`, package metadata, or release wiring.
+5) **Full gate (when needed):** `just test` or `just check` for SQLite + Postgres.
+
+Run `just test-smoke` when you specifically need the MCP smoke flow.
 
 If testmon is “cold,” the first run may be long. Subsequent runs get much faster.
 
+### Consolidated Agent Package Checks
+
+The monorepo ships several host-native packages alongside the Python core. Use the root justfile as the canonical entry point:
+
+- `just package-check` — validates every copied package and generated bundle path.
+- `just package-check-claude-code` — validates the root and plugin-local Claude marketplace manifests, the SessionStart/PreCompact hooks, the bundled output style, and the seed schemas, then runs `claude plugin validate . --strict`.
+- `just package-check-skills` — validates every top-level `skills/memory-*/SKILL.md` frontmatter block.
+- `just package-check-hermes` — validates `integrations/hermes/plugin.yaml`, the Hermes provider entrypoint, bundled skill, and runs the hermetic unit suite.
+- `just package-check-openclaw` — runs the OpenClaw package install, copies top-level skills into the generated bundle, typechecks, lints, builds `dist/`, runs Bun tests, and performs `npm pack --dry-run`.
+- `just agent-harness-check` — checks the host-specific harnesses without the shared markdown-only skills target.
+
+Package-local justfiles live in `plugins/claude-code/`, `skills/`, `integrations/hermes/`, and `integrations/openclaw/`. Prefer the root targets for PR verification so command names stay stable as package internals evolve.
+
+### PR CI Gate
+
+Before opening or updating a PR, run the checks that mirror the common required CI failures:
+
+- Run `just typecheck` in addition to targeted `ruff` and `pytest` commands when tests were added or changed.
+- Sign commits with `git commit -s` so DCO passes. If a PR branch already has unsigned commits, rewrite the branch with signed-off commits before asking for review.
+- Use a semantic PR title accepted by `.github/workflows/pr-title.yml`: `type(scope): summary`.
+- Use one of the allowed scopes: `core`, `cli`, `api`, `mcp`, `sync`, `ui`, `ci`, `deps`, `installer`, `plugins`, `skills`, `integrations`.
+
 ### Test Structure
 
 - `tests/` - Unit tests for individual components (mocked, fast)
@@ -77,13 +108,34 @@ If testmon is “cold,” the first run may be long. Subsequent runs get much fa
 - Follow the repository pattern for data access
 - Tools communicate to api routers via the httpx ASGI client (in process)
 
+### Programming Style
+
+See [docs/ENGINEERING_STYLE.md](docs/ENGINEERING_STYLE.md) for the fuller house style. The
+short version for agents:
+
+- Prefer type-safe, explicit designs over object-heavy indirection. Use Python 3.12 `type`
+  aliases, full annotations, and narrow `Protocol`s when a caller only needs a capability.
+- Use dataclasses for internal value objects and operation results; use Pydantic v2 at API,
+  CLI, MCP, and persistence boundaries where validation and serialization matter.
+- Keep async boundaries obvious. Resource-owning code should use context managers, propagate
+  cancellation, and avoid hidden background work unless the lifecycle is explicit.
+- Fail fast. Do not add silent fallback logic, broad exception swallowing, speculative
+  `getattr`, or casts that hide an unclear model shape.
+- Keep control flow simple and local. Push branching decisions up, keep leaf helpers focused,
+  and name values after the domain concept they carry.
+- Use evidence-first testing. Add or update meaningful regression tests for bugs and risky
+  behavior, prefer real code paths over mocks, and run the narrowest command that proves the
+  change before widening verification.
+- Comments should explain why a branch, invariant, or constraint exists. Avoid comments that
+  merely narrate obvious code.
+
 ### Code Change Guidelines
 
 - **Full file read before edits**: Before editing any file, read it in full first to ensure complete context; partial reads lead to corrupted edits
 - **Minimize diffs**: Prefer the smallest change that satisfies the request. Avoid unrelated refactors or style rewrites unless necessary for correctness
-- **No speculative getattr**: Never use `getattr(obj, "attr", default)` when unsure about attribute names. Check the class definition or source code first
-- **Fail fast**: Write code with fail-fast logic by default. Do not swallow exceptions with errors or warnings
-- **No fallback logic**: Do not add fallback logic unless explicitly told to and agreed with the user
+- **House style is canonical**: Follow the Programming Style section above for type-safe,
+  fail-fast code; do not hide unclear models with speculative attributes, broad exception
+  handling, casts, or unapproved fallback logic
 - **No guessing**: Do not say "The issue is..." before you actually know what the issue is. Investigate first.
 
 ### Literate Programming Style
@@ -151,6 +203,10 @@ See [docs/ARCHITECTURE.md](docs/ARCHITECTURE.md) for detailed architecture docum
 - `/schemas` - Pydantic models for validation
 - `/services` - Business logic layer
 - `/sync` - File synchronization services + `coordinator.py` for lifecycle management
+- `/plugins/claude-code` - Claude Code plugin marketplace package, hooks, skills, and agent harness
+- `/skills` - Canonical framework-agnostic Basic Memory `SKILL.md` source
+- `/integrations/hermes` - Hermes memory-provider plugin
+- `/integrations/openclaw` - OpenClaw npm/TypeScript plugin
 
 **Composition Roots:**
 Each entrypoint (API, MCP, CLI) has a composition root that:
@@ -242,6 +298,33 @@ async_client.set_client_factory(your_custom_factory)
 
 See SPEC-16 for full context manager refactor details.
 
+### Release Process
+
+Releases are driven by `just release` / `just beta` — never by a bare `git tag`. The recipes bump version metadata, run pre-flight checks, commit, tag, and push. GitHub Actions then publishes to PyPI and updates the Homebrew formula.
+
+**Stable release:**
+
+```
+just release v0.21.3
+```
+
+The recipe runs `just lint` + `just typecheck`, then updates every release manifest through `scripts/update_versions.py`: `src/basic_memory/__init__.py`, `server.json`, the root Claude marketplace, the Claude Code plugin manifest and local marketplace, the Hermes `plugin.yaml`, and the OpenClaw `package.json`. It commits as `chore: update version to X.Y.Z for vX.Y.Z release`, creates the `vX.Y.Z` tag, and pushes both the commit and the tag to `origin/main`. After the tag lands, the `Release` workflow builds the Python package, publishes to PyPI, creates the GitHub release with auto-generated notes, publishes the OpenClaw npm package, and updates the Homebrew formula. The recipe finishes by printing the post-release tasks the workflow doesn't cover.
+
+**Beta release:** `just beta v0.21.3b1` — same flow with a beta-suffixed tag. PyPI consumers install with `pip install basic-memory --pre`.
+
+**Release dry run:** `just release-dry-run v0.21.4` previews the consolidated version update without writing files.
+
+**Development builds:** every commit to `main` publishes a `0.21.3.dev26+468a22f`-style version to PyPI automatically via `.github/workflows/dev-release.yml`. No human action.
+
+**Do not tag releases by hand.** A bare `git tag vX.Y.Z` skips the in-code version bump. Package metadata is still correct (uv-dynamic-versioning derives it from the git tag) but `basic-memory --version` reports the previous release, which is what happened with v0.21.2 → v0.21.3.
+
+**Post-release tasks** the recipe surfaces but doesn't run:
+- `docs.basicmemory.com` — add notes to `src/pages/latest-releases.mdx`
+- `basicmachines.co` — bump version in `src/components/sections/hero.tsx`
+- MCP Registry — `mcp-publisher publish` from the repo root
+
+See `.claude/commands/release/release.md` (and `beta.md`, `release-check.md`, `changelog.md` alongside it) for the full release + post-release runbook, including the slash commands.
+
 ## BASIC MEMORY PRODUCT USAGE
 
 ### Knowledge Structure
@@ -286,6 +369,13 @@ See SPEC-16 for full context manager refactor details.
 - Manage snapshots: `basic-memory cloud snapshot [create|list|delete|show|browse]`
 - Restore from snapshot: `basic-memory cloud restore <path> --snapshot <id>`
 
+**Cloud Sync Commands (Personal and Team workspaces):**
+- Fetch cloud changes (cloud -> local): `basic-memory cloud pull --name "name"` (Team-safe; additive, never deletes local)
+- Upload local changes (local -> cloud): `basic-memory cloud push --name "name"` (Team-safe; additive, never deletes cloud)
+- Resolve conflicts on push/pull: `--on-conflict [fail|keep-local|keep-cloud|keep-both]` (default `fail` lists conflicts and aborts, git-style)
+- One-way mirror (local -> cloud): `basic-memory cloud sync --name "name"` (Personal workspaces only; deletes cloud files missing locally)
+- Two-way mirror (local <-> cloud): `basic-memory cloud bisync --name "name"` (Personal workspaces only)
+
 ### MCP Capabilities
 
 - Basic Memory exposes these MCP tools to LLMs:
@@ -442,5 +532,9 @@ With GitHub integration, the development workflow includes:
 3. **Branch management** - Claude can create feature branches for implementations
 4. **Documentation maintenance** - Claude can keep documentation updated as the code evolves
 5. **Code Commits**: ALWAYS sign off commits with `git commit -s`
+6. **Pull Request Titles**: PR titles must follow the semantic format enforced by `.github/workflows/pr-title.yml`: `type(scope): summary`
+   - Allowed types: `feat`, `fix`, `chore`, `docs`, `style`, `refactor`, `perf`, `test`, `build`, `ci`
+   - Allowed scopes: `core`, `cli`, `api`, `mcp`, `sync`, `ui`, `ci`, `deps`, `installer`, `plugins`, `skills`, `integrations`
+   - Example: `fix(cli): propagate cloud workspace routing`
 
 This level of integration represents a new paradigm in AI-human collaboration, where the AI assistant becomes a full-fledged team member rather than just a tool for generating code snippets.

CHANGELOG.md

@@ -1,14 +1,492 @@
 # CHANGELOG
 
-## Unreleased
+## v0.22.0 (2026-06-11)
+
+Team-safe cloud sync. New additive `bm cloud push` and `bm cloud pull`
+commands work safely on shared Team workspaces, while the destructive mirror
+commands are gated to Personal workspaces. Also: a large batch of MCP tool
+fixes, search improvements, and embedding reliability work.
 
 ### Features
 
+- **#917**: Added Team-safe `bm cloud push` / `bm cloud pull`. Both are
+  additive (they never delete on the destination) and abort on conflicts by
+  default, git-style, with `--on-conflict {fail|keep-local|keep-cloud|keep-both}`.
+  The destructive `bm cloud sync` / `bm cloud bisync` mirrors are now gated
+  to Personal workspaces.
+- **#920**: Team push/pull uses per-workspace rclone remotes, so remotes and
+  credentials stay scoped to each workspace.
+- **#908**: Search supports an observation category filter.
+- **#809**: Added an experimental LiteLLM embedding provider for semantic
+  search (marked experimental, see **#899**).
+- **#907**: `bm tool write-note` accepts `--type`.
+- **#906**: `bm status` accepts `--wait` and `--timeout`.
+- Added the `bm tool delete-note` command.
+- **#905**: Improved workspace and cloud bisync command discoverability.
+
+### Bug Fixes
+
+- **#931 / #946**: Truncated observation permalinks are disambiguated to
+  prevent search-index collisions, and `build_context` resolves observations
+  by the same permalink the search index uses (**#909**, **#929**).
+- **#934**: `edit_note` recovers when the file exists on disk but is not yet
+  indexed (**#581**).
+- **#911 / #932 / #941**: Comma-separated tags are split consistently in
+  `parse_tags` and the `search_notes` tags parameter, with input normalized
+  for direct callers (**#910**).
+- **#933**: `read_note` accepts `page`/`page_size` for parity with sibling
+  tools (**#883**).
+- **#914 / #904 / #916**: `move_note` resolves `memory://` URLs, stops
+  falsely rejecting same-project moves as cross-project, no longer reports
+  false success across project boundaries, and mismatch guidance points at
+  the landing path.
+- **#915**: Navigation pagination is validated, and `recent_activity` shows
+  the correct project.
+- **#913**: `bm tool` commands align with MCP behavior (error exit codes,
+  overwrite handling, category support, defaults).
+- **#923**: `bm cloud setup` no longer overwrites an existing rclone remote
+  (**#922**).
+- **#912**: The `note_types` search filter is case-insensitive.
+- `build_context` allows cross-project context traversal, `write_note`
+  resolves overwrite conflicts, and sync uses strict deferred relation
+  resolution.
+- Embedding reliability: FastEmbed vectors are L2-normalized (**#843**),
+  corrupt FastEmbed model caches self-heal (**#900**), a single embedding
+  provider is reused per process (**#903**), `sqlite-vec` loads for the
+  embedding-status query (**#901**), and engine disposal no longer crashes
+  on the Postgres backend (**#902**).
+
+### Maintenance
+
+- Leaner, faster CI: testmon-selected branch builds, sharded Postgres jobs,
+  and faster default test fixtures (**#928**, **#938**, **#945**).
+- Documented personal-vs-team cloud sync semantics (**#947**, closes
+  **#851**).
+- Fixed npx skill install docs (**#927**).
+- Added `glama.json` to claim the Glama MCP directory listing (**#953**).
+
+## v0.21.6 (2026-06-04)
+
+Monorepo consolidation plus a redesigned Claude Code plugin. The satellite
+repositories now live in the main `basic-memory` tree, and the plugin is
+rebuilt as a memory bridge with a guided setup interview, capture skills, and
+team workspaces. Codex, Hermes, and OpenClaw integration packages ship
+alongside it.
+
+### Core
+
+- Consolidated the Basic Memory satellite repositories into the main
+  `basic-memory` tree as the canonical source, discovery, documentation,
+  issue, and release home.
+- Added root marketplace/package validation for the consolidated repository
+  layout.
+- Added root and package-local justfile targets so Claude Code, Codex, skills,
+  Hermes, and OpenClaw builds can be verified from the monorepo.
+- Removed the legacy `ui/` directory.
+
+### API
+
+- Entity resolution now exposes the owning project on resolved entities so
+  callers can route follow-up reads to the correct workspace/project.
+
+### CLI
+
+- Added an "auto BM" GitHub CI workflow that captures notes from CI runs.
+- Exposed project sync-support metadata and surfaced copyable workspace
+  identifiers in project listings.
+- `cloud login` now surfaces non-subscription errors instead of masking them.
+- Team workspaces block `rclone sync`, with the team-workspace guard limited
+  to `bisync`.
+
+### Claude Code
+
+- Rebuilt the Claude Code plugin as a memory bridge with SessionStart and
+  PreCompact hooks, a bundled output style, and seeded note schemas.
+- Added the `/basic-memory:setup` bootstrap interview and the
+  `/basic-memory:remember`, `/basic-memory:status`, and `/basic-memory:share`
+  skills.
+- Added team workspace support with attribution for shared writes.
+- Prefixed plugin skills with `bm-` and installed the shared `memory-*` skills
+  instead of duplicating them in the plugin.
+- Hooks fall back to `uvx`/`uv` when no CLI binary is on PATH.
+
+### Codex
+
+- Added the Codex plugin under `plugins/codex/` with its native
+  `.codex-plugin`, hooks, seeded schemas, and `bm-*` skills.
+
+### Skills
+
+- Added the shared Basic Memory `SKILL.md` collection under top-level
+  `skills/` and ported useful retired plugin skills into the `memory-*` set.
+- Fixed invalid picoschema enum YAML in the memory skills.
+
+### Hermes
+
+- Added the Hermes memory provider plugin under `integrations/hermes/` with
+  its native Python module, `plugin.yaml`, skill, tests, docs, and release
+  metadata.
+
+### OpenClaw
+
+- Added the OpenClaw plugin under `integrations/openclaw/` with its native
+  npm/TypeScript package shape, tests, docs, and release metadata; skill
+  bundling copies from the top-level `skills/` source.
+
+## v0.21.5 (2026-05-26)
+
+Workspace/project routing fixes for MCP, plus a SQLite vector reindex stability fix.
+
+### Bug Fixes
+
+- **#854**: MCP project listing now keeps duplicate cloud project rows distinct by
+  workspace, so only the selected workspace row inherits local project state.
+- **#853**: `write_note` returns workspace-qualified permalinks for cloud
+  workspace writes, allowing follow-up `memory://` reads to route back to the
+  correct workspace/project.
+- **#852**: Full SQLite vector reindex now loads `sqlite-vec` before dropping
+  `vec0` virtual tables, preventing reindex crashes.
+- **#838**: Local ASGI database initialization is preloaded so MCP routing can
+  safely enter local project contexts.
+
+## v0.21.1 (2026-05-16)
+
+CI-only release. No user-facing changes.
+
+### Maintenance
+
+- **#833**: Replace `mislav/bump-homebrew-formula-action` with an inline
+  bash bump step. The action's `resolveRedirect` HEAD-requests
+  `api.github.com /repos/.../tarball/<ref>` expecting HTTP 302; GitHub now
+  returns 303 for authenticated requests on that endpoint, which broke
+  the v0.21.0 Homebrew job and required a manual tap bump. The inline
+  step does the same work (curl + sha256sum, clone tap, sed-bump
+  `url`/`sha256`, commit, push) without any third-party dependency.
+
+## v0.21.0 (2026-05-16)
+
+Workspace-aware everywhere: every MCP tool and CLI command now routes through the
+same workspace/project model, the search and sync paths are noticeably faster,
+and a handful of long-standing parsing and routing footguns are gone.
+
+### Breaking Changes
+
+- Relation parsing no longer treats unquoted multi-word text before a wikilink as a
+  custom relation type. Use single-token relation types like `relates_to [[Target]]`,
+  or quote multi-word relation types like `"relates to" [[Target]]` or
+  `'relates to' [[Target]]`.
+  - Bare list wikilinks like `- [[Target]]` now index as `links_to`.
+  - Prose list items like `- some other thing [[Target]]` now index as `links_to`.
+  - To preserve existing multi-word relation types on re-sync, quote them before upgrading.
+  - See **#824**.
+
+### Features
+
+- **#816**: `bm orphan` CLI command surfaces entities whose underlying markdown
+  files are gone, with a flag to clean them out.
+- **#789**: Create projects directly by cloud workspace slug from MCP
+  (`create_memory_project(workspace=...)`).
+- **#757**: Discover projects across every accessible cloud workspace in MCP's
+  project list — no more per-workspace blind spots.
+- **#766**: MCP tools accept training-data-friendly parameter aliases
+  (`q`/`search`/`text` for `query`, etc.) so models reach for them naturally.
+- **#776**: `bm db reset` refuses to run while a `basic-memory mcp` process is
+  alive, so resets can no longer corrupt an open session.
+- **#791**: Search responses include result totals so pagers can stop guessing.
+- **#719**: Cloud `note_content` tenant schema primitive lands on the backend.
+- **#715**: `bm project add` accepts a `--visibility` flag for cloud projects.
+
+### Bug Fixes
+
+#### Search and recent activity
+- **#832**: SQLite project deletion now sweeps `search_index`,
+  `search_vector_chunks`, and `search_vector_embeddings`, so a project that
+  reuses a recycled auto-increment id can't inherit the previous tenant's content.
+- **#812**: `recent_activity` orders and filters by `updated_at`, so edits bubble
+  to the top instead of staying pinned to creation time.
+- **#807**: Multi-project `search_notes` is opt-in (`search_all_projects=True`);
+  default search stays scoped to the resolved project.
+- **#785**: `recent_activity` caps responses and emits an explicit truncation
+  footer instead of silently dropping rows.
+- **#713**: Eliminated an N+1 query in search result hydration.
+
+#### Workspace / project routing
+- **#822**: `bm project list` now includes projects from every workspace, not
+  just the current one.
+- **#813**, **#808**, **#806**, **#803**, **#801**, **#795**, **#790**, **#778**,
+  **#777**, **#722**, **#712**, **#704**: Workspace-qualified permalink routing
+  is centralized and applied consistently across `edit_note`, `delete_project`,
+  `build_context`, `memory://` URLs, factory-mode project listing, cloud uploads,
+  and the API client.
+
+#### Sync
+- **#827**: `rclone bisync` filters from `.bmignore` are preserved across syncs.
+- **#815**: Watch service ignores hidden paths relative to the watched project,
+  not just relative to `$HOME`.
+- **#814**: `scan` subprocesses no longer go through the shell, avoiding quoting
+  issues with paths that contain special characters.
+- **#759**: Watch service stays inside `--project` scope.
+- **#746**: Canonical markdown is preserved during single-file sync.
+
+#### Parsing
+- **#796**: Picoschema modifier descriptions (`field?(modifier, description)`)
+  parse correctly.
+- **#769**: Obsidian callout blocks are skipped by the observation parser
+  instead of being mis-extracted as observations.
+
+#### CLI
+- **#775**: `bm project set-cloud` / `set-local` cleans up local DB state for
+  the affected project.
+- **#773**: `bm cloud logout` clears `default_workspace`.
+- **#780**: `bm cloud setup` hint points at `bm cloud sync-setup`.
+- **#734**: `bm project info` shows cloud index freshness.
+- **#718**: Private cloud projects display under their `display_name` instead of
+  raw UUID.
+- **#768**: `read_note` / `view_note` drop no-op pagination params from their
+  signatures.
+
+#### Stability
+- **#774**: `sqlite-vec` failures during init degrade gracefully to keyword-only
+  search instead of crashing startup.
+- **#733**: Delete-vector and cloud-sync cleanup is now consistent.
+- **#702**: Race conditions in concurrent `delete_entity` are resolved.
+- **#724**: `external_id` is preserved when entities are re-upserted during a
+  re-index.
+- **#728**: Vector init no longer issues runtime `ALTER TABLE`.
+- **#744**: `BASIC_MEMORY_CONFIG_DIR` is honored across remaining call sites.
+- **#743**: FastEmbed cache lives under the data dir instead of `/tmp`.
+- **#752**: Cloud projects report `source=cloud` in factory mode.
+- Stripped null bytes from markdown content before DB insert.
+- Allowed long `relation_type` values in API responses.
+
+#### Installer
+- **#772**: Docker-compose config volume mounts under the `appuser` home.
+- **#695**: Bumped `brew outdated` timeout from 15s to 60s.
+
+### Performance
+
+- **#828**: CLI startup no longer pulls in the local ASGI FastAPI app when it
+  isn't needed.
+- **#751**, **#726**, **#717**, **#714**, single-file/batch indexing: marked
+  speedups on the sync hot path; unchanged markdown is skipped entirely.
+- **#731**, **#723**: Vector sync is faster on both backends, with tuned
+  fastembed defaults.
+
+### Maintenance
+
+- **#825**: Dependency refresh + security hardening.
+- Updated to `fastmcp` 3.3.1.
+- **#754**: Removed in-house telemetry wrappers in favor of direct `logfire`
+  usage.
+- **#736**: `ty` is now the default typechecker.
+- **#771**, **#770**, **#716**: New regression guards for vector-row cleanup,
+  long relation types, and recent activity hydration.
+
+## v0.20.3 (2026-03-26)
+
+### Bug Fixes
+
+- **#698**: CLI cloud commands now use API key when configured
+  - `get_authenticated_headers()` only checked OAuth tokens, ignoring `config.cloud_api_key`
+  - All CLI cloud commands (`upload`, `status`, `snapshot`, `restore`, etc.) failed for API-key-only users while MCP tools worked fine
+  - Now mirrors the same credential priority as MCP: API key first, OAuth fallback
+  - Fixes `bm cloud upload --project` returning "project does not exist" when authenticated with `bmc_*` API key
+
+## v0.20.2 (2026-03-10)
+
+### Bug Fixes
+
+- Fix auto-update Homebrew detection: `brew outdated` exits 1 when a formula is outdated, not on error
+  - Previously treated exit code 1 as a failure, causing "Automatic update check failed" instead of detecting the available update
+
+## v0.20.1 (2026-03-10)
+
+### Bug Fixes
+
+- **#661**: Fix `bm project list` MCP column to show transport type (stdio/https) instead of DB presence
+  - Renamed "MCP (stdio)" column to "MCP"
+  - Shows actual routing mode: `stdio` for local, `https` for cloud projects
+  - Clears local path display for cloud-mode projects
+- **#662**: Invalidate config cache when file is modified by another process
+  - Adds mtime-based cache validation to `ConfigManager.load_config()`
+  - Long-lived processes (MCP stdio server) now detect external config changes
+  - Fixes `bm project set-cloud` having no effect on running MCP server
+
+## v0.20.0 (2026-03-10)
+
+### Features
+
+- **#643**: Default-on auto-update system and `bm update` command
+  - Automatic background update checks for CLI installs (uv tool, Homebrew)
+  - Install-source detection (homebrew, uv_tool, uvx, unknown) with uvx skip behavior
+  - Periodic check gating via `auto_update_last_checked_at` + `update_check_interval` config
+  - Manager-specific update flows: Homebrew (`brew upgrade`) and uv tool (`uv tool upgrade`)
+  - Silent, non-blocking MCP behavior via daemon thread before server run
+  - Manual commands: `bm update` (force check + apply) and `bm update --check` (check only)
+  - New config fields: `auto_update`, `update_check_interval`, `auto_update_last_checked_at`
+
+## v0.19.2 (2026-03-09)
+
+### Bug Fixes
+
+- **#657**: Coerce string params to list/dict in MCP tools
+  - MCP clients that serialize `list`/`dict` arguments as JSON strings no longer fail Pydantic validation
+  - Adds `BeforeValidator` coercion to `search_notes` (`entity_types`, `note_types`, `tags`, `metadata_filters`), `write_note` (`metadata`), and `canvas` (`nodes`, `edges`)
+- **#655**: Handle SQLite and Windows semantic search regressions
+  - Fix embedding status query for non-semantic SQLite databases
+  - Windows-safe log file rotation with per-process log filenames
+  - Robust `setup_logging` that handles all environments cleanly
+
+## v0.19.1 (2026-03-08)
+
+### Bug Fixes
+
+- **#649**: Enforce strict entity resolution in destructive MCP tools (`edit_note`, `move_note`, `delete_note`)
+  - Prevents fuzzy-match fallback from silently editing/moving/deleting the wrong note
+  - DST-related timeframe validation fix (round instead of truncate days)
+
+### Features
+
+- **#648**: Add `insert_before_section` and `insert_after_section` edit operations
+- Add `GET /knowledge/graph` endpoint for full graph visualization
+
+### Dependencies
+
+- Bump authlib from 1.6.6 to 1.6.7
+
+## v0.19.0 (2026-03-07)
+
+### Highlights
+
+- **Semantic vector search** for SQLite and Postgres with FastEmbed embeddings
+- **Schema system** for validating and inferring knowledge base structure
+- **Per-project cloud routing** with API key authentication
+- **Upgraded to FastMCP 3.0** with tool annotations
+- **CLI overhaul** with JSON output, workspace awareness, and project dashboard
+
+### Features
+
+- **#550**: Add semantic vector search for SQLite and Postgres
+  - FastEmbed-based embeddings with automatic backfill
+  - Hybrid search combining full-text and vector similarity
+  - Score-based fusion replacing RRF for better ranking
+  - `min_similarity` override for tuning search precision
+  - Semantic dependencies are now default, with optional extras fallback
+
+- **#549**: Schema system for Basic Memory
+  - `schema_infer` — infer schema from existing notes
+  - `schema_validate` — validate notes against a schema definition
+  - `schema_diff` — compare schemas across projects
+  - Frontmatter validation support (#597)
+  - Read schema definitions from file instead of stale DB metadata (#635)
+
+- **#555**: Per-project local/cloud routing with API key auth
+  - Individual projects route through cloud while others stay local
+  - `basic-memory cloud set-key` and `basic-memory project set-cloud/set-local`
+  - Stdio MCP honors per-project cloud routing (#590)
+
+- **#598**: Upgrade FastMCP 2.12.3 to 3.0.1 with tool annotations
+
+- **#585**: Add JSON output mode for MCP tools (default text)
+  - `--json` output for CLI commands for scripting and CI
+
+- **#576**: Add workspace selection flow for MCP and CLI
+  - Workspace-aware cloud project listing
+  - CLI refactoring for workspace support
+
+- **#544**: Project-prefixed permalinks and memory URL routing
+
+- **#632**: Add overwrite guard to `write_note` tool
+
+- **#614**: `edit_note` append/prepend auto-creates note if not found
+
+- **#609**: Richer content context in search results
+  - Return matched chunk text in search results (#601)
+  - Improved content hit rate
+
+- **#602**: Add `created_by` and `last_updated_by` user tracking to Entity
+
+- **#600**: Rename `entity_type` to `note_type` across codebase
+
+- **#574**: Add `display_name` and `is_private` to ProjectItem
+
+- **#569**: Expose `external_id` in EntityResponse and link resolver
+
+- **#567**: Isolate default SQLite DB by config dir
+
+- **#560**: Enable `default_project_mode` by default
+
+- **#559**: Add `basic-memory watch` CLI command
+
+- **#546**: Add cloud discovery touchpoints to CLI and MCP
+
+- **#572**: CLI analytics via Umami event collector
+
+- Replace project info with htop-inspired dashboard
+
+- Merge `search_by_metadata` into `search_notes` with optional query
+
 - Add `--strip-frontmatter` to `basic-memory tool read-note`
-  - Default behavior is unchanged: `content` still includes raw markdown with frontmatter.
-  - With `--strip-frontmatter`, both text and JSON modes return body-only markdown content.
-  - JSON output now includes an additive `frontmatter` field with parsed YAML metadata (or `null`
-    when no valid opening frontmatter block exists).
+
+- Add `destination_folder` parameter to `move_note` tool
+
+### Bug Fixes
+
+- **#644**: Fix default project resolution in cloud mode
+  - ChatGPT search/fetch tools broken in cloud mode
+  - `resolve_project_parameter` falls back to projects API
+
+- **#638**: Restore API backward compatibility for v0.18.x clients
+
+- **#637**: Create backup before config migration overwrites old format
+
+- **#636**: `list_workspaces` bypasses factory pattern on cloud MCP server
+
+- **#631**: `build_context` related_results schema validation failure
+
+- **#613**: Reduce excessive log volume by demoting per-request noise to DEBUG
+
+- **#612**: Handle quoted picoschema enum strings in YAML frontmatter
+
+- **#607**: Guard against closed streams in promo and missing vector tables
+
+- **#606**: Accept null for `expected_replacements` in `edit_note`
+
+- **#595**: `recent_activity` dedup and pagination across MCP tools
+
+- **#593**: Backend-specific distance-to-similarity conversion
+
+- **#582**: Use LinkResolver fallback in `build_context` for flexible identifier matching
+
+- **#577**: Replace RRF with score-based fusion in hybrid search
+
+- **#575**: Remove hardcoded "main" default from `default_project`
+
+- **#534**: Speed up `bm --version` startup
+
+- Fix semantic embeddings not generated on fresh DB or upgrade
+
+- Clarify `search_notes` parameter naming and fix `note_types` case sensitivity
+
+- Parse `tag:` prefix at MCP tool level to avoid hybrid search failure
+
+- Cap sqlite-vec knn k parameter at 4096 limit
+
+- Parameterize SQL queries in search repository type filters
+
+- Coerce list frontmatter values to strings for title and type fields
+
+- Avoid `Post(**metadata)` crash when frontmatter contains 'content' or 'handler' keys
+
+- Upgrade cryptography and python-multipart for security advisories
+
+### Internal
+
+- **#594**: Add `ty` as supplemental type checker
+- Batched vector sync orchestration across repositories
+- FastEmbed parallel guardrails and provider caching
+- Improved cloud CLI status and error messages
+- CI coverage and Postgres test fixes
 
 ## v0.18.5 (2026-02-13)
 
@@ -1927,12 +2405,12 @@ Signed-off-by: phernandez <paul@basicmachines.co>
 - Update CLAUDE.md ([#33](https://github.com/basicmachines-co/basic-memory/pull/33),
   [`dfaf0fe`](https://github.com/basicmachines-co/basic-memory/commit/dfaf0fea9cf5b97d169d51a6276ec70162c21a7e))
 
-fix spelling in CLAUDE.md: enviroment -> environment Signed-off-by: Ikko Eltociear Ashimine
+fix spelling in CLAUDE.md: environment typo Signed-off-by: Ikko Eltociear Ashimine
   <eltociear@gmail.com>
 
 ### Refactoring
 
-- Move project stats into projct subcommand
+- Move project stats into project subcommand
   ([`2a881b1`](https://github.com/basicmachines-co/basic-memory/commit/2a881b1425c73947f037fbe7ac5539c015b62526))
 
 Signed-off-by: phernandez <paul@basicmachines.co>
@@ -2361,7 +2839,7 @@ Co-authored-by: phernandez <phernandez@basicmachines.co>
 
 ### Bug Fixes
 
-- Refix vitual env in installer build
+- Refix virtual env in installer build
   ([`052f491`](https://github.com/basicmachines-co/basic-memory/commit/052f491fff629e8ead629c9259f8cb46c608d584))
 
 
@@ -2380,7 +2858,7 @@ Co-authored-by: phernandez <phernandez@basicmachines.co>
 
 ### Bug Fixes
 
-- Fix path to intaller app artifact
+- Fix path to installer app artifact
   ([`53d220d`](https://github.com/basicmachines-co/basic-memory/commit/53d220df585561f9edd0d49a9e88f1d4055059cf))
 
 
@@ -2388,7 +2866,7 @@ Co-authored-by: phernandez <phernandez@basicmachines.co>
 
 ### Bug Fixes
 
-- Activate vitualenv in installer build
+- Activate virtualenv in installer build
   ([`d4c8293`](https://github.com/basicmachines-co/basic-memory/commit/d4c8293687a52eaf3337fe02e2f7b80e4cc9a1bb))
 
 - Trigger installer build on release

CONTRIBUTING.md

@@ -224,40 +224,6 @@ See `test-int/BENCHMARKS.md` for detailed benchmark documentation.
 - **Fixtures**: Use async pytest fixtures for setup and teardown
 - **Markers**: Use `@pytest.mark.benchmark` for benchmarks, `@pytest.mark.slow` for slow tests
 
-## Release Process
-
-Basic Memory uses automatic versioning based on git tags with `uv-dynamic-versioning`. Here's how releases work:
-
-### Version Management
-- **Development versions**: Automatically generated from git commits (e.g., `0.12.4.dev26+468a22f`)
-- **Beta releases**: Created by tagging with beta suffixes (e.g., `git tag v0.13.0b1`)
-- **Stable releases**: Created by tagging with version numbers (e.g., `git tag v0.13.0`)
-
-### Release Workflows
-
-#### Development Builds
-- Automatically published to PyPI on every commit to `main`
-- Version format: `0.12.4.dev26+468a22f` (base version + dev + commit count + hash)
-- Users install with: `pip install basic-memory --pre --force-reinstall`
-
-#### Beta Releases
-1. Create and push a beta tag: `git tag v0.13.0b1 && git push origin v0.13.0b1`
-2. GitHub Actions automatically builds and publishes to PyPI
-3. Users install with: `pip install basic-memory --pre`
-
-#### Stable Releases
-1. Create and push a version tag: `git tag v0.13.0 && git push origin v0.13.0`
-2. GitHub Actions automatically:
-   - Builds the package with version `0.13.0`
-   - Creates GitHub release with auto-generated notes
-   - Publishes to PyPI
-3. Users install with: `pip install basic-memory`
-
-### For Contributors
-- No manual version bumping required
-- Versions are automatically derived from git tags
-- Focus on code changes, not version management
-
 ## Creating Issues
 
 If you're planning to work on something, please create an issue first to discuss the approach. Include:

SECURITY.md

@@ -8,6 +8,71 @@
 
 ## Reporting a Vulnerability
 
-Use this section to tell people how to report a vulnerability.
+If you find a vulnerability, please contact hello@basicmachines.co.
 
-If you find a vulnerability, please contact hello@basicmachines.co
+Please do not open a public GitHub issue for security vulnerabilities. We aim
+to respond within 72 hours and will coordinate a fix and disclosure timeline
+with you.
+
+## Threat Model
+
+Basic Memory is a local-first MCP server that reads and writes markdown files
+inside configured project directories. It runs on your machine with your user
+permissions, so local configuration deserves the same care as any other
+developer tool that can access your files.
+
+### What Basic Memory Controls
+
+- Filesystem-touching tools validate paths against the configured project root
+  with `validate_project_path()`, resolved paths, and `Path.is_relative_to()`.
+  Path traversal attempts such as `../../etc/passwd` are blocked at this layer.
+- Scan optimizations in `sync_service.py` call `find` through
+  `asyncio.create_subprocess_exec()` with explicit argument lists. Project paths
+  are passed as data, not interpolated into shell strings.
+- Auto-update code uses hardcoded commands, list-form arguments, and
+  `stdin=DEVNULL`. User-controlled strings do not reach a shell there.
+
+### MCP Client-Side Risk
+
+Recent MCP ecosystem research has highlighted a client-side pattern where an
+MCP host can be configured to run arbitrary commands as "servers." That risk is
+in the host configuration, not in notes or Basic Memory tool input.
+
+The recommended Basic Memory MCP configuration uses a known command with
+explicit arguments:
+
+```json
+{
+  "mcpServers": {
+    "basic-memory": {
+      "command": "uvx",
+      "args": ["basic-memory", "mcp"]
+    }
+  }
+}
+```
+
+Only add MCP server entries from sources you trust. Avoid inline shell scripts
+or command strings copied from untrusted sources. Treat third-party MCP server
+configuration with the same scrutiny as any locally executed program.
+
+Related ecosystem context:
+
+- OX Security: The Mother of All AI Supply Chains
+- CSO Online: RCE by design: MCP architectural choice haunts AI agent ecosystem
+
+### Out Of Scope
+
+- Basic Memory does not execute note content as code. Notes are returned as
+  data to the LLM.
+- Basic Memory does not open network ports by default. The MCP server uses
+  stdio; the optional REST API is intended for localhost use.
+- Basic Memory is designed for single-user local knowledge bases and does not
+  implement access controls between operating-system users.
+
+## Secure Configuration Checklist
+
+- MCP config `command` points to `uvx` or a trusted binary, not a shell string.
+- Project paths in Basic Memory config come from trusted local configuration.
+- If exposing the REST API, bind it only to localhost.
+- Review any third-party MCP servers before adding them to your host config.

docker-compose-postgres.yml

@@ -1,5 +1,8 @@
 # Docker Compose configuration for Basic Memory with PostgreSQL
-# Use this for local development and testing with Postgres backend
+# Use this for local development and testing with Postgres backend.
+#
+# The Postgres backend requires the pgvector extension (semantic search).
+# This image bundles pgvector; plain postgres:17 will not work for vector search.
 #
 # Usage:
 #   docker-compose -f docker-compose-postgres.yml up -d
@@ -7,7 +10,7 @@
 
 services:
   postgres:
-    image: postgres:17
+    image: pgvector/pgvector:pg17
     container_name: basic-memory-postgres
     environment:
       # Local development/test credentials - NOT for production

docker-compose.yml

@@ -17,7 +17,9 @@ services:
     volumes:
 
       # Persistent storage for configuration and database
-      - basic-memory-config:/root/.basic-memory:rw
+      # Container runs as `appuser` (Dockerfile USER directive), so the CLI
+      # config dir lives under /home/appuser, not /root.
+      - basic-memory-config:/home/appuser/.basic-memory:rw
 
       # Mount your knowledge directory (required)
       # Change './knowledge' to your actual Obsidian vault or knowledge directory

docs/ENGINEERING_STYLE.md

@@ -0,0 +1,64 @@
+# Basic Memory Engineering Style
+
+Style is how we make code easier to verify. Prefer explicit, typed, local-first code that
+preserves the file system as the source of truth while keeping the database, API, and MCP
+surfaces in sync.
+
+## Design Center
+
+- Basic Memory is local-first. Markdown files are the durable source; SQLite/Postgres indexes
+  are derived state that should be rebuilt or reconciled from files when needed.
+- Keep the existing boundary order: CLI/MCP/API entrypoints compose dependencies, services own
+  business behavior, repositories own database access, and file services own filesystem writes.
+- MCP tools should remain atomic and composable. They should call API routers through typed MCP
+  clients, not reach around into services.
+- Prefer small, explicit abstractions that match a real domain boundary. Avoid object
+  hierarchies when a function, dataclass, type alias, or protocol describes the concept better.
+
+## Types And Data
+
+- Use full type annotations and Python 3.12 syntax. Introduce `type` aliases for repeated
+  structured shapes, callback signatures, or domain concepts that would otherwise become
+  anonymous `dict[str, Any]` values.
+- Use dataclasses for internal values, operation inputs, and service results. Prefer
+  `frozen=True` when the value should not change and `slots=True` when identity/dynamic
+  attributes are not needed.
+- Use Pydantic v2 at boundaries that validate, serialize, or deserialize data: API payloads,
+  CLI/MCP schemas, configuration, and persistence-adjacent schemas.
+- Use narrow `Protocol`s when a caller needs a capability rather than a concrete repository or
+  service. Keep protocols small enough that fake implementations in tests are obvious.
+- Avoid speculative `getattr`, broad casts, or `Any` as a way to paper over uncertainty. Read
+  the model or schema definition and make the type relationship explicit.
+
+## Control Flow And Resources
+
+- Fail fast when an invariant is broken. Do not swallow exceptions, add warning-only error
+  handling, or introduce fallback behavior unless the user explicitly agrees to that behavior.
+- Keep control flow simple and close to the domain decision. Push `if` statements up into the
+  function that owns orchestration; keep leaf helpers focused on computation or one side effect.
+- Make async/resource boundaries visible with context managers and explicit lifecycles. Do not
+  start background work without a clear owner, cancellation story, and verification path.
+- Keep file mutations centralized through the existing file utilities/services so checksum,
+  atomic write, and index synchronization behavior stays coherent.
+
+## Testing And Verification
+
+- Use evidence-first testing, not mechanical TDD. For bugs and risky behavior, add or update a
+  regression test that would catch the failure. For small documentation-only edits, use the
+  relevant doc/repo hygiene checks.
+- Prefer tests that exercise real code paths. Use mocks, doubles, or `monkeypatch` only when
+  the external boundary would be slow, nondeterministic, or impossible to trigger directly.
+- Keep coverage at 100% for new code. Use `# pragma: no cover` only for code that would require
+  disproportionate mocking and is covered through an integration or runtime path.
+- Start with targeted commands, then widen as risk grows: focused pytest, `just fast-check`,
+  `just doctor`, package checks for agent packaging changes, and full SQLite/Postgres gates
+  when behavior crosses shared boundaries.
+
+## Comments And Names
+
+- Name values after the domain concept they carry: project, entity, permalink, tenant, route,
+  checksum, observation, relation, batch, or index state.
+- Comments should say why a branch, invariant, retry, lifecycle, or compatibility constraint
+  exists. Section headers are useful when a function or file has clear phases.
+- Avoid comments that restate the code. If a comment cannot explain a decision, simplify the
+  code or improve the name instead.

docs/NOTE-FORMAT.md

@@ -106,7 +106,7 @@ The parser excludes these list item patterns:
 |---------|---------|--------|
 | Checkboxes | `- [ ] Todo item`, `- [x] Done`, `- [-] Cancelled` | Task list syntax |
 | Markdown links | `- [text](url)` | URL link syntax |
-| Bare wiki links | `- [[Target]]` | Treated as a relation instead |
+| Bare wiki links | `- [[Target]]` | Treated as a `links_to` relation instead |
 
 A list item with `#tags` but no `[category]` is still parsed — the tags are extracted and the category defaults to `Note`.
 
@@ -116,47 +116,65 @@ Relations connect documents to form the knowledge graph. There are two kinds.
 
 ### Explicit Relations
 
-Written as list items with a relation type and a `[[wiki link]]` target.
+Written as list items with a relation type and a `[[wiki link]]` target. Unquoted
+relation types are single tokens. Quote relation types that contain spaces.
 
 **Syntax:**
 
 ```
 - relation_type [[Target Entity]] (context)
+- "multi word relation type" [[Target Entity]] (context)
+- 'multi word relation type' [[Target Entity]] (context)
 ```
 
 | Part | Required | Description |
 |------|----------|-------------|
-| `relation_type` | No | Text before `[[`. Defaults to `relates_to` if omitted. |
+| `relation_type` | Yes | Single unquoted token before `[[`, or quoted text for multi-word labels. |
 | `[[Target]]` | Yes | Wiki link to the target entity. Matched by title or permalink. |
 | `(context)` | No | Parenthesized text after `]]`. Supporting details. |
 
 ### Examples
 
+Explicit relations:
+
 ```markdown
 - implements [[Search Design]]
 - depends_on [[Database Schema]]
 - works_at [[Y Combinator]] (co-founder)
-- [[Some Entity]]
+- "based on" [[Customer Interview]]
+- 'in response to' [[Incident Review]]
 ```
 
-The last example — a bare `[[wiki link]]` in a list item — gets relation type `relates_to`.
+Bare wiki links and prose list items create implicit `links_to` relations:
+
+```markdown
+- [[Some Entity]]
+- some other thing [[Some Entity]]
+```
+
+Both examples above create `links_to [[Some Entity]]`. Use quotes when the words before
+`[[` are meant to be a multi-word relation type.
 
 Common relation types:
 - `implements`, `depends_on`, `relates_to`, `inspired_by`
 - `extends`, `part_of`, `contains`, `pairs_with`
 - `works_at`, `authored`, `collaborated_with`
 
-Any text works as a relation type. These are conventions, not a fixed set.
+Any single-token text or quoted text works as a relation type. These are conventions,
+not a fixed set.
 
 ### Inline References
 
-Wiki links appearing in regular prose (not as list items) create implicit `links_to` relations.
+Wiki links appearing in regular prose create implicit `links_to` relations. This includes
+list items that do not match the explicit relation grammar above.
 
 ```markdown
 This builds on [[Core Design]] and uses [[Utility Functions]].
+- We should revisit [[Search Design]] after the API changes.
 ```
 
-This creates two relations: `links_to [[Core Design]]` and `links_to [[Utility Functions]]`.
+This creates three relations: `links_to [[Core Design]]`, `links_to [[Utility Functions]]`,
+and `links_to [[Search Design]]`.
 
 ### Forward References
 

docs/ai-assistant-guide-extended.md

@@ -91,10 +91,11 @@ SQLite Database (Index)
 # List all projects
 projects = await list_memory_projects()
 
-# Response structure:
+# Response structure (each entry includes external_id you can pass as project_id):
 # [
 #   {
 #     "name": "main",
+#     "external_id": "550e8400-e29b-41d4-a716-446655440000",
 #     "path": "/Users/name/notes",
 #     "is_default": True,
 #     "note_count": 156,
@@ -102,6 +103,7 @@ projects = await list_memory_projects()
 #   },
 #   {
 #     "name": "work",
+#     "external_id": "9f86d081-884c-42a3-b5e3-1c0c5b4c8e52",
 #     "path": "/Users/name/work-notes",
 #     "is_default": False,
 #     "note_count": 89,
@@ -164,6 +166,44 @@ active_project = "main"
 results = await search_notes(query="topic", project=active_project)
 ```
 
+### `project` vs `project_id`
+
+Every project has two identifiers:
+
+- **`project`** — human-readable name (e.g., `"main"`). Easy to use, but can collide across cloud workspaces.
+- **`project_id`** — stable `external_id` UUID. Always unambiguous; takes precedence over `project` when both are passed.
+
+**When to prefer `project_id`:**
+
+1. **Cloud multi-workspace setups.** If the user belongs to more than one workspace (personal + organization, or several organizations) and the same project name might exist in more than one of them, pass `project_id` to route to the exact project. Without it, name resolution falls back to the default workspace, which may not be the one the user means.
+2. **After `list_memory_projects()`.** Once you have the `external_id`, prefer using it — it's the same number of characters in JSON and saves a name-resolution round-trip.
+3. **When persisting a project choice across a long session.** UUIDs are stable; names can be renamed.
+
+**When `project` (name) is fine:**
+
+- Local single-workspace setups (no collision risk).
+- One-off operations where the name is clearly visible to the user (e.g., quick `search_notes(project="main", ...)`).
+- The user explicitly references a project by name in their message.
+
+**Example — cloud multi-workspace pattern:**
+
+```python
+# Discover and pick the right project for this user
+projects = await list_memory_projects()
+target = next(p for p in projects if p["name"] == "research" and p["workspace"]["slug"] == "acme")
+
+# Use the UUID for all subsequent operations — no ambiguity
+await write_note(
+    title="Meeting Notes",
+    content="...",
+    folder="meetings",
+    project_id=target["external_id"],
+)
+results = await search_notes(query="kickoff", project_id=target["external_id"])
+```
+
+**Precedence rule:** When both are passed, `project_id` wins. This lets you safely supply `project="main"` for backward compatibility while still routing precisely with `project_id`.
+
 ### Cross-Project Operations
 
 **Some tools work across all projects when project parameter omitted:**

docs/cloud-cli.md

@@ -8,9 +8,25 @@ The cloud CLI enables you to:
 - **Authenticate cloud access** - OAuth/API key credentials are stored locally for cloud operations
 - **Project-scoped sync** - Each project independently manages its sync configuration
 - **Explicit operations** - Sync only what you want, when you want
-- **Bidirectional sync** - Keep local and cloud in sync with rclone bisync
+- **Team-safe push/pull** - Additive, git-style transfers that work on shared Team workspaces
+- **Bidirectional sync** - Keep local and cloud in sync with rclone bisync (Personal workspaces)
 - **Offline access** - Work locally, sync when ready
 
+### Personal vs Team workspaces
+
+The transfer commands fall into two groups:
+
+| Command | Direction | Behavior | Personal | Team |
+|---|---|---|---|---|
+| `bm cloud pull` | cloud → local | **additive** — never deletes local | ✅ | ✅ |
+| `bm cloud push` | local → cloud | **additive** — never deletes cloud | ✅ | ✅ |
+| `bm cloud sync` | local → cloud | **mirror** — deletes cloud files missing locally | ✅ | ❌ |
+| `bm cloud bisync` | local ↔ cloud | **mirror** — two-way, deletes on both sides | ✅ | ❌ |
+
+`sync` and `bisync` are mirror operations: one local tree becomes authoritative and files missing on the other side get deleted. That is correct for a Personal workspace (one user, one source of truth) but unsafe on a shared Team bucket, where it could delete a teammate's files. On Team workspaces these commands exit early with a clear error and point you at `push`/`pull`.
+
+`push` and `pull` are additive (they use `rclone copy`, which never deletes on the destination), so they are safe on both Personal and Team workspaces.
+
 ## Prerequisites
 
 Before using Basic Memory Cloud, you need:
@@ -55,8 +71,8 @@ bm project add work --cloud --local-path ~/work-notes
 bm project add temp --cloud  # No local sync
 
 # Now you can sync individually (after initial --resync):
-bm project bisync --name research
-bm project bisync --name work
+bm cloud bisync --name research
+bm cloud bisync --name work
 # temp stays cloud-only
 ```
 
@@ -94,13 +110,18 @@ bm cloud setup
 ```
 
 **What this does:**
-1. Installs rclone automatically (if needed)
+1. Installs rclone with a supported package manager (if needed)
 2. Fetches your tenant information from cloud
 3. Generates scoped S3 credentials for sync
 4. Configures single rclone remote: `basic-memory-cloud`
 
 **Result:** You're ready to sync projects. No sync directories created yet - those come with project setup.
 
+Rclone setup uses package managers such as Homebrew, MacPorts, apt, dnf, yum, pacman,
+zypper, snap, winget, Chocolatey, or Scoop when available. It does not run remote
+install scripts with `sudo`; if no supported package manager is found, the CLI prints
+manual install instructions.
+
 ### 3. Add Projects with Sync
 
 Create projects with optional local sync paths:
@@ -113,7 +134,7 @@ bm project add research --cloud
 bm project add research --cloud --local-path ~/Documents/research
 
 # Or configure sync for existing project
-bm project sync-setup research ~/Documents/research
+bm cloud sync-setup research ~/Documents/research
 ```
 
 **What happens under the covers:**
@@ -132,10 +153,10 @@ Establish the initial sync baseline. **Best practice:** Always preview with `--d
 
 ```bash
 # Step 1: Preview the initial sync (recommended)
-bm project bisync --name research --resync --dry-run
+bm cloud bisync --name research --resync --dry-run
 
 # Step 2: If all looks good, run the actual sync
-bm project bisync --name research --resync
+bm cloud bisync --name research --resync
 ```
 
 **What happens under the covers:**
@@ -162,7 +183,7 @@ This will effectively make both Path1 and Path2 filesystems contain a matching s
 After the first sync, just run bisync without `--resync`:
 
 ```bash
-bm project bisync --name research
+bm cloud bisync --name research
 ```
 
 **What happens:**
@@ -230,13 +251,13 @@ bm project add research --cloud --local-path ~/Documents/research
 - Stores sync config in `~/.basic-memory/config.json`
 - Prepares for bisync (but doesn't sync yet)
 
-**Result:** Project ready to sync. Run `bm project bisync --name research --resync` to establish baseline.
+**Result:** Project ready to sync. Run `bm cloud bisync --name research --resync` to establish baseline.
 
 **Use case 3: Add sync to existing cloud project**
 
 ```bash
 # Project already exists on cloud
-bm project sync-setup research ~/Documents/research
+bm cloud sync-setup research ~/Documents/research
 ```
 
 **What this does:**
@@ -289,18 +310,98 @@ For MCP stdio, routing is always local.
 
 ### Understanding the Sync Commands
 
-**There are three sync-related commands:**
+**There are five sync-related commands:**
 
-1. `bm project sync` - One-way: local → cloud (make cloud match local)
-2. `bm project bisync` - Two-way: local ↔ cloud (recommended)
-3. `bm project check` - Verify files match (no changes)
+| Command | Direction | Workspace | Summary |
+|---|---|---|---|
+| `bm cloud pull` | cloud → local | Personal + Team | Fetch cloud changes, additively (git-style) |
+| `bm cloud push` | local → cloud | Personal + Team | Upload local changes, additively (git-style) |
+| `bm cloud sync` | local → cloud | Personal only | One-way mirror (cloud becomes identical to local) |
+| `bm cloud bisync` | local ↔ cloud | Personal only | Two-way mirror (recommended for solo use) |
+| `bm cloud check` | — | Personal only | Verify mirror integrity (no changes) |
 
-### One-Way Sync: Local → Cloud
+If you collaborate on a shared Team workspace, use **`push`/`pull`** (see [Team Workspaces](#team-workspaces-push--pull-additive-git-style)). If you are the only writer (a Personal workspace), the mirror commands `sync`/`bisync` give you a single source of truth.
+
+### Team Workspaces: push / pull (additive, git-style)
+
+`push` and `pull` are the Team-safe transfer commands. They model `git push` / `git pull`:
+
+- **`bm cloud pull`** fetches changes from the cloud into your local directory.
+- **`bm cloud push`** uploads your local changes to the cloud.
+
+Both use `rclone copy`, so they are **additive — they never delete on the destination**. A conflict (a file that differs on both sides) is never resolved silently: by default the command aborts and lists the conflicting files, exactly like git refusing to clobber your changes.
+
+#### Pull: fetch cloud changes
+
+```bash
+# Preview first (recommended)
+bm cloud pull --name research --dry-run
+
+# Fetch new/changed cloud files into local
+bm cloud pull --name research
+```
+
+**What happens:**
+1. Compares cloud and local with `rclone check`
+2. Downloads files that are new or changed on the cloud
+3. Leaves your local-only files untouched (never deletes local)
+4. If any file differs on both sides, aborts and lists the conflicts (unless you pass `--on-conflict`)
+
+#### Push: upload local changes
+
+```bash
+bm cloud push --name research --dry-run
+bm cloud push --name research
+```
+
+**What happens:**
+1. Compares local and cloud with `rclone check`
+2. Uploads files that are new or changed locally
+3. Leaves cloud-only files untouched (never deletes cloud)
+4. If any file differs on both sides, aborts and lists the conflicts — pull first, like a rejected `git push`
+
+#### Resolving conflicts
+
+When `push`/`pull` reports conflicts, re-run with `--on-conflict` to choose how differing files are handled. The value names exactly what survives, so it reads the same in both directions:
+
+| `--on-conflict` | Behavior |
+|---|---|
+| `fail` *(default)* | List the conflicting files and exit without transferring anything |
+| `keep-cloud` | Take the cloud version (pull: overwrite local; push: skip those files) |
+| `keep-local` | Keep the local version (pull: skip those files; push: overwrite cloud) |
+| `keep-both` | Keep both — write the incoming version beside the existing one as `name.conflict-<date>.md` |
+
+```bash
+# A teammate edited notes you also changed locally — pull reports a conflict:
+bm cloud pull --name research
+# pull aborted: 1 file(s) differ between local and cloud.
+#   * notes/decisions.md
+# Re-run with one of: --on-conflict keep-cloud | keep-local | keep-both
+
+# Take the cloud copy:
+bm cloud pull --name research --on-conflict keep-cloud
+
+# Or keep both versions to merge by hand:
+bm cloud pull --name research --on-conflict keep-both
+```
+
+#### Limitations
+
+`push`/`pull` are deliberately simple, conflict-aware byte transfers — not a full reconciler. Without a sync baseline:
+
+- **Deletions are not propagated.** A note deleted on one side is not removed from the other (we cannot tell an intentional delete from a file the other side never had). This is surfaced in the command output.
+- **Every divergence is treated as a conflict.** We cannot tell a teammate's edit from your stale copy, so any differing file prompts a decision rather than auto-resolving.
+
+For conflict-aware *editing*, write through the MCP/API tools (which merge at the note level). A Team-safe bidirectional reconciler with a real baseline is tracked in [issue #862](https://github.com/basicmachines-co/basic-memory/issues/862).
+
+### One-Way Sync: Local → Cloud (Personal only)
 
 **Use case:** You made changes locally and want to push to cloud (overwrite cloud).
 
+> **Personal workspaces only.** `sync` is a destructive mirror — it deletes cloud files that are not present locally. On a Team workspace it would delete a teammate's files, so it is blocked there. Use `bm cloud push` (additive) on Team workspaces.
+
 ```bash
-bm project sync --name research
+bm cloud sync --name research
 ```
 
 **What happens:**
@@ -316,16 +417,18 @@ bm project sync --name research
 - You want to force cloud to match local
 - You don't care about cloud changes
 
-### Two-Way Sync: Local ↔ Cloud (Recommended)
+### Two-Way Sync: Local ↔ Cloud (Personal only, recommended for solo use)
 
 **Use case:** You edit files both locally and in cloud UI, want both to stay in sync.
 
+> **Personal workspaces only.** `bisync` is a two-way mirror that can delete and overwrite on both sides. It is blocked on Team workspaces — use `bm cloud pull` then `bm cloud push` there. A Team-safe bidirectional reconciler is tracked separately ([issue #862](https://github.com/basicmachines-co/basic-memory/issues/862)).
+
 ```bash
 # First time - establish baseline
-bm project bisync --name research --resync
+bm cloud bisync --name research --resync
 
 # Subsequent syncs
-bm project bisync --name research
+bm cloud bisync --name research
 ```
 
 **What happens:**
@@ -344,7 +447,7 @@ echo "Local change" > ~/Documents/research/notes.md
 # Cloud now has: "Cloud change"
 
 # Run bisync
-bm project bisync --name research
+bm cloud bisync --name research
 
 # Result: Newer file wins (based on modification time)
 # If cloud was more recent, cloud version kept
@@ -356,12 +459,14 @@ bm project bisync --name research
 - You edit in multiple places
 - You want automatic conflict resolution
 
-### Verify Sync Integrity
+### Verify Sync Integrity (Personal only)
 
 **Use case:** Check if local and cloud match without making changes.
 
+> **Personal workspaces only.** `check` compares against the Personal workspace mirror remote, like `sync`/`bisync`. On Team workspaces use `bm cloud pull --dry-run` / `bm cloud push --dry-run` to preview differences instead.
+
 ```bash
-bm project check --name research
+bm cloud check --name research
 ```
 
 **What happens:**
@@ -373,7 +478,7 @@ bm project check --name research
 
 ```bash
 # One-way check (faster)
-bm project check --name research --one-way
+bm cloud check --name research --one-way
 ```
 
 ### Preview Changes (Dry Run)
@@ -381,7 +486,7 @@ bm project check --name research --one-way
 **Use case:** See what would change without actually syncing.
 
 ```bash
-bm project bisync --name research --dry-run
+bm cloud bisync --name research --dry-run
 ```
 
 **What happens:**
@@ -427,20 +532,20 @@ bm project add work --cloud --local-path ~/work-notes
 bm project add personal --cloud --local-path ~/personal
 
 # Establish baselines
-bm project bisync --name research --resync
-bm project bisync --name work --resync
-bm project bisync --name personal --resync
+bm cloud bisync --name research --resync
+bm cloud bisync --name work --resync
+bm cloud bisync --name personal --resync
 
 # Daily workflow: sync everything
-bm project bisync --name research
-bm project bisync --name work
-bm project bisync --name personal
+bm cloud bisync --name research
+bm cloud bisync --name work
+bm cloud bisync --name personal
 ```
 
 **Future:** `--all` flag will sync all configured projects:
 
 ```bash
-bm project bisync --all  # Coming soon
+bm cloud bisync --all  # Coming soon
 ```
 
 ### Mixed Usage
@@ -457,8 +562,8 @@ bm project add archive --cloud
 bm project add temp-notes --cloud
 
 # Sync only the configured ones
-bm project bisync --name research
-bm project bisync --name work
+bm cloud bisync --name research
+bm cloud bisync --name work
 
 # Archive and temp-notes stay cloud-only
 ```
@@ -485,6 +590,9 @@ bm cloud create-key "my-laptop"    # Creates key and saves it locally
 ```
 
 The API key is account-level — it grants access to all your cloud projects. It's stored in `~/.basic-memory/config.json` as `cloud_api_key`.
+On POSIX systems, Basic Memory writes `~/.basic-memory/` as user-private (`0700`) and
+`config.json` as user-read/write only (`0600`). Treat this config file as a credential
+file when an API key is saved.
 
 ### Setting Project Modes
 
@@ -574,33 +682,62 @@ bm cloud logout
 **Default patterns:**
 
 ```gitignore
-# Version control
-.git/**
-
-# Python
-__pycache__/**
-*.pyc
-.venv/**
-venv/**
-
-# Node.js
-node_modules/**
+# Hidden files and directories
+.*
 
 # Basic Memory internals
-memory.db/**
-memory.db-shm/**
-memory.db-wal/**
-config.json/**
-watch-status.json/**
-.bmignore.rclone/**
+*.db
+*.db-shm
+*.db-wal
+config.json
+
+# Version control
+.git
+.svn
+
+# Python
+__pycache__
+*.pyc
+*.pyo
+*.pyd
+.pytest_cache
+.coverage
+*.egg-info
+.tox
+.mypy_cache
+.ruff_cache
+
+# Virtual environments
+.venv
+venv
+env
+.env
+
+# Node.js
+node_modules
+
+# Build artifacts
+build
+dist
+.cache
+
+# IDE
+.idea
+.vscode
 
 # OS files
-.DS_Store/**
-Thumbs.db/**
+.DS_Store
+Thumbs.db
+desktop.ini
 
-# Environment files
-.env/**
-.env.local/**
+# Obsidian
+.obsidian
+
+# Temporary files
+*.tmp
+*.swp
+*.swo
+*~
 ```
 
 **How it works:**
@@ -609,6 +746,11 @@ Thumbs.db/**
 3. Rclone uses filters during sync
 4. Same patterns used by all projects
 
+During conversion, file patterns exclude the direct match and recursive contents.
+For example, `config.json` becomes both `- config.json` and `- config.json/**`,
+while `.*` becomes both `- .*` and `- .*/**`. Directory-only patterns keep
+their trailing slash, so `cache/` becomes `- cache/` and `- cache/**`.
+
 **Customizing:**
 
 ```bash
@@ -616,14 +758,41 @@ Thumbs.db/**
 code ~/.basic-memory/.bmignore
 
 # Add custom patterns
-echo "*.tmp/**" >> ~/.basic-memory/.bmignore
+echo "*.tmp" >> ~/.basic-memory/.bmignore
 
 # Next sync uses updated patterns
-bm project bisync --name research
+bm cloud bisync --name research
 ```
 
 ## Troubleshooting
 
+### Rclone Setup Cannot Install Automatically
+
+**Problem:** `bm cloud setup` cannot find a supported package manager, or package-manager
+installation fails.
+
+**Explanation:** The CLI avoids remote privileged install scripts. It only invokes known
+package managers and otherwise asks you to install rclone manually.
+
+**Solution:** Install rclone with your OS package manager, then rerun setup:
+
+```bash
+# macOS
+brew install rclone
+
+# Debian/Ubuntu
+sudo apt install rclone
+
+# Fedora
+sudo dnf install rclone
+
+# Arch
+sudo pacman -S rclone
+
+# After rclone is on PATH
+bm cloud setup
+```
+
 ### Authentication Issues
 
 **Problem:** "Authentication failed" or "Invalid token"
@@ -655,7 +824,7 @@ bm cloud login
 **Solution:**
 
 ```bash
-bm project bisync --name research --resync
+bm cloud bisync --name research --resync
 ```
 
 **What this does:**
@@ -678,7 +847,7 @@ bm project bisync --name research --resync
 echo "# Research Notes" > ~/Documents/research/README.md
 
 # Now run bisync
-bm project bisync --name research --resync
+bm cloud bisync --name research --resync
 ```
 
 **Why this happens:** Bisync creates listing files that track the state of each side. When both directories are completely empty, these listing files are considered invalid by rclone.
@@ -695,10 +864,10 @@ bm project bisync --name research --resync
 
 ```bash
 # Clear bisync state
-bm project bisync-reset research
+bm cloud bisync-reset research
 
 # Re-establish baseline
-bm project bisync --name research --resync
+bm cloud bisync --name research --resync
 ```
 
 **What this does:**
@@ -718,16 +887,16 @@ bm project bisync --name research --resync
 
 ```bash
 # Check what would be deleted
-bm project bisync --name research --dry-run
+bm cloud bisync --name research --dry-run
 
 # If correct, establish new baseline
-bm project bisync --name research --resync
+bm cloud bisync --name research --resync
 ```
 
 **Solution 2:** Use one-way sync if you know local is correct:
 
 ```bash
-bm project sync --name research
+bm cloud sync --name research
 ```
 
 ### Project Not Configured for Sync
@@ -739,8 +908,8 @@ bm project sync --name research
 **Solution:**
 
 ```bash
-bm project sync-setup research ~/Documents/research
-bm project bisync --name research --resync
+bm cloud sync-setup research ~/Documents/research
+bm cloud bisync --name research --resync
 ```
 
 ### Connection Issues
@@ -759,8 +928,10 @@ If instance is down, wait a few minutes and retry.
 
 - **Authentication**: OAuth 2.1 with PKCE flow
 - **Tokens**: Stored securely in `~/.basic-memory/basic-memory-cloud.json`
+- **API keys**: Stored in `~/.basic-memory/config.json`, which is written with private file permissions on POSIX systems
 - **Transport**: All data encrypted in transit (HTTPS)
 - **Credentials**: Scoped S3 credentials (read-write to your tenant only)
+- **Rclone setup**: Uses package managers or manual instructions; no remote privileged install-script fallback
 - **Isolation**: Your data isolated from other tenants
 - **Ignore patterns**: Sensitive files automatically excluded via `.bmignore`
 
@@ -785,7 +956,7 @@ bm cloud create-key <name>  # Create API key via cloud API (requires OAuth login
 ### Setup
 
 ```bash
-bm cloud setup              # Install rclone and configure credentials
+bm cloud setup              # Install rclone via package manager and configure credentials
 ```
 
 ### Project Management
@@ -795,7 +966,7 @@ bm project list --local                   # Local project list
 bm project list --cloud                   # Cloud project list
 bm project add <name> --cloud             # Create cloud project (no sync)
 bm project add <name> --cloud --local-path <path> # Create with local sync
-bm project sync-setup <name> <path>       # Add sync to existing project
+bm cloud sync-setup <name> <path>       # Add sync to existing project
 bm project rm <name>                      # Delete project
 ```
 
@@ -809,20 +980,30 @@ bm project set-local <name>  # Revert project to local mode
 ### File Synchronization
 
 ```bash
-# One-way sync (local → cloud)
-bm project sync --name <project>
-bm project sync --name <project> --dry-run
-bm project sync --name <project> --verbose
+# Pull: fetch cloud changes (cloud → local) - Personal + Team, additive
+bm cloud pull --name <project>
+bm cloud pull --name <project> --dry-run
+bm cloud pull --name <project> --on-conflict [fail|keep-local|keep-cloud|keep-both]
 
-# Two-way sync (local ↔ cloud) - Recommended
-bm project bisync --name <project>          # After first --resync
-bm project bisync --name <project> --resync # First time / force baseline
-bm project bisync --name <project> --dry-run
-bm project bisync --name <project> --verbose
+# Push: upload local changes (local → cloud) - Personal + Team, additive
+bm cloud push --name <project>
+bm cloud push --name <project> --dry-run
+bm cloud push --name <project> --on-conflict [fail|keep-local|keep-cloud|keep-both]
 
-# Integrity check
-bm project check --name <project>
-bm project check --name <project> --one-way
+# One-way mirror (local → cloud) - Personal workspaces only
+bm cloud sync --name <project>
+bm cloud sync --name <project> --dry-run
+bm cloud sync --name <project> --verbose
+
+# Two-way mirror (local ↔ cloud) - Personal workspaces only
+bm cloud bisync --name <project>          # After first --resync
+bm cloud bisync --name <project> --resync # First time / force baseline
+bm cloud bisync --name <project> --dry-run
+bm cloud bisync --name <project> --verbose
+
+# Integrity check - Personal workspaces only
+bm cloud check --name <project>
+bm cloud check --name <project> --one-way
 
 # List project files by route
 bm project ls --name <project>          # Default target: local
@@ -838,15 +1019,25 @@ bm project ls --name <project> --cloud --path <subpath>
 1. **Authenticate cloud access** - `bm cloud login`
 2. **Install rclone** - `bm cloud setup`
 3. **Add projects with sync** - `bm project add research --cloud --local-path ~/Documents/research`
-4. **Preview first sync** - `bm project bisync --name research --resync --dry-run`
-5. **Establish baseline** - `bm project bisync --name research --resync`
-6. **Daily workflow** - `bm project bisync --name research`
+
+**Personal workspace (solo, mirror) workflow:**
+
+4. **Preview first sync** - `bm cloud bisync --name research --resync --dry-run`
+5. **Establish baseline** - `bm cloud bisync --name research --resync`
+6. **Daily workflow** - `bm cloud bisync --name research`
+
+**Team workspace (shared, additive) workflow:**
+
+4. **Fetch teammates' changes** - `bm cloud pull --name research`
+5. **Upload your changes** - `bm cloud push --name research`
+6. **Resolve conflicts explicitly** - re-run with `--on-conflict keep-cloud|keep-local|keep-both`
 
 **Key benefits:**
 - ✅ Each project independently syncs (or doesn't)
 - ✅ Projects can live anywhere on disk
 - ✅ Explicit sync operations (no magic)
-- ✅ Safe by design (max delete limits, conflict resolution)
+- ✅ Team-safe push/pull that never delete on the destination
+- ✅ Safe by design (max delete limits, conflict resolution, git-style conflict aborts)
 - ✅ Full offline access (work locally, sync when ready)
 
 **Future enhancements:**

docs/litellm-provider.md

@@ -0,0 +1,300 @@
+# LiteLLM Provider
+
+Basic Memory can use the LiteLLM SDK for semantic search embeddings. This lets you
+keep Basic Memory's vector indexing and search behavior while routing embedding calls
+to OpenAI-compatible and provider-specific backends such as OpenAI, Azure OpenAI,
+Cohere, Bedrock, NVIDIA NIM, and other LiteLLM-supported embedding providers.
+
+Use this page when you want to try a non-default embedding model, validate a provider,
+or tune LiteLLM-specific settings.
+
+> **Experimental — advanced users only.** The LiteLLM provider is experimental and
+> intended for users who are comfortable operating remote embedding backends. It makes
+> paid, networked API calls, requires per-model dimension and input-role configuration,
+> and reindexing a real corpus can be slow and spend provider quota (see
+> [Reindexing with a remote provider](#reindexing-with-a-remote-provider)). For most
+> users, the default local **FastEmbed** provider is the recommended choice. Use LiteLLM
+> only if you know what you're doing.
+
+## Quick Start
+
+The default LiteLLM model is OpenAI `text-embedding-3-small` through the LiteLLM
+model string `openai/text-embedding-3-small`.
+
+```bash
+export BASIC_MEMORY_SEMANTIC_SEARCH_ENABLED=true
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_PROVIDER=litellm
+export OPENAI_API_KEY=sk-...
+
+bm reindex --embeddings
+```
+
+Then use vector or hybrid search:
+
+```python
+search_notes("login token flow", search_type="hybrid")
+```
+
+## Basic Memory Options
+
+All options can be set in config or as environment variables.
+
+| Config Field | Env Var | Default | Notes |
+|---|---|---|---|
+| `semantic_search_enabled` | `BASIC_MEMORY_SEMANTIC_SEARCH_ENABLED` | Auto | Set to `true` to force vector/hybrid support on. |
+| `semantic_embedding_provider` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_PROVIDER` | `fastembed` | Set to `litellm` for the LiteLLM provider. |
+| `semantic_embedding_model` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_MODEL` | `bge-small-en-v1.5` | With `litellm`, the default is remapped to `openai/text-embedding-3-small`. |
+| `semantic_embedding_dimensions` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_DIMENSIONS` | Provider default | Required for non-default LiteLLM models because vector tables are dimensioned before the first API call. |
+| `semantic_embedding_forward_dimensions` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_FORWARD_DIMENSIONS` | Auto | Sends `dimensions` to LiteLLM only when supported. Auto is enabled for `text-embedding-3` model strings. |
+| `semantic_embedding_document_input_type` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_DOCUMENT_INPUT_TYPE` | Auto | LiteLLM `input_type` for indexed notes/passages. |
+| `semantic_embedding_query_input_type` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_QUERY_INPUT_TYPE` | Auto | LiteLLM `input_type` for search queries. |
+| `semantic_embedding_batch_size` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_BATCH_SIZE` | `2` | Number of text chunks per provider request. |
+| `semantic_embedding_request_concurrency` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_REQUEST_CONCURRENCY` | `4` | Maximum concurrent LiteLLM embedding requests. |
+| `semantic_embedding_sync_batch_size` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_SYNC_BATCH_SIZE` | `2` | Number of prepared vector jobs flushed through the sync pipeline together. |
+
+## Dimensions
+
+Basic Memory needs the vector dimension before it can create SQLite or Postgres
+vector tables. The OpenAI default is known, so this works without an explicit
+dimension:
+
+```bash
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_PROVIDER=litellm
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_MODEL=openai/text-embedding-3-small
+```
+
+For every other LiteLLM model, set the dimension explicitly:
+
+```bash
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_PROVIDER=litellm
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_MODEL=cohere/embed-english-v3.0
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_DIMENSIONS=1024
+```
+
+For fixed-size models, `semantic_embedding_dimensions` is Basic Memory's local
+schema and validation size. For OpenAI/Azure `text-embedding-3` models, LiteLLM
+can also forward `dimensions` as a provider-side reduced-output request. Basic
+Memory enables that automatically when the model string contains `text-embedding-3`.
+
+If you use an Azure deployment alias such as `azure/<deployment-name>`, the model
+string may not reveal that the underlying model supports reduced output dimensions.
+Set this only when your deployment supports it:
+
+```bash
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_FORWARD_DIMENSIONS=true
+```
+
+## Asymmetric Models
+
+Some embedding models use different request roles for indexed documents and
+search queries. Basic Memory automatically sets these for known LiteLLM families:
+
+| Model Family | Document `input_type` | Query `input_type` |
+|---|---|---|
+| Cohere v3 embeddings | `search_document` | `search_query` |
+| NVIDIA NIM retrieval embeddings | `passage` | `query` |
+
+For any other asymmetric model, configure both roles explicitly:
+
+```bash
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_DOCUMENT_INPUT_TYPE=passage
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_QUERY_INPUT_TYPE=query
+```
+
+Changing provider, model, dimensions, dimension-forwarding, or document/query
+roles changes the meaning of stored vectors. Rebuild embeddings after any of
+those changes:
+
+```bash
+bm reindex --embeddings
+```
+
+## Reindexing with a remote provider
+
+Embedding a real corpus through a network API is far slower than local FastEmbed, and
+the defaults are tuned for the local case. Two things to know before you run a full
+reindex.
+
+**Raise the sync batch size.** `semantic_embedding_sync_batch_size` defaults to `2`, and
+it — not `semantic_embedding_batch_size` — governs throughput on the sync pipeline. With
+the default, a full reindex can take tens of seconds *per note* against a remote provider.
+Raising both to a larger value turns a multi-minute (or longer) reindex into well under a
+minute for the same corpus:
+
+```bash
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_SYNC_BATCH_SIZE=32
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_BATCH_SIZE=64
+```
+
+Stay within the provider's per-request size and rate limits — Cohere v3, for example,
+accepts up to 96 inputs per embedding request.
+
+**Changing dimensions requires recreating the vector table.** Basic Memory dimensions the
+vector table on first index and refuses to mix sizes. Switching to a model with a
+different dimension (for example FastEmbed 384 → OpenAI 1536 → Cohere 1024) makes a plain
+`bm reindex` raise an `Embedding dimension mismatch` error. Recreate the table with a full
+rebuild — files are the source of truth, so this re-indexes from disk and re-embeds
+everything:
+
+```bash
+bm reset --reindex
+```
+
+To trial a provider without disturbing your existing index, point Basic Memory at a
+throwaway config + database instead:
+
+```bash
+export BASIC_MEMORY_CONFIG_DIR=/tmp/bm-litellm-trial
+```
+
+## Provider Setup Examples
+
+LiteLLM reads provider credentials from the environment. These are the examples
+covered by Basic Memory's live validation harness.
+
+### OpenAI Through LiteLLM
+
+```bash
+export OPENAI_API_KEY=sk-...
+export BASIC_MEMORY_SEMANTIC_SEARCH_ENABLED=true
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_PROVIDER=litellm
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_MODEL=openai/text-embedding-3-small
+```
+
+### Cohere v3
+
+```bash
+export COHERE_API_KEY=...
+export BASIC_MEMORY_SEMANTIC_SEARCH_ENABLED=true
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_PROVIDER=litellm
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_MODEL=cohere/embed-english-v3.0
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_DIMENSIONS=1024
+```
+
+The provider auto-selects `search_document` for indexed chunks and `search_query`
+for search queries.
+
+### Azure OpenAI
+
+```bash
+export AZURE_API_KEY=...
+export AZURE_API_BASE=https://<resource-name>.openai.azure.com
+export AZURE_API_VERSION=2024-02-01
+
+export BASIC_MEMORY_SEMANTIC_SEARCH_ENABLED=true
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_PROVIDER=litellm
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_MODEL=azure/<deployment-name>
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_DIMENSIONS=1536
+```
+
+If your Azure deployment is a reduced-dimension `text-embedding-3` deployment,
+set the dimension you want and enable forwarding:
+
+```bash
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_DIMENSIONS=512
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_FORWARD_DIMENSIONS=true
+```
+
+### NVIDIA NIM
+
+```bash
+export NVIDIA_NIM_API_KEY=...
+# Optional when using a custom or self-hosted NIM endpoint:
+export NVIDIA_NIM_API_BASE=https://integrate.api.nvidia.com/v1
+
+export BASIC_MEMORY_SEMANTIC_SEARCH_ENABLED=true
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_PROVIDER=litellm
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_MODEL=nvidia_nim/nvidia/embed-qa-4
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_DIMENSIONS=1024
+```
+
+The provider auto-selects `passage` for indexed chunks and `query` for search
+queries.
+
+## Testing LiteLLM Providers
+
+Run the non-live LiteLLM unit and harness tests first:
+
+```bash
+uv run pytest tests/repository/test_litellm_provider.py \
+  test-int/semantic/test_litellm_live_harness.py -q
+```
+
+Run the SQLite and Postgres vector identity regressions when changing model
+identity, role, or vector sync behavior:
+
+```bash
+uv run pytest \
+  tests/repository/test_sqlite_vector_search_repository.py::test_sqlite_embedding_model_key_includes_litellm_role_settings \
+  -q
+
+BASIC_MEMORY_TEST_POSTGRES=1 uv run pytest \
+  tests/repository/test_postgres_search_repository.py::test_postgres_litellm_role_change_reembeds_existing_chunks \
+  -q
+```
+
+The Postgres command uses testcontainers, so Docker must be running.
+
+## Live Provider Harness
+
+The live harness makes real LiteLLM API calls and spends provider quota. It is
+opt-in by design:
+
+```bash
+export OPENAI_API_KEY=sk-...
+export COHERE_API_KEY=...
+
+just test-litellm-live
+```
+
+Built-in cases run when their API keys are present:
+
+| Case | Required Env Var | Validates |
+|---|---|---|
+| `openai-text-embedding-3-small` | `OPENAI_API_KEY` | OpenAI via LiteLLM, 1536 dimensions, normalized vectors, ranking sanity. |
+| `cohere-embed-english-v3` | `COHERE_API_KEY` | Cohere v3 role handling, 1024 dimensions, normalized vectors, ranking sanity. |
+
+Add provider aliases or new backends with a custom cases file:
+
+```bash
+cat > /tmp/litellm-cases.json <<'JSON'
+[
+  {
+    "name": "azure-text-embedding-3-small-512",
+    "model": "azure/<deployment-name>",
+    "dimensions": 512,
+    "api_key_env": "AZURE_API_KEY",
+    "forward_dimensions": true
+  },
+  {
+    "name": "nvidia-embed-qa-4",
+    "model": "nvidia_nim/nvidia/embed-qa-4",
+    "dimensions": 1024,
+    "api_key_env": "NVIDIA_NIM_API_KEY",
+    "document_input_type": "passage",
+    "query_input_type": "query"
+  }
+]
+JSON
+
+just test-litellm-live --cases-file /tmp/litellm-cases.json
+```
+
+For CI-style output:
+
+```bash
+just test-litellm-live --cases-file /tmp/litellm-cases.json --json
+```
+
+The harness embeds two documents and one query, validates dimension and vector
+normalization, checks that the authentication query ranks the authentication
+document above a distractor, and reports latency plus role/dimension settings.
+
+## Provider Reference
+
+LiteLLM's own provider and embedding docs are the source of truth for current
+model strings and credential names:
+
+- [LiteLLM embedding models](https://docs.litellm.ai/docs/embedding/supported_embedding)
+- [LiteLLM Azure OpenAI provider](https://docs.litellm.ai/docs/providers/azure)
+- [LiteLLM NVIDIA NIM provider](https://docs.litellm.ai/docs/providers/nvidia_nim)

docs/logfire-instrumentation-strategy.md

@@ -0,0 +1,499 @@
+# Logfire Instrumentation Strategy
+
+## Why
+
+We want Logfire in Basic Memory for two specific use cases:
+
+1. Local development and performance investigation
+2. Cloud deployments where Basic Memory runs inside Basic Memory Cloud
+
+This instrumentation must be:
+
+- Disabled by default
+- Useful when enabled
+- Safe for local-first users
+- Searchable in Logfire over time
+
+The previous integration added telemetry, but it leaned too much on generic framework instrumentation. That created noisy spans with weak names and made the trace view harder to navigate. This strategy favors manual instrumentation around Basic Memory's real units of work.
+
+## Core Principles
+
+### 1. Default-off
+
+Basic Memory should ship with Logfire disabled unless the operator explicitly enables it.
+
+That means:
+
+- no required token for normal local usage
+- no surprise outbound telemetry
+- no behavior change for existing users
+
+### 2. Manual spans over automatic framework spans
+
+We should not rely on broad auto-instrumentation for FastAPI, MCP, SQLAlchemy, or HTTP as the primary experience.
+
+Why:
+
+- auto-generated span names are often generic
+- routes and middleware produce too many low-signal spans
+- it becomes harder to answer product questions like "why was `write_note` slow?" or "where did sync time go?"
+
+The preferred model is:
+
+- one meaningful root span per high-level operation
+- a small number of child spans for important phases
+- optional targeted instrumentation only where it adds clear value
+
+### 3. Logs must live inside traces
+
+Basic Memory already uses `loguru` pervasively. The Logfire integration should preserve that and make those logs visible inside the active trace/span context.
+
+If traces exist but the logs are detached from them, the integration is not doing its job.
+
+### 4. Stable names, selective attributes
+
+Span names should describe the operation class, not the specific input.
+
+Good:
+
+- `mcp.tool.write_note`
+- `sync.project.scan`
+- `search.execute`
+- `routing.resolve_project`
+
+Bad:
+
+- `Searching for "foo bar baz"`
+- `POST /v2/projects/123/search/`
+- `write note to /specs/api.md`
+
+Dynamic values belong in attributes, not in the span name.
+
+## What We Should Not Do
+
+### Avoid broad FastAPI auto-instrumentation
+
+We should not turn on `instrument_fastapi()` and treat that as the main telemetry story.
+
+It may still be useful in narrowly scoped debugging, but it should not define the production trace shape. The meaningful root spans should come from Basic Memory's own entrypoints and service boundaries.
+
+### Avoid per-file spans by default
+
+`sync` can process many files. A span per file will explode trace cardinality and make performance views noisy.
+
+Default behavior should be:
+
+- one span for the project sync
+- child spans for scan, move handling, delete handling, markdown sync batch, relation resolution, embedding sync, watermark update
+- per-file spans only for failures or very slow outliers
+
+### Avoid high-cardinality attributes on every span
+
+Do not attach large or highly variable values everywhere:
+
+- raw note content
+- file bodies
+- long search text
+- arbitrary metadata blobs
+- unique IDs that make every span shape distinct
+
+Prefer compact, queryable attributes:
+
+- `project_name`
+- `workspace_id`
+- `route_mode`
+- `scan_type`
+- `file_count`
+- `result_count`
+- `search_type`
+- `retrieval_mode`
+- `duration_ms`
+
+## Proposed Architecture
+
+Add a dedicated telemetry module in core Basic Memory, separate from logging setup.
+
+Suggested shape:
+
+```python
+# basic_memory/telemetry.py
+
+def configure_telemetry(service_name: str, *, enable_logfire: bool) -> None: ...
+def telemetry_enabled() -> bool: ...
+def span(name: str, **attrs): ...
+def bind_telemetry_context(**attrs): ...
+```
+
+This module should:
+
+- configure Logfire only when explicitly enabled
+- set up the Logfire `loguru` handler
+- expose lightweight helpers so application code does not import `logfire` directly everywhere
+- degrade cleanly to no-op behavior when disabled
+
+This keeps the rest of the codebase readable and makes it easy to reason about what telemetry is doing.
+
+## Logging Integration Strategy
+
+### Goal
+
+When a span is active, logs emitted through `loguru` during that operation should show up in the same trace.
+
+### Preferred design
+
+1. Configure Logfire once in the telemetry bootstrap
+2. Add the Logfire `loguru` handler to the existing `loguru` configuration
+3. At operation boundaries, bind stable contextual fields with `loguru`
+4. Let logs emitted inside the span inherit the active trace context
+
+### Context to bind
+
+Bind only the fields that help correlate work across the system:
+
+- `service_name`
+- `entrypoint`
+- `project_name`
+- `workspace_id`
+- `route_mode`
+- `tool_name`
+- `command_name`
+
+This binding should happen at the root of an operation, not deep in leaf functions.
+
+### Important nuance
+
+We should not try to encode the entire trace model into logger extras. The logger context should be a human-meaningful slice of the active operation. Trace linkage comes from the active Logfire/OpenTelemetry context; logger extras are there to improve searchability and readability.
+
+## Span Model
+
+### Root spans
+
+Each user-visible or system-visible operation should get one root span.
+
+Examples:
+
+- `cli.command.status`
+- `cli.command.project_sync`
+- `api.request.search`
+- `mcp.tool.write_note`
+- `mcp.tool.read_note`
+- `mcp.tool.search_notes`
+- `sync.project.run`
+- `db.semantic_backfill`
+
+### Child spans
+
+Child spans should represent real phases whose duration we care about.
+
+Examples:
+
+- `routing.client_session`
+- `routing.resolve_project`
+- `routing.resolve_workspace`
+- `api.search.execute`
+- `sync.project.scan`
+- `sync.project.detect_moves`
+- `sync.project.apply_changes`
+- `sync.project.resolve_relations`
+- `sync.project.sync_embeddings`
+- `sync.file.markdown`
+- `sync.file.regular`
+- `search.execute`
+- `search.relaxed_fts_retry`
+- `db.init`
+- `db.migrate`
+
+### Span naming rules
+
+- Use dot-separated names
+- Start with subsystem
+- Keep the verb at the end
+- Keep names stable across runs
+- Never include request-specific text in the span name
+
+## Attribute Taxonomy
+
+### Required attributes on root spans
+
+Every root span should have a small common set:
+
+- `service_name`
+- `entrypoint`
+- `project_name` when applicable
+- `workspace_id` when applicable
+- `route_mode` with values like `local_asgi`, `cloud_proxy`, `factory`
+
+### Operation-specific attributes
+
+Examples:
+
+For search:
+
+- `search_type`
+- `retrieval_mode`
+- `page`
+- `page_size`
+- `result_count`
+- `fallback_used`
+
+For sync:
+
+- `scan_type`
+- `force_full`
+- `new_count`
+- `modified_count`
+- `deleted_count`
+- `move_count`
+- `skipped_count`
+- `embeddings_enabled`
+
+For note operations:
+
+- `tool_name`
+- `note_type`
+- `directory`
+- `overwrite`
+- `output_format`
+
+### Attributes to avoid by default
+
+- full `query.text`
+- full note titles if they create privacy or cardinality issues
+- file content
+- raw frontmatter
+- raw HTTP bodies
+
+If we need richer payloads for a local debugging session, that should be an explicit temporary mode, not the default telemetry shape.
+
+## Instrumentation Plan By Layer
+
+### 1. Entrypoints
+
+Instrument these first:
+
+- `cli.app` callback and major commands
+- API lifespan and selected routers
+- MCP server lifespan
+- MCP tool entrypoints
+
+Why:
+
+- this establishes clean root spans
+- it gives us trace boundaries that match how users think about the product
+
+### 2. Routing and context resolution
+
+Instrument:
+
+- client routing decisions
+- workspace resolution
+- project resolution
+- default-project fallback
+
+Why:
+
+- Basic Memory has local/cloud/per-project routing logic
+- when something is slow or surprising, we need to know which path was taken
+
+### 3. Sync and indexing
+
+This is the highest-value area to instrument deeply.
+
+Instrument:
+
+- sync root
+- scan strategy decision
+- filesystem scan
+- move detection
+- delete handling
+- markdown sync phase
+- relation resolution
+- vector embedding sync
+- scan watermark update
+
+Why:
+
+- this is where performance work will happen
+- cloud and local both benefit from this visibility
+
+### 4. Search
+
+Instrument:
+
+- search execution
+- retrieval mode
+- relaxed FTS fallback
+- result shaping
+
+Why:
+
+- search is user-facing and latency-sensitive
+- hybrid/vector/FTS paths need to be distinguishable
+
+### 5. Database and initialization
+
+Instrument selectively:
+
+- DB init
+- migrations
+- semantic backfill
+- connection mode selection
+
+Avoid full automatic SQL span firehose by default.
+
+## Recommended Rollout Phases
+
+## Task List
+
+- [x] Phase 1: Bootstrap and config gating
+- [x] Phase 2: Root spans for entrypoints and primary operations
+- [x] Phase 3: Child spans for sync, search, and routing
+- [x] Phase 4: Failure-focused detail and final verification
+- [x] Phase 5: Loguru context binding and scoped context inheritance
+
+## Recommended Rollout Phases
+
+### Phase 1: Bootstrap and config gating
+
+Add:
+
+- telemetry bootstrap module
+- config/env gating
+- `loguru` + Logfire handler integration
+
+This gives immediate value with low noise.
+
+### Phase 2: Root spans for entrypoints and primary operations
+
+Add:
+
+- root spans for CLI, API, MCP, and main MCP tools
+- stable root attributes for project, workspace, route mode, and operation type
+
+This gives us clean top-level traces that match how users think about the product.
+
+### Phase 3: Child spans for sync, search, and routing
+
+Add child spans to:
+
+- sync
+- search
+- routing
+
+This is the main performance-investigation layer.
+
+### Phase 4: Failure-focused detail
+
+Add selective deeper spans/log enrichment for:
+
+- sync failures
+- relation resolution failures
+- slow file operations
+- cloud routing/auth failures
+
+This keeps normal traces clean while improving debuggability.
+
+### Phase 5: Loguru context binding and scoped context inheritance
+
+Add:
+
+- context-local telemetry state in `basic_memory.telemetry`
+- a shared `scope(...)` helper that opens a span and binds stable logger context together
+- context inheritance for routing, sync, and search so downstream `loguru` logs carry the active operation fields
+
+This makes the trace view and the log stream tell the same story without forcing logger rewrites across the codebase.
+
+## Local Dev Playbook
+
+The fastest way to sanity-check the current trace shape is:
+
+```bash
+LOGFIRE_TOKEN=lf_... just telemetry-smoke
+```
+
+What this does:
+
+- creates an isolated temp home, config dir, and project path
+- enables Logfire for the run
+- automatically exports to Logfire when `LOGFIRE_TOKEN` is present
+- defaults `BASIC_MEMORY_SEMANTIC_SEARCH_ENABLED=false` so the smoke run stays fast and trace-friendly
+- disables promo telemetry so the trace is about Basic Memory work, not analytics noise
+- runs a small CLI workflow:
+  - `project add`
+  - `tool write-note`
+  - `tool read-note`
+  - `tool edit-note`
+  - `tool build-context`
+  - `tool search-notes`
+  - `doctor`
+
+If you want to exercise the instrumentation without exporting anything upstream:
+
+```bash
+BASIC_MEMORY_LOGFIRE_SEND_TO_LOGFIRE=false just telemetry-smoke
+```
+
+If you want the smoke run to include vector or hybrid retrieval spans too:
+
+```bash
+LOGFIRE_TOKEN=lf_... BASIC_MEMORY_SEMANTIC_SEARCH_ENABLED=true just telemetry-smoke
+```
+
+The recipe sets `BASIC_MEMORY_LOGFIRE_ENVIRONMENT=telemetry-smoke` by default so these traces are easy to isolate in Logfire. Override it if you want the smoke traces grouped under a different environment name.
+
+### What to look for
+
+You should see a small set of comparable root spans rather than a framework-generated span forest:
+
+- `cli.command.project`
+- `cli.command.tool`
+- `mcp.tool.write_note`
+- `mcp.tool.read_note`
+- `mcp.tool.edit_note`
+- `mcp.tool.build_context`
+- `mcp.tool.search_notes`
+- `sync.project.run`
+
+You should also see correlated logs under those traces with stable fields like:
+
+- `project_name`
+- `route_mode`
+- `tool_name`
+- `entrypoint`
+
+### Expected nuance
+
+`doctor` creates its own temporary project on purpose. That means the sync trace will usually show a different project name than the `telemetry-smoke` write/search traces. That is fine for smoke testing because the goal is to confirm:
+
+- root span names are meaningful
+- scoped logs stay attached to the active trace
+- routing, tool, search, and sync phases are easy to distinguish
+
+## Validation Checklist
+
+We should consider the integration successful when the following are true:
+
+1. With telemetry disabled, Basic Memory behaves exactly as it does today.
+2. With telemetry enabled, one user action produces one obvious root span.
+3. Logs emitted during that action are visible inside the same trace.
+4. A search in Logfire for `mcp.tool.write_note` or `sync.project.run` returns comparable spans across runs.
+5. Trace views show phase timing clearly without drowning in framework noise.
+6. Sensitive payloads are not captured by default.
+
+## Immediate Implementation Direction
+
+When we start coding, the first pass should be:
+
+1. Add `basic_memory.telemetry`
+2. Add config/env switches for `enabled`, `send_to_logfire`, and service name
+3. Wire telemetry bootstrap into CLI, API, and MCP entrypoints
+4. Configure `loguru` to emit to both existing sinks and the Logfire handler when enabled
+5. Add manual root spans around:
+   - CLI commands
+   - API request handlers we care about
+   - MCP tool entrypoints
+   - sync root
+   - search root
+6. Add child spans to the sync and routing phases only after the root span model feels clean
+
+That gives us a strong foundation without repeating the earlier "turn on instrumentation everywhere" approach.

docs/semantic-search.md

@@ -99,10 +99,13 @@ All settings are fields on `BasicMemoryConfig` and can be set via environment va
 | Config Field | Env Var | Default | Description |
 |---|---|---|---|
 | `semantic_search_enabled` | `BASIC_MEMORY_SEMANTIC_SEARCH_ENABLED` | Auto (`true` when semantic deps are available) | Enable semantic search. Required before vector/hybrid modes work. |
-| `semantic_embedding_provider` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_PROVIDER` | `"fastembed"` | Embedding provider: `"fastembed"` (local) or `"openai"` (API). |
+| `semantic_embedding_provider` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_PROVIDER` | `"fastembed"` | Embedding provider: `"fastembed"` (local), `"openai"` (API), or `"litellm"` (multi-provider API, **experimental** — advanced users only). |
 | `semantic_embedding_model` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_MODEL` | `"bge-small-en-v1.5"` | Model identifier. Auto-adjusted per provider if left at default. |
-| `semantic_embedding_dimensions` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_DIMENSIONS` | Auto-detected | Vector dimensions. 384 for FastEmbed, 1536 for OpenAI. Override only if using a non-default model. |
-| `semantic_embedding_batch_size` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_BATCH_SIZE` | `64` | Number of texts to embed per batch. |
+| `semantic_embedding_dimensions` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_DIMENSIONS` | Provider default | Vector dimensions. 384 for FastEmbed, 1536 for OpenAI/LiteLLM OpenAI. Required when using a non-default LiteLLM model. |
+| `semantic_embedding_forward_dimensions` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_FORWARD_DIMENSIONS` | Auto | LiteLLM-only override for whether configured dimensions are sent as a provider-side output-size request. |
+| `semantic_embedding_batch_size` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_BATCH_SIZE` | `2` | Number of texts to embed per batch. |
+| `semantic_embedding_document_input_type` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_DOCUMENT_INPUT_TYPE` | Auto for known LiteLLM models | Optional LiteLLM `input_type` for indexed document/passages. |
+| `semantic_embedding_query_input_type` | `BASIC_MEMORY_SEMANTIC_EMBEDDING_QUERY_INPUT_TYPE` | Auto for known LiteLLM models | Optional LiteLLM `input_type` for search queries. |
 | `semantic_vector_k` | `BASIC_MEMORY_SEMANTIC_VECTOR_K` | `100` | Candidate count for vector nearest-neighbour retrieval. Higher values improve recall at the cost of latency. |
 
 ## Embedding Providers
@@ -135,7 +138,114 @@ export BASIC_MEMORY_SEMANTIC_EMBEDDING_PROVIDER=openai
 export OPENAI_API_KEY=sk-...
 ```
 
-When switching from FastEmbed to OpenAI (or vice versa), you must rebuild embeddings since the vector dimensions differ:
+### LiteLLM
+
+> **Experimental — advanced users only.** The LiteLLM provider is experimental and aimed at users comfortable operating remote embedding backends: paid API calls, per-model dimension and input-role configuration, and slower reindexing of large corpora. For most users, FastEmbed (local, default) is recommended. See [LiteLLM Provider](litellm-provider.md) for the caveats and tuning.
+
+Uses the LiteLLM SDK to call embedding models from providers such as OpenAI, Cohere, Azure, Bedrock, NVIDIA NIM, and other LiteLLM-supported backends. Requires the provider's API credentials.
+For the full option reference, provider setup examples, and live validation harness, see [LiteLLM Provider](litellm-provider.md).
+
+```bash
+export BASIC_MEMORY_SEMANTIC_SEARCH_ENABLED=true
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_PROVIDER=litellm
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_MODEL=cohere/embed-english-v3.0
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_DIMENSIONS=1024
+export COHERE_API_KEY=...
+```
+
+Basic Memory creates vector tables before the first embedding call, so non-default LiteLLM models must set `BASIC_MEMORY_SEMANTIC_EMBEDDING_DIMENSIONS`. The LiteLLM OpenAI default (`openai/text-embedding-3-small`) uses 1536 dimensions automatically.
+
+For fixed-size LiteLLM models, dimensions are used as Basic Memory's local vector schema and
+validation size. Basic Memory automatically sends dimensions as a provider-side output-size
+request for `text-embedding-3` model strings, where LiteLLM/OpenAI support reduced output
+dimensions. If an Azure/OpenAI deployment uses an arbitrary LiteLLM model string such as
+`azure/<deployment-name>` and the underlying model supports reduced dimensions, set
+`BASIC_MEMORY_SEMANTIC_EMBEDDING_FORWARD_DIMENSIONS=true`.
+
+Some retrieval models are asymmetric: indexed passages and search queries must be embedded with different provider parameters. Basic Memory automatically sets LiteLLM `input_type` for known asymmetric model families:
+
+- Cohere v3: documents use `search_document`, queries use `search_query`
+- NVIDIA NIM retrieval models: documents use `passage`, queries use `query`
+
+For other asymmetric LiteLLM models, set the input types explicitly:
+
+```bash
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_DOCUMENT_INPUT_TYPE=passage
+export BASIC_MEMORY_SEMANTIC_EMBEDDING_QUERY_INPUT_TYPE=query
+```
+
+#### Live LiteLLM Validation
+
+Provider APIs differ in subtle ways: some accept `dimensions`, some require separate
+document/query roles, and some route through deployment aliases that do not reveal the
+underlying model name. Before adding or changing LiteLLM model support, run the opt-in live
+evaluation harness:
+
+```bash
+export OPENAI_API_KEY=sk-...
+export COHERE_API_KEY=...
+just test-litellm-live
+```
+
+The built-in live cases cover:
+
+| Case | Required key | What it validates |
+|---|---|---|
+| `openai/text-embedding-3-small` | `OPENAI_API_KEY` | Standard LiteLLM OpenAI embedding calls and normalized 1536-dimensional output. |
+| `cohere/embed-english-v3.0` | `COHERE_API_KEY` | Cohere v3 asymmetric `search_document` / `search_query` handling and fixed 1024-dimensional output. |
+
+The harness embeds two documents and one query, checks vector dimensions and normalization,
+then verifies the authentication query ranks the authentication document above the distractor.
+It prints a table with per-model scores, norms, latency, role settings, and dimension-forwarding
+mode.
+
+To validate provider aliases or additional LiteLLM backends, save custom JSON cases:
+
+```bash
+export AZURE_API_KEY=...
+export AZURE_API_BASE=https://example.openai.azure.com
+export AZURE_API_VERSION=2024-02-01
+
+cat > /tmp/litellm-azure-cases.json <<'JSON'
+[
+  {
+    "name": "azure-text-embedding-3-small-512",
+    "model": "azure/<deployment-name>",
+    "dimensions": 512,
+    "api_key_env": "AZURE_API_KEY",
+    "forward_dimensions": true
+  }
+]
+JSON
+
+just test-litellm-live --cases-file /tmp/litellm-azure-cases.json
+```
+
+NVIDIA NIM retrieval models can be checked the same way:
+
+```bash
+export NVIDIA_NIM_API_KEY=...
+
+cat > /tmp/litellm-nvidia-cases.json <<'JSON'
+[
+  {
+    "name": "nvidia-embed-qa-4",
+    "model": "nvidia_nim/nvidia/embed-qa-4",
+    "dimensions": 1024,
+    "api_key_env": "NVIDIA_NIM_API_KEY",
+    "document_input_type": "passage",
+    "query_input_type": "query"
+  }
+]
+JSON
+
+just test-litellm-live --cases-file /tmp/litellm-nvidia-cases.json
+```
+
+For repeatable local runs, put the same JSON array in a file and pass
+`just test-litellm-live --cases-file path/to/litellm-cases.json`.
+
+When switching providers, models, dimensions, or LiteLLM document/query input types, rebuild embeddings:
 
 ```bash
 bm reindex --embeddings
@@ -203,9 +313,10 @@ bm reindex -p my-project
 
 - **Upgrade note**: Migration now performs a one-time automatic embedding backfill on upgrade.
 - **Manual enable case**: If you explicitly had `semantic_search_enabled=false` and then turn it on
-- **Provider change**: After switching between `fastembed` and `openai`
+- **Provider change**: After switching between `fastembed`, `openai`, and `litellm`
 - **Model change**: After changing `semantic_embedding_model`
 - **Dimension change**: After changing `semantic_embedding_dimensions`
+- **LiteLLM role change**: After changing `semantic_embedding_document_input_type` or `semantic_embedding_query_input_type`
 
 The reindex command shows progress with embedded/skipped/error counts:
 
@@ -263,6 +374,7 @@ The sqlite-vec extension is loaded per-connection. Vector tables are created laz
 ### Postgres (cloud)
 
 - **Vector storage**: [pgvector](https://github.com/pgvector/pgvector) with HNSW indexing
+- **Local Docker**: use `docker-compose-postgres.yml` (`pgvector/pgvector:pg17`). Plain `postgres:17` lacks the extension; run `CREATE EXTENSION IF NOT EXISTS vector;` on any external instance before first migration.
 - **Chunk metadata table**: Created via Alembic migration (`search_vector_chunks` with `BIGSERIAL` primary key)
 - **Embedding table**: `search_vector_embeddings` created at runtime (dimension-dependent, same pattern as SQLite)
 - **Index**: HNSW index on the embedding column for fast approximate nearest-neighbour queries

docs/specs/SPEC-LOCAL-GRAPH-INTELLIGENCE-IMPLEMENTATION-PLAN.md

@@ -1,594 +0,0 @@
-# SPEC-LOCAL-GRAPH-INTELLIGENCE-IMPLEMENTATION-PLAN
-
-**Status:** Draft (Decision-Complete)  
-**Date:** 2026-03-05  
-**Owner:** Basic Memory Engineering  
-**Implementation Status (2026-03-05):** Phase 1 contract skeleton implemented in `basic-memory` branch `codex/graph-intelligence-phase1`.  
-**Related Specs:**
-1. `/docs/specs/SPEC-LOCAL-GRAPH-INTELLIGENCE-MASTER.md`
-2. `/docs/specs/SPEC-LOCAL-GRAPH-INTELLIGENCE-TECHNICAL-ADDENDUM.md`
-3. `/docs/specs/SPEC-LOCAL-GRAPH-INTELLIGENCE.md`
-
-## Scope and Intent
-
-This document is the execution handoff for Local+ Graph Intelligence.
-
-It defines exactly how we will deliver graph and FCM capabilities inside the existing Basic Memory architecture:
-1. FastAPI-first business logic.
-2. MCP and CLI as thin facades.
-3. Local/cloud contract parity.
-4. Tight build-test-iterate loop for fast delivery.
-
-This file is intentionally implementation-oriented and does not duplicate pricing narrative from the master spec.
-
-## Architecture Alignment (FastAPI-first, MCP/CLI facade)
-
-Locked architecture alignment for implementation:
-1. MCP tools remain thin proxy facades.
-2. CLI `bm tool` commands call MCP tools in JSON mode.
-3. Core logic lives in FastAPI routers and services.
-4. Cloud and local share the same REST contracts.
-5. Per-project routing continues through existing project client patterns.
-
-Execution mapping:
-1. API routers define public contracts in `/graph` and `/fcm` domains.
-2. Services own traversal, scoring, simulation, and fallback logic.
-3. Repositories and index providers own data access and graph index operations.
-4. MCP typed clients call REST endpoints and return JSON-first tool output.
-5. CLI passthrough executes tool calls and prints machine-friendly JSON.
-
-## Locked Decisions
-
-1. SQLite remains operational source for entities, relations, embeddings, and project state.
-2. Markdown remains source of truth.
-3. Oxigraph/pyoxigraph is the derived graph index for deep traversal.
-4. FCM simulation runs in Python service layer; it is not delegated to graph DB query engines.
-5. Graph index is rebuildable and disposable; stale index never blocks user workflows.
-6. FCM model and scenario artifacts persist in app database.
-7. Local+ features are gated by config flags first; entitlement wiring follows later.
-8. Graph-first vertical slices ship before deep FCM expansion.
-9. Atomic tools ship first; orchestration workflows are deferred.
-10. Existing `build_context` and `search_notes` remain backward compatible with no breaking change.
-
-## Progress Snapshot (as of 2026-03-05)
-
-Completed in Phase 1:
-1. Added `/graph` and `/fcm` v2 routers with all required contract endpoints.
-2. Added graph/FCM request and response schemas for all public API contracts.
-3. Added service-layer implementations for graph and FCM contract endpoints.
-4. Added typed MCP clients for graph and FCM API calls.
-5. Added MCP tools: `graph_lineage`, `graph_impact`, `graph_health`, `graph_reindex`, `fcm_simulate`, `fcm_rank_actions`, `fcm_import_model`, `fcm_export_model`.
-6. Added CLI passthrough commands under `bm tool ...` for all planned graph/FCM operations.
-7. Added scheduler task names for graph lifecycle: `sync_graph_entity`, `sync_graph_project`, `reindex_graph_project`.
-8. Added focused tests for API, MCP clients/tools, and CLI graph/FCM passthrough.
-9. Added fast-loop `just` targets: `test-graph-intel-api`, `test-graph-intel-mcp`, `test-graph-intel-cli`, `test-graph-intel`.
-
-Validation completed:
-1. `just test-graph-intel` passes.
-2. `ruff check` passes on changed files.
-3. `pyright` passes on changed files.
-
-Still pending after Phase 1:
-1. SQL-backed traversal/scoring for graph `lineage`, `impact`, and `health`.
-2. Oxigraph provider integration and stale-index catch-up flow.
-3. Persistent FCM model/scenario state and interop round-trip guarantees.
-4. Config-flag and entitlement gating at API/tool boundaries.
-5. Performance instrumentation and p95 envelope enforcement.
-
-## Delivery Phases
-
-### Phase 1: Contract skeleton
-
-Status: Completed (2026-03-05)
-
-Deliverables:
-1. Add `/graph` and `/fcm` API routers with request/response schemas.
-2. Add typed MCP clients for graph and FCM endpoints.
-3. Add MCP tool passthrough commands for all new operations.
-4. Add CLI `bm tool` passthrough commands mirroring MCP surface.
-5. Add minimal smoke tests for route reachability and schema validation.
-
-Exit criteria:
-1. All endpoints return structured success and error envelopes.
-2. MCP/CLI paths execute end-to-end with stubbed service responses.
-
-### Phase 2: Graph capabilities on SQL-backed logic
-
-Status: Next active phase
-
-Deliverables:
-1. Implement `lineage`, `impact`, and `health` in service layer using SQL-backed traversal and scoring.
-2. Add provenance/evidence linking in graph outputs.
-3. Add deterministic graph-health calculations for fixed snapshots.
-
-Exit criteria:
-1. `graph_lineage`, `graph_impact`, and `graph_health` pass contract tests.
-2. SQL fallback behavior is explicit and covered by tests.
-
-### Phase 3: Oxigraph derived index provider
-
-Status: Planned
-
-Deliverables:
-1. Introduce Oxigraph provider behind graph-query interface.
-2. Add lazy catch-up jobs and project-wide reindex operation.
-3. Preserve SQL fallback when index is missing or stale.
-
-Exit criteria:
-1. Stale index path serves results via SQL and schedules catch-up.
-2. Index rebuild can be triggered and completed without data loss.
-
-### Phase 4: FCM import/simulate/rank/export
-
-Status: Planned (contract endpoints complete, full behavior pending)
-
-Deliverables:
-1. Implement CSV-first import/export contracts.
-2. Implement deterministic simulation core with convergence metadata.
-3. Implement action ranking with evidence references and confidence output.
-4. Persist scenario inputs and result artifacts.
-
-Exit criteria:
-1. Research flow scenario passes: import -> simulate -> rank -> export.
-2. Interop round-trip preserves node/edge counts and signed weights.
-
-### Phase 5: Hardening
-
-Status: Planned
-
-Deliverables:
-1. Performance tuning against published latency envelopes.
-2. Local/cloud parity tests for semantics and error behavior.
-3. MCP prompt/docs updates for new graph and FCM tools.
-4. Operational docs for reindex, fallback, and troubleshooting.
-
-Exit criteria:
-1. `just check` passes before merge.
-2. Acceptance criteria in this document are fully met.
-
-## API and Interface Additions
-
-### Shared API conventions
-
-1. All endpoints are project-scoped under `/v2/projects/{project_id}`.
-2. Request and response bodies are JSON-first and agent-friendly.
-3. Success envelope is endpoint-specific payload with deterministic fields and optional probabilistic fields.
-4. Error envelope:
-```json
-{
-  "error": {
-    "code": "INVALID_ARGUMENT|NOT_FOUND|INDEX_NOT_READY|MODEL_INVALID|RESOURCE_LIMIT_EXCEEDED|INTERNAL_ERROR",
-    "message": "string",
-    "details": {}
-  }
-}
-```
-5. Latency and scale targets are p95 targets for local default hardware profile.
-
-### 1) `POST /v2/projects/{project_id}/graph/lineage`
-
-Purpose: explain decision lineage and supporting evidence paths.
-
-Request schema:
-```json
-{
-  "start": "string",
-  "goal": "string|null",
-  "max_hops": 4,
-  "relation_filters": ["string"]
-}
-```
-
-Response schema:
-```json
-{
-  "root": {"id": "string", "title": "string", "permalink": "string"},
-  "paths": [
-    {
-      "path_id": "string",
-      "nodes": [{"id": "string", "title": "string"}],
-      "edges": [{"relation": "string", "direction": "outgoing|incoming"}],
-      "deterministic_path_score": 0.0,
-      "confidence": 0.0,
-      "evidence_refs": ["memory://..."]
-    }
-  ],
-  "generated_at": "RFC3339"
-}
-```
-
-Deterministic fields: `root`, `paths.nodes`, `paths.edges`, `deterministic_path_score`, `generated_at`.  
-Probabilistic fields: `confidence`.  
-Latency target: p95 <= 450ms with `max_hops<=4`.  
-Scale envelope: up to 50k nodes and 300k edges.
-
-### 2) `POST /v2/projects/{project_id}/graph/impact`
-
-Purpose: preview impact radius before edits or decisions.
-
-Request schema:
-```json
-{
-  "target": "string",
-  "horizon": 2,
-  "relation_filters": ["string"],
-  "include_reasons": true
-}
-```
-
-Response schema:
-```json
-{
-  "target": {"id": "string", "title": "string"},
-  "affected": [
-    {
-      "id": "string",
-      "title": "string",
-      "distance": 1,
-      "impact_score": 0.0,
-      "confidence": 0.0,
-      "reasons": ["string"],
-      "evidence_refs": ["memory://..."]
-    }
-  ],
-  "summary": {"total_considered": 0, "total_returned": 0}
-}
-```
-
-Deterministic fields: membership, distance, summary counts.  
-Probabilistic fields: `impact_score`, `confidence`.  
-Latency target: p95 <= 650ms for `horizon<=3`.  
-Scale envelope: default 200 results, hard cap 1000 with pagination token.
-
-### 3) `GET /v2/projects/{project_id}/graph/health`
-
-Purpose: report deterministic graph quality and actionable issues.
-
-Query params:
-1. `scope` optional directory prefix.
-2. `timeframe` optional window like `30d`.
-
-Response schema:
-```json
-{
-  "metrics": {
-    "orphan_rate": 0.0,
-    "stale_central_nodes": 0,
-    "overloaded_hubs": 0,
-    "contradiction_candidates": 0
-  },
-  "issues": [
-    {
-      "issue_type": "orphan|stale_central|overloaded_hub|contradiction_candidate",
-      "entity_id": "string",
-      "severity": "low|medium|high",
-      "reason": "string",
-      "suggested_action": "string",
-      "confidence": 0.0
-    }
-  ],
-  "computed_at": "RFC3339"
-}
-```
-
-Deterministic fields: `metrics`, issue membership for fixed snapshot.  
-Probabilistic fields: contradiction confidence when applicable.  
-Latency target: p95 <= 1500ms project-wide, <= 700ms scoped.
-
-### 4) `POST /v2/projects/{project_id}/graph/reindex`
-
-Purpose: force project-wide graph index rebuild.
-
-Request schema:
-```json
-{
-  "mode": "full|incremental",
-  "reason": "string|null"
-}
-```
-
-Response schema:
-```json
-{
-  "job_id": "string",
-  "status": "queued|running|completed|failed",
-  "scheduled_at": "RFC3339"
-}
-```
-
-Deterministic fields: job metadata and status transitions.  
-Probabilistic fields: none.  
-Latency target: enqueue response p95 <= 120ms.
-
-### 5) `POST /v2/projects/{project_id}/fcm/simulate`
-
-Purpose: run FCM scenario simulation.
-
-Request schema:
-```json
-{
-  "actions": [{"node_id": "string", "delta": 0.2}],
-  "scenario": {
-    "steps": 12,
-    "activation": "tanh|sigmoid|bounded_linear",
-    "decay": 0.05
-  },
-  "clamp_rules": [{"node_id": "string", "min": -1.0, "max": 1.0}]
-}
-```
-
-Response schema:
-```json
-{
-  "baseline": [{"node_id": "string", "state": 0.0}],
-  "projected": [{"node_id": "string", "state": 0.0}],
-  "deltas": [{"node_id": "string", "delta": 0.0}],
-  "stability": {"converged": true, "iterations_used": 0, "residual": 0.0},
-  "confidence": 0.0,
-  "explanations": [{"node_id": "string", "top_influencers": [{"source": "string", "weight": 0.0}]}],
-  "evidence_refs": ["memory://..."]
-}
-```
-
-Deterministic fields: baseline, projected, deltas, stability for fixed model and params.  
-Probabilistic fields: confidence.  
-Latency target: p95 <= 1000ms for <=500 nodes and <=5000 edges.
-
-### 6) `POST /v2/projects/{project_id}/fcm/rank-actions`
-
-Purpose: rank candidate interventions by expected outcome and risk.
-
-Request schema:
-```json
-{
-  "goal": "string",
-  "constraints": {
-    "max_negative_impact": 0.25,
-    "required_tags": ["string"],
-    "disallowed_nodes": ["string"]
-  },
-  "top_k": 10
-}
-```
-
-Response schema:
-```json
-{
-  "goal": {"node_id": "string", "label": "string"},
-  "recommendations": [
-    {
-      "action_node_id": "string",
-      "expected_goal_delta": 0.0,
-      "risk_penalty": 0.0,
-      "net_score": 0.0,
-      "confidence": 0.0,
-      "rationale": ["string"],
-      "evidence_refs": ["memory://..."]
-    }
-  ]
-}
-```
-
-Deterministic fields: candidate set and constraint compliance.  
-Probabilistic fields: expected delta, penalty, net score, confidence.  
-Latency target: p95 <= 1500ms for top-10 from <=100 candidates.
-
-### 7) `POST /v2/projects/{project_id}/fcm/import`
-
-Purpose: import FCM model from CSV-first contract.
-
-Request schema:
-```json
-{
-  "source": "string",
-  "format": "csv_bundle_v1",
-  "merge_mode": "replace|upsert"
-}
-```
-
-Response schema:
-```json
-{
-  "import_id": "string",
-  "nodes_loaded": 0,
-  "edges_loaded": 0,
-  "warnings": ["string"],
-  "errors": ["string"]
-}
-```
-
-Deterministic fields: counts and validation diagnostics.  
-Probabilistic fields: none.  
-Latency target: p95 <= 2500ms for 10k edges import.
-
-### 8) `POST /v2/projects/{project_id}/fcm/export`
-
-Purpose: export FCM model for interoperability.
-
-Request schema:
-```json
-{
-  "format": "csv_bundle_v1",
-  "selection": {
-    "scope": "all|tag|subgraph",
-    "tag": "string|null",
-    "seed_nodes": ["string"]
-  }
-}
-```
-
-Response schema:
-```json
-{
-  "export_id": "string",
-  "format": "csv_bundle_v1",
-  "files": [{"name": "nodes.csv", "path": "string"}, {"name": "edges.csv", "path": "string"}],
-  "node_count": 0,
-  "edge_count": 0
-}
-```
-
-Deterministic fields: file names and counts for fixed selection.  
-Probabilistic fields: none.  
-Latency target: p95 <= 1800ms for 50k edges export.
-
-## Data Model and Storage Boundaries
-
-1. SQLite is mandatory operational source for entities, relations, embeddings, and project metadata.
-2. Oxigraph stores derived knowledge graph index only.
-3. FCM state persists in app database with scenario artifacts and run history.
-4. Graph index is rebuildable and disposable by design.
-5. Markdown files remain canonical source of truth.
-
-Implementation data boundaries:
-1. Knowledge graph schema tracks descriptive nodes and typed edges plus provenance.
-2. FCM schema tracks signed weighted causal edges and node states.
-3. Provenance model requires `evidence_refs`, `confidence`, and `updated_at`.
-4. Scenario model stores interventions, constraints, run parameters, and output deltas.
-5. Interop schema starts with CSV-first Mental Modeler contract.
-
-## Background Jobs and Index Lifecycle
-
-Scheduler tasks to add:
-1. `sync_graph_entity`
-2. `sync_graph_project`
-3. `reindex_graph_project`
-
-Lifecycle rules:
-1. Note writes, edits, moves, and deletes schedule graph-index sync tasks.
-2. Scheduling pattern mirrors existing vector sync behavior.
-3. On stale or missing graph index, request path serves via SQL fallback and schedules catch-up.
-4. Reindex is idempotent and safe to rerun.
-5. Index version metadata is tracked per project for staleness checks.
-
-Operational behaviors:
-1. Foreground requests never block on full reindex completion.
-2. Background job failures surface in health endpoints with actionable status.
-3. Reindex job can run incremental or full mode.
-4. Phase 1 note: scheduler task names and reindex enqueue path are implemented; write/edit/move/delete sync hooks still need explicit wiring.
-
-## MCP and CLI Surface
-
-New MCP tools:
-1. `graph_lineage`
-2. `graph_impact`
-3. `graph_health`
-4. `fcm_simulate`
-5. `fcm_rank_actions`
-6. `fcm_import_model`
-7. `fcm_export_model`
-
-CLI passthrough additions:
-1. `bm tool graph-lineage ...`
-2. `bm tool graph-impact ...`
-3. `bm tool graph-health ...`
-4. `bm tool fcm-simulate ...`
-5. `bm tool fcm-rank-actions ...`
-6. `bm tool fcm-import-model ...`
-7. `bm tool fcm-export-model ...`
-
-Output conventions:
-1. Default output is JSON for MCP and CLI.
-2. MCP supports optional `output_format="text"` for human-readable summaries.
-3. CLI remains JSON-first to keep agent integration deterministic.
-
-## Test Strategy (fast loop + gates)
-
-### Slice-by-slice loop
-
-For each vertical slice, implement in this order:
-1. API contract and schema.
-2. Typed MCP client.
-3. MCP tool passthrough.
-4. CLI passthrough.
-5. Focused tests for API/MCP/CLI.
-
-Fast checks per slice:
-1. Targeted `pytest` for changed API, MCP, and CLI modules.
-2. `just fast-check`.
-3. `just doctor`.
-4. `just test-graph-intel` for graph/FCM-only iteration loop.
-
-Milestone gates:
-1. SQLite unit and integration pass first.
-2. Selective Postgres parity tests for new graph and FCM contracts.
-3. Full `just check` before merge.
-
-### Required test cases and scenarios
-
-1. Casual user impact preview before note edit.
-2. Decision audit: lineage plus evidence references explain recommendation.
-3. Graph health deterministic output for fixed snapshot.
-4. Research flow: import model -> simulate -> rank -> export.
-5. Sparse and contradictory graph input degrades gracefully.
-6. Interop round-trip preserves node/edge counts and signed weights.
-7. Local and cloud parity on contract semantics and error model.
-8. Stale index fallback path returns valid response and schedules catch-up.
-
-## Rollout and Feature Flagging
-
-Rollout controls:
-1. Gate graph and FCM endpoints behind config flags first.
-2. Add entitlement enforcement after behavior and reliability stabilize.
-3. Keep existing tools and endpoints fully backward compatible.
-
-Suggested flags:
-1. `feature_graph_intelligence_enabled`
-2. `feature_fcm_enabled`
-3. `feature_graph_oxigraph_provider_enabled`
-4. `feature_graph_sql_fallback_enabled`
-
-Rollout sequence:
-1. Enable contract skeleton in dev.
-2. Enable graph features for internal alpha users.
-3. Enable Oxigraph provider with fallback-on by default.
-4. Enable FCM import/simulate/rank/export for research alpha users.
-5. Promote to Local+ beta when acceptance criteria are met.
-
-## Risks and Mitigations
-
-1. Risk: graph query complexity increases p95 latency.  
-Mitigation: strict query caps, fallback path, and performance budgets per endpoint.
-2. Risk: stale index produces confusing outputs.  
-Mitigation: explicit staleness checks, SQL fallback, and background catch-up scheduling.
-3. Risk: FCM recommendations appear opaque.  
-Mitigation: require evidence references, confidence fields, and deterministic simulation metadata.
-4. Risk: local/cloud contract drift.  
-Mitigation: shared schemas, contract tests, and parity checks in CI gates.
-5. Risk: integration surface grows faster than team can validate.  
-Mitigation: phase gates and vertical-slice completion before opening next phase.
-
-## Improvement Backlog (Post-Phase 1)
-
-1. Refactor `bm tool` graph/FCM commands into a dedicated CLI module to reduce `tool.py` size and improve maintainability.
-2. Consolidate repeated MCP text-formatting helpers for graph/FCM outputs.
-3. Replace deterministic placeholder graph behavior with SQL-backed lineage/impact/health implementations.
-4. Add explicit config/entitlement enforcement for graph/FCM endpoints and tools.
-5. Add performance telemetry and p95 reporting for graph and FCM routes.
-6. Add parity and degradation tests for stale-index fallback and contradictory/sparse inputs.
-
-## Acceptance Criteria
-
-1. All required sections in this document are complete with no unresolved decisions.
-2. API/interface contracts are implementation-ready with request, response, error, latency, and scale details.
-3. Architecture alignment is explicit: FastAPI logic core, MCP/CLI facades, shared local/cloud contracts.
-4. Delivery phases define concrete outputs and exit criteria.
-5. Test strategy includes tight iteration loop and milestone gates.
-6. Required scenario matrix is covered in test plan and mapped to implementation phases.
-7. Rollout plan includes feature flags and backward compatibility guarantees.
-8. An implementer can execute this plan without additional architecture clarification.
-
-## Assumptions and Defaults
-
-1. Config-flag gating first; entitlement wiring later.
-2. Graph-first vertical slices before deep FCM expansion.
-3. Atomic tools first; orchestration layer deferred.
-4. JSON-first contracts for agent usability.
-5. No breaking changes to existing `build_context` and `search_notes`.
-
-## Out of Scope
-
-1. Implementation details unrelated to graph/FCM delivery phases in this document.
-2. Migration execution.
-3. Pricing and positioning rewrites.
-4. Cloud infrastructure changes in this phase.

docs/specs/SPEC-LOCAL-GRAPH-INTELLIGENCE-MASTER.md

@@ -1,782 +0,0 @@
-# SPEC-LOCAL-GRAPH-INTELLIGENCE-MASTER: Local+ Graph Intelligence Blueprint
-
-**Status:** Draft (Iteration 2, Decision-Complete)  
-**Date:** 2026-03-05  
-**Owner:** Basic Memory  
-**Primary Audience:** Internal build team (Product, Engineering, GTM)  
-**Current Phase (2026-03-05):** Implementation Plan Phase 1 is complete; Phase 2 (SQL-backed graph logic) is the active engineering phase.  
-**Related Specs:**
-1. `/docs/specs/SPEC-LOCAL-GRAPH-INTELLIGENCE.md`
-2. `/docs/specs/SPEC-LOCAL-GRAPH-INTELLIGENCE-TECHNICAL-ADDENDUM.md`
-3. `/docs/specs/SPEC-LOCAL-GRAPH-INTELLIGENCE-IMPLEMENTATION-PLAN.md`
-
-Reading guide:
-1. Sections 1-5 define the business and product decisions.
-2. Sections 6-10 define architecture and interface contracts.
-3. Sections 11-14 define pricing, rollout, and decision gates for execution.
-
-## 1) Executive Thesis
-
-Basic Memory will ship **Local+ Graph Intelligence** as a premium local capability that upgrades the product from retrieval to decision support.
-
-Positioning statement:
-1. "Keep your local workflow. Add decision intelligence as complexity grows."
-2. The product sells safer decisions and explainable recommendations, not graph database mechanics.
-
-Locked thesis decisions:
-1. SQLite will remain the operational core.
-2. Markdown will remain source of truth.
-3. Graph and FCM indexes will be derived and rebuildable.
-4. Oxigraph/pyoxigraph will be the v1 graph index path.
-5. FCM simulation will run in a Python service layer.
-6. SurrealDB and FalkorDB will not be core dependencies in v1 due license-roadmap mismatch.
-7. Product messaging will sell outcomes (safer decisions, explainable recommendations), not database internals.
-
-## 2) Problem and Opportunity
-
-Current state after v0.19:
-1. Recursive SQL traversal can retrieve connected notes but becomes expensive and noisy after a few hops.
-2. Users still do manual synthesis for impact analysis, decision lineage, and contradiction resolution.
-3. Researchers need causal reasoning and scenario modeling, not only graph navigation.
-
-Opportunity:
-1. Deliver a premium local tier that materially improves decision quality while keeping data local.
-2. Create a bridge from knowledge graph navigation to causal simulation (FCM).
-3. Open a research-heavy market segment that values explainability and model interoperability.
-
-Business opportunity:
-1. Add a middle tier between free OSS and cloud subscription.
-2. Preserve an upgrade path to hosted collaboration for research teams later.
-3. Differentiate Basic Memory for research-grade workflows without forcing cloud adoption.
-
-## 3) User Segments and Jobs-to-be-Done
-
-| Segment | Primary Job-to-be-Done | Pain Today | Value Trigger |
-|---|---|---|---|
-| Casual local builder | Avoid breaking related notes when editing | Hidden dependencies and rework | Impact preview before edits |
-| Solo technical founder | Keep architecture and decision context coherent | Context overload and drift | Decision lineage + impact radius |
-| Research user | Model and test intervention strategies | No integrated causal simulation with notes | FCM simulation + action ranking |
-| Product/research lead | Synthesize evidence quickly across many docs | Fragmented understanding | Path exploration + priority briefs |
-
-## 4) Product Outcomes (not feature list)
-
-Local+ Graph Intelligence will optimize for these outcomes:
-1. **Change Safety:** users catch downstream impacts before they edit.
-2. **Decision Clarity:** users can explain why an answer or recommendation was produced.
-3. **Knowledge Health:** users keep larger graphs coherent with less manual audit work.
-4. **Research Leverage:** users run scenario-level reasoning tied to explicit evidence.
-
-Outcome metrics (for 30-day retained Local+ cohorts):
-1. Median time-to-understanding for complex topics decreases by at least 35% for active Local+ users.
-2. User-reported surprise side effects after note edits decrease by at least 30%.
-3. At least 60% of active Local+ users invoke graph intelligence features weekly.
-4. At least 40% of research-profile Local+ users invoke one FCM workflow weekly.
-
-## 5) Feature Set v1/v1.5/v2
-
-### v1 (post-v0.19 launch scope)
-
-Included:
-1. Decision Lineage
-2. Impact Radius
-3. Path Explorer (guided)
-4. Graph Health (orphans, stale-central nodes, overloaded hubs)
-5. CSV FCM import/export (nodes and edges)
-6. FCM simulation for explicit action scenarios
-7. FCM action ranking with evidence-linked rationale
-
-Excluded:
-1. Native Mental Modeler project format write support
-2. Team governance and shared model policy controls
-3. Cloud-only enhancements
-
-### v1.5
-
-Included:
-1. Contradiction Watch with reconciliation queue
-2. Priority Briefs (graph + FCM leverage summary)
-3. Stronger uncertainty propagation in FCM scoring
-4. Cloud execution optionality for heavy simulation jobs
-
-### v2
-
-Included:
-1. Team-shared model governance
-2. Hosted collaboration features for research teams
-3. Optional native model translators beyond CSV baseline
-
-Cut line policy:
-1. If a capability cannot meet explainability requirements, it moves to v1.5+.
-2. If a capability requires cloud to function, it cannot be marked v1.
-3. If a capability cannot meet local performance envelopes, it cannot be promoted into default workflows.
-
-## 6) Technical Architecture (Two-Graph Model)
-
-### High-level architecture
-
-```mermaid
-flowchart LR
-    A[Markdown Files Source of Truth] --> B[Parser + Sync Pipeline]
-    B --> C[SQLite Operational Store]
-    B --> D[Derived Knowledge Graph Index Oxigraph]
-    C --> E[Graph Intelligence Service]
-    D --> E
-    E --> F[Lineage Impact Path Health APIs]
-    C --> G[FCM Service Python]
-    D --> G
-    G --> H[Simulation Ranking Interop APIs]
-```
-
-### Two-graph model
-
-1. **Knowledge Graph (descriptive):** notes, decisions, concepts, and typed relations.
-2. **FCM Graph (causal):** signed weighted influence links between goals, drivers, risks, and interventions.
-
-### Core architectural decisions
-
-1. SQLite is authoritative for entities, observations, relations, metadata, embeddings, and project state.
-2. Oxigraph is a derived index for multi-hop graph traversal and graph-pattern retrieval.
-3. FCM calculations run in Python using explicit model state and deterministic numerical steps.
-4. Local mode runs fully offline.
-5. Cloud mode can execute the same contracts via adjunct services while Neon remains system of record.
-
-## 7) Backend Decision and Trade-Offs
-
-### Final recommendation
-
-Use this stack for v1:
-1. SQLite (existing): primary operational store.
-2. Oxigraph/pyoxigraph: derived knowledge graph index.
-3. Python FCM service: causal simulation and ranking.
-
-Decision rationale:
-1. This preserves local-first UX while enabling deeper traversal and causal simulation.
-2. This avoids restrictive licensing dependencies in the core product path.
-3. This keeps a clean cloud portability path where Neon remains the hosted system of record.
-
-### Trade-off matrix
-
-| Option | Strengths | Risks | Decision |
-|---|---|---|---|
-| SQLite + Oxigraph + Python FCM | Local-first, permissive licensing, clear service boundaries, cloud-portable | Requires translation layer for query ergonomics | **Adopt v1** |
-| Apache AGE on Postgres | SQL+graph in one engine, good cloud-side graph semantics | Neon extension support uncertainty, weaker local/cloud parity with SQLite local baseline | Defer |
-| SurrealDB | Strong integrated multi-model experience | BSL posture conflicts with future hosted/open strategy timing | Reject for v1 core |
-| FalkorDB | Graph performance and Redis ecosystem familiarity | SSPL posture conflicts with hosted/open strategy | Reject for v1 core |
-
-## 8) Public APIs / Interfaces
-
-All APIs are proposed MCP tool contracts for Local+ mode.
-
-### Common conventions
-
-1. `project` parameter is optional and follows existing Basic Memory project routing.
-2. Deterministic fields are reproducible with identical inputs and index state.
-3. Probabilistic fields are model-derived scores and include confidence metadata.
-4. Error model uses structured codes and fail-fast behavior.
-
-Shared error codes:
-1. `INVALID_ARGUMENT`
-2. `NOT_FOUND`
-3. `MODEL_INVALID`
-4. `INDEX_NOT_READY`
-5. `RESOURCE_LIMIT_EXCEEDED`
-6. `INTERNAL_ERROR`
-
----
-
-### 8.1 `graph_lineage(start, goal?)`
-
-**Input schema:**
-```json
-{
-  "start": "string (required, permalink or memory URL)",
-  "goal": "string (optional, concept or decision target)",
-  "max_hops": "integer (optional, default 4, range 1-6)",
-  "relation_filters": ["string"],
-  "project": "string (optional)"
-}
-```
-
-**Output schema:**
-```json
-{
-  "root": {"id": "string", "title": "string", "permalink": "string"},
-  "paths": [
-    {
-      "path_id": "string",
-      "nodes": [{"id": "string", "title": "string"}],
-      "edges": [{"relation": "string", "direction": "outgoing|incoming"}],
-      "deterministic_path_score": 0.0,
-      "confidence": 0.0,
-      "evidence_refs": ["memory://..."]
-    }
-  ],
-  "generated_at": "RFC3339"
-}
-```
-
-**Deterministic fields:** root, nodes, edges, deterministic path score, generated timestamp.  
-**Probabilistic fields:** confidence.
-
-**Latency target:** p95 <= 450ms for `max_hops<=4`, graph envelope up to 50k nodes / 300k edges.
-
-**Scale envelope:**
-1. Tested local baseline: 50k nodes, 300k edges.
-2. Expected degradation: path expansion can exceed latency target when candidate paths > 20k.
-
----
-
-### 8.2 `graph_impact(target, horizon, relation_filters?)`
-
-**Input schema:**
-```json
-{
-  "target": "string (required)",
-  "horizon": "integer (required, range 1-4)",
-  "relation_filters": ["string"],
-  "include_reasons": "boolean (default true)",
-  "project": "string (optional)"
-}
-```
-
-**Output schema:**
-```json
-{
-  "target": {"id": "string", "title": "string"},
-  "affected": [
-    {
-      "id": "string",
-      "title": "string",
-      "distance": 2,
-      "impact_score": 0.0,
-      "confidence": 0.0,
-      "reasons": ["string"]
-    }
-  ],
-  "summary": {"total_considered": 0, "total_returned": 0}
-}
-```
-
-**Deterministic fields:** membership, distance, summary counts.  
-**Probabilistic fields:** impact score, confidence.
-
-**Latency target:** p95 <= 650ms for `horizon<=3` under baseline envelope.
-
-**Scale envelope:**
-1. `affected` default cap: 200 items.
-2. Hard cap: 1000 items with pagination token.
-
----
-
-### 8.3 `graph_health(scope?, timeframe?)`
-
-**Input schema:**
-```json
-{
-  "scope": "string (optional, directory prefix or project-wide)",
-  "timeframe": "string (optional, e.g. 30d, 90d)",
-  "project": "string (optional)"
-}
-```
-
-**Output schema:**
-```json
-{
-  "metrics": {
-    "orphan_rate": 0.0,
-    "stale_central_nodes": 0,
-    "overloaded_hubs": 0,
-    "contradiction_candidates": 0
-  },
-  "issues": [
-    {
-      "issue_type": "orphan|stale_central|overloaded_hub|contradiction_candidate",
-      "entity_id": "string",
-      "severity": "low|medium|high",
-      "reason": "string",
-      "suggested_action": "string"
-    }
-  ],
-  "computed_at": "RFC3339"
-}
-```
-
-**Deterministic fields:** metrics and issue list membership for a fixed graph snapshot.  
-**Probabilistic fields:** contradiction candidate confidence when present.
-
-**Latency target:** p95 <= 1500ms project-wide; <= 700ms for scoped directory mode.
-
-**Scale envelope:** project-wide scans tested to 50k nodes.
-
----
-
-### 8.4 `fcm_simulate(actions, scenario?, clamp_rules?)`
-
-**Input schema:**
-```json
-{
-  "actions": [
-    {"node_id": "string", "delta": 0.2}
-  ],
-  "scenario": {
-    "steps": 12,
-    "activation": "tanh|sigmoid|bounded_linear",
-    "decay": 0.05
-  },
-  "clamp_rules": [
-    {"node_id": "string", "min": -1.0, "max": 1.0}
-  ],
-  "project": "string (optional)"
-}
-```
-
-**Output schema:**
-```json
-{
-  "baseline": [{"node_id": "string", "state": 0.12}],
-  "projected": [{"node_id": "string", "state": 0.43}],
-  "deltas": [{"node_id": "string", "delta": 0.31}],
-  "stability": {
-    "converged": true,
-    "iterations_used": 9,
-    "residual": 0.002
-  },
-  "confidence": 0.0,
-  "explanations": [
-    {"node_id": "string", "top_influencers": [{"source": "string", "weight": 0.7}]}
-  ]
-}
-```
-
-**Deterministic fields:** baseline, projected, deltas, convergence metadata for fixed model and parameters.  
-**Probabilistic fields:** confidence (derived from edge confidence and evidence coverage).
-
-**Latency target:** p95 <= 1000ms for up to 500 nodes / 5000 edges and <=12 steps.
-
-**Scale envelope:**
-1. Soft limit: 2000 nodes / 20000 edges.
-2. Over soft limit: return `RESOURCE_LIMIT_EXCEEDED` with remediation guidance.
-
----
-
-### 8.5 `fcm_rank_actions(goal, constraints?, top_k?)`
-
-**Input schema:**
-```json
-{
-  "goal": "string (required node_id)",
-  "constraints": {
-    "max_negative_impact": 0.25,
-    "required_tags": ["string"],
-    "disallowed_nodes": ["string"]
-  },
-  "top_k": "integer (default 10, range 1-25)",
-  "project": "string (optional)"
-}
-```
-
-**Output schema:**
-```json
-{
-  "goal": {"node_id": "string", "label": "string"},
-  "recommendations": [
-    {
-      "action_node_id": "string",
-      "expected_goal_delta": 0.0,
-      "risk_penalty": 0.0,
-      "net_score": 0.0,
-      "confidence": 0.0,
-      "rationale": ["string"],
-      "evidence_refs": ["memory://..."]
-    }
-  ]
-}
-```
-
-**Deterministic fields:** candidate action set, constraints compliance.  
-**Probabilistic fields:** expected goal delta, risk penalty, net score, confidence.
-
-**Latency target:** p95 <= 1500ms for top 10 from up to 100 candidate actions.
-
-**Scale envelope:**
-1. Candidate actions hard cap: 1000.
-2. For larger sets, require pre-filtering via tags/scope.
-
----
-
-### 8.6 `fcm_import_model(source, format)`
-
-**Input schema:**
-```json
-{
-  "source": "string (required path or URI)",
-  "format": "csv_bundle_v1 (required)",
-  "merge_mode": "replace|upsert (default upsert)",
-  "project": "string (optional)"
-}
-```
-
-**Output schema:**
-```json
-{
-  "import_id": "string",
-  "nodes_loaded": 0,
-  "edges_loaded": 0,
-  "warnings": ["string"],
-  "errors": ["string"]
-}
-```
-
-**Deterministic fields:** load counts and validation results.  
-**Probabilistic fields:** none.
-
-**Latency target:** p95 <= 2500ms for 10k edges CSV bundle.
-
-**Scale envelope:**
-1. Maximum CSV rows per import: 250k.
-2. Above limit returns `RESOURCE_LIMIT_EXCEEDED`.
-
----
-
-### 8.7 `fcm_export_model(format, selection?)`
-
-**Input schema:**
-```json
-{
-  "format": "csv_bundle_v1 (required)",
-  "selection": {
-    "scope": "all|tag|subgraph",
-    "tag": "string (optional)",
-    "seed_nodes": ["string"]
-  },
-  "project": "string (optional)"
-}
-```
-
-**Output schema:**
-```json
-{
-  "export_id": "string",
-  "format": "csv_bundle_v1",
-  "files": [
-    {"name": "nodes.csv", "path": "string"},
-    {"name": "edges.csv", "path": "string"}
-  ],
-  "node_count": 0,
-  "edge_count": 0
-}
-```
-
-**Deterministic fields:** file set and row counts for fixed selection.  
-**Probabilistic fields:** none.
-
-**Latency target:** p95 <= 1800ms for 50k edges export.
-
-**Scale envelope:**
-1. Max export rows: 500k total.
-2. Pagination or scoped export required above cap.
-
-## 9) Data Model and Storage Boundaries
-
-### 9.1 Knowledge graph schema (descriptive)
-
-`KnowledgeNode`:
-1. `id: str`
-2. `kind: note|decision|spec|concept|person|project`
-3. `title: str`
-4. `permalink: str`
-5. `tags: list[str]`
-6. `updated_at: datetime`
-
-`KnowledgeEdge`:
-1. `id: str`
-2. `src_id: str`
-3. `dst_id: str`
-4. `relation: str`
-5. `directionality: directed|bidirectional`
-6. `evidence_refs: list[str]`
-7. `confidence: float [0,1]`
-8. `updated_at: datetime`
-
-### 9.2 FCM schema (causal signed weighted)
-
-`FCMNode`:
-1. `id: str`
-2. `label: str`
-3. `node_type: goal|driver|risk|intervention|context`
-4. `state: float [-1,1]`
-5. `clamp_min: float`
-6. `clamp_max: float`
-7. `metadata: map`
-
-`FCMEdge`:
-1. `id: str`
-2. `source_id: str`
-3. `target_id: str`
-4. `weight: float [-1,1]`
-5. `confidence: float [0,1]`
-6. `time_decay: float [0,1]`
-7. `evidence_refs: list[str]`
-8. `updated_at: datetime`
-
-### 9.3 Provenance model
-
-`ProvenanceRecord`:
-1. `entity_id: str`
-2. `evidence_refs: list[str]`
-3. `confidence: float [0,1]`
-4. `updated_at: datetime`
-5. `source_type: extracted|user_authored|imported`
-
-### 9.4 Scenario model
-
-`Scenario`:
-1. `id: str`
-2. `name: str`
-3. `interventions: list[{node_id, delta}]`
-4. `constraints: list[{node_id, min, max}]`
-5. `steps: int`
-6. `activation: tanh|sigmoid|bounded_linear`
-7. `created_at: datetime`
-8. `created_by: str`
-
-`ScenarioResult`:
-1. `scenario_id: str`
-2. `converged: bool`
-3. `iterations_used: int`
-4. `residual: float`
-5. `goal_deltas: list[{node_id, delta}]`
-6. `confidence: float [0,1]`
-
-### 9.5 Storage boundaries
-
-| Layer | System of Record | Purpose | Rebuildable |
-|---|---|---|---|
-| Markdown files | File system | Canonical knowledge content | No |
-| SQLite entities/relations/embeddings | SQLite | Operational queries and project state | Yes (from markdown + embedding pipeline) |
-| Knowledge graph triples | Oxigraph | Fast graph traversal and pattern queries | Yes |
-| FCM model and snapshots | SQLite + optional artifacts | Causal model state and scenario history | Yes (from imports and authored model definitions) |
-
-## 10) Mental Modeler Interoperability
-
-### v1 interoperability contract
-
-Format: `csv_bundle_v1`
-1. `nodes.csv`
-2. `edges.csv`
-3. Optional `scenarios.csv`
-
-`nodes.csv` required columns:
-1. `node_id`
-2. `label`
-3. `node_type`
-4. `state`
-5. `clamp_min`
-6. `clamp_max`
-
-`edges.csv` required columns:
-1. `edge_id`
-2. `source_id`
-3. `target_id`
-4. `weight`
-5. `confidence`
-6. `evidence_refs` (semicolon-delimited)
-
-### Import rules
-
-1. Missing required columns fail with `MODEL_INVALID`.
-2. Unknown node types fail fast.
-3. Weight and confidence ranges are strictly validated.
-4. Import returns warnings for dangling evidence references.
-
-### Export rules
-
-1. Preserve stable IDs for round-trip compatibility.
-2. Preserve signed weights exactly.
-3. Preserve confidence values exactly.
-4. Non-portable metadata is emitted to `metadata.json` sidecar when present.
-
-### Native file translators
-
-1. Native project-format translation is deferred to v2.
-2. CSV remains the guaranteed compatibility baseline in v1 and v1.5.
-
-## 11) Pricing and Packaging
-
-### Tier structure
-
-| Tier | Price Monthly | Price Annual | Beta Price (25% off) | Target Persona | Core Value |
-|---|---:|---:|---:|---|---|
-| OSS Local | $0 | $0 | $0 | Casual local users | Retrieval and memory basics |
-| Local+ Graph Intelligence | $9 | $90 | $6.75 monthly / $67.50 annual | Founders, consultants, researchers | Safer changes + explainable graph + FCM simulation |
-| Cloud Pro (current anchor) | $19 | $190 | $14.25 monthly / $142.50 annual | Users who need hosted sync and cloud workflows | Managed cloud + sync + collaboration path |
-
-Pricing principles:
-1. Local+ is intentionally priced between free OSS and cloud to capture users who need deeper intelligence but not hosted sync.
-2. Cloud Pro remains the hosted convenience anchor and future collaboration path.
-3. Local+ must stand on standalone local value and cannot depend on cloud features.
-
-### Feature gate mapping
-
-| Capability | OSS Local | Local+ | Cloud Pro |
-|---|---|---|---|
-| Search and basic context tools | Yes | Yes | Yes |
-| Decision Lineage | No | Yes | Yes |
-| Impact Radius | No | Yes | Yes |
-| Graph Health | No | Yes | Yes |
-| FCM simulate + rank | No | Yes | Yes |
-| CSV model import/export | No | Yes | Yes |
-| Hosted collaboration controls | No | No | Future add-on |
-
-### Packaging decisions
-
-1. Local+ remains fully local-capable and does not require cloud auth to run.
-2. Cloud Pro remains the hosted convenience and collaboration anchor.
-3. Future hosted research add-on will layer on Cloud Pro after v2 readiness.
-
-## 12) Rollout Strategy
-
-### 12.1 Document production iterations (locked)
-
-Iteration 1 (draft complete):
-1. Complete all 15 sections in one pass.
-2. Include v1/v1.5/v2 cut lines.
-3. Include pricing and scenario definitions.
-4. Ensure no placeholders.
-
-Iteration 2 (hardening and decision lock):
-1. Resolve cross-section contradictions.
-2. Convert uncertain language to locked decisions.
-3. Add measurable acceptance criteria and risk owners.
-4. Finalize execution-ready API contracts.
-
-### 12.2 Product rollout phases
-
-Phase A: Foundation release (v1)
-1. Graph lineage, impact, health.
-2. CSV model import/export.
-3. FCM simulation and ranking.
-4. Advanced mode UX gating for research-grade controls.
-
-Phase B: Quality and confidence (v1.5)
-1. Contradiction Watch.
-2. Priority Briefs.
-3. Improved uncertainty propagation.
-4. Confidence calibration pass using real-world model feedback.
-
-Phase C: Team expansion (v2)
-1. Hosted team governance.
-2. Shared model controls.
-3. Extended translator support.
-
-### 12.3 Go/No-Go release gates
-
-Gate to ship v1 default workflows:
-1. p95 latency targets are met within the declared scale envelopes.
-2. Scenario round-trip fidelity tests pass for CSV import/export.
-3. Every recommendation and simulation path exposes evidence references and confidence.
-
-Gate to ship v1.5:
-1. Contradiction Watch precision is acceptable for default-on use.
-2. Confidence calibration reduces false-confidence reports in user testing.
-
-Gate to ship v2 team features:
-1. Clear willingness-to-pay signal from team and research buyers.
-2. Cloud execution path preserves local-cloud semantic parity for core contracts.
-
-## 13) Risks, Counterarguments, and Mitigations
-
-| Risk | Counterargument | Severity | Likelihood | Mitigation | Owner |
-|---|---|---|---|---|---|
-| "This is just better search" | Positioning can collapse into technical jargon | High | Medium | Lead with decision safety and explainability outcomes in product copy and onboarding | Product Lead |
-| FCM feels opaque or invented | Users distrust black-box scoring | High | Medium | Require evidence refs and confidence disclosure on every recommendation | Applied AI Lead |
-| Local performance regressions | Multi-hop and simulation can feel slow on laptops | Medium | Medium | Enforce envelopes, caps, and fail-fast limit errors with guidance | Engineering Lead |
-| Research features overwhelm casual users | UX complexity can reduce adoption | Medium | High | Default to guided flows and hide advanced controls behind explicit advanced mode | Design Lead |
-| License/roadmap conflict if backend changes | Later swap to restrictive engines creates GTM risk | High | Low | Lock permissive v1 stack and require leadership sign-off for any license-restricted dependency | Product + Legal |
-| Interop mismatch with external tools | Round-trip drift harms trust with researchers | Medium | Medium | Validate node and edge parity in import/export tests and version interop schema | Integrations Lead |
-| Pricing confusion between Local+ and Cloud Pro | Buyers may not understand which tier fits | Medium | Medium | Publish explicit tier comparison focused on local intelligence vs hosted collaboration | GTM Lead |
-
-## 14) Acceptance Criteria
-
-### 14.1 Document acceptance criteria
-
-1. Product, technical, and pricing decisions are explicit and unambiguous.
-2. API contracts include input/output schemas, error models, deterministic versus probabilistic fields, and performance envelopes.
-3. v1/v1.5/v2 cut lines are explicit and consistent.
-4. Risk register includes severity, likelihood, owner, and mitigation.
-5. Document can be handed to implementation without additional architecture decisions.
-6. Leadership can use this document directly for pricing and positioning decisions.
-
-### 14.2 Product acceptance criteria for v1 delivery
-
-1. `graph_lineage`, `graph_impact`, `graph_health`, `fcm_simulate`, `fcm_rank_actions`, `fcm_import_model`, and `fcm_export_model` are available as Local+ contracts.
-2. Local mode executes all v1 contracts without cloud dependency.
-3. API p95 latency targets are met within defined scale envelopes.
-4. Every ranked or simulated output includes evidence-linked rationale and confidence.
-5. CSV round-trip preserves node count, edge count, and signed weights exactly.
-
-### 14.3 Scenario test matrix (required)
-
-1. **Casual local user impact check**
-Expected pass:
-`graph_impact` returns ranked affected notes with reasons before a note edit.
-
-2. **Research workflow simulation**
-Expected pass:
-User imports a model bundle, runs `fcm_simulate`, and receives converged deltas and rationale.
-
-3. **Decision audit traceability**
-Expected pass:
-`graph_lineage` returns path and evidence references that explain recommendation origin.
-
-4. **Cloud and local parity**
-Expected pass:
-Identical query inputs return semantically equivalent outputs in local and cloud modes with local fallback behavior when cloud is unavailable.
-
-5. **Sparse or contradictory graph behavior**
-Expected pass:
-System degrades gracefully with explicit uncertainty and does not fabricate high-confidence recommendations.
-
-6. **Interop round-trip fidelity**
-Expected pass:
-`fcm_export_model` then `fcm_import_model` preserves node and edge counts and signed weights without mutation.
-
-## 15) Appendix (license notes, terminology, examples)
-
-### 15.1 License notes (verified 2026-03-05)
-
-1. SurrealDB core licensing is published under BSL 1.1 with DBaaS-related restrictions in its conversion window.
-2. FalkorDB is published under SSPLv1.
-3. pyoxigraph is dual-licensed Apache-2.0 or MIT.
-4. Neon extension catalog currently does not list Apache AGE as a supported extension.
-
-These notes support the v1 dependency decisions in this document.
-
-### 15.2 Terminology
-
-1. **Knowledge graph:** descriptive relation graph derived from markdown knowledge.
-2. **FCM:** fuzzy cognitive model with signed weighted causal edges.
-3. **Deterministic field:** reproducible output field from fixed input and fixed model/index snapshot.
-4. **Probabilistic field:** score influenced by confidence weights and model uncertainty.
-
-### 15.3 Assumptions and defaults
-
-1. Markdown remains source of truth.
-2. SQLite remains mandatory baseline.
-3. Graph and FCM capabilities are premium Local+ features, not OSS defaults.
-4. Research-heavy features are advanced mode, while mainstream UX stays guided.
-5. Mental Modeler interoperability starts with CSV contract first; native translator support is deferred.
-
-### 15.4 Out of scope for this document
-
-1. Implementation code changes.
-2. Database migrations.
-3. Cloud infrastructure edits.
-4. Full instrumentation pilot plan as the primary artifact.
-
-### 15.5 External references
-
-1. [SurrealDB licensing](https://surrealdb.com/license)
-2. [FalkorDB licensing](https://docs.falkordb.com/References/license.html)
-3. [pyoxigraph package and license](https://pypi.org/project/pyoxigraph/)
-4. [Neon Postgres extension catalog](https://neon.com/docs/extensions/pg-extensions)

docs/specs/SPEC-LOCAL-GRAPH-INTELLIGENCE-TECHNICAL-ADDENDUM.md

@@ -1,299 +0,0 @@
-# SPEC-LOCAL-GRAPH-INTELLIGENCE: Technical Addendum (Graph + FCM)
-
-**Status:** Draft  
-**Date:** 2026-03-05  
-**Owner:** Basic Memory
-**Current Phase (2026-03-05):** Contract skeleton implementation is complete; next active phase is SQL-backed graph capabilities.
-
-Related product spec:
-`/docs/specs/SPEC-LOCAL-GRAPH-INTELLIGENCE.md`
-
-Related execution spec:
-`/docs/specs/SPEC-LOCAL-GRAPH-INTELLIGENCE-IMPLEMENTATION-PLAN.md`
-
-## Why This Addendum Exists
-
-The product spec defines user value. This addendum defines the technical shape that can deliver that value without
-breaking local-first principles.
-
-This addendum also introduces a second graph layer:
-
-1. Knowledge graph for relationships between notes, entities, and decisions.
-2. Fuzzy Cognitive Model (FCM) graph for weighted causal reasoning over actions and outcomes.
-
-Both are derived from markdown and optional user-provided models.
-
-## Strategic Reality Check
-
-This is a strong idea if we stage it correctly.
-
-It is not a pipe dream if we avoid one trap: building a big "graph platform" before proving users repeatedly use
-decision simulation workflows.
-
-The correct strategy is:
-
-1. Launch high-precision graph insights first.
-2. Add FCM scoring where it changes user behavior (not as a novelty dashboard).
-3. Expand to hosted/team workflows only after local usage proves repeat value.
-
-## Constraints and Design Principles
-
-1. SQLite remains the operational source for entities, observations, relations, and embeddings.
-2. Markdown remains source of truth.
-3. Graph indexes are derived, rebuildable, and disposable.
-4. Premium local mode must run fully offline.
-5. Cloud deployment should support both single-tenant and SaaS later.
-6. Avoid licenses that constrain hosted/open-source strategy.
-
-## Backend Recommendation
-
-### Primary Recommendation
-
-Use a dual-store architecture:
-
-1. SQLite (existing): operational data, metadata filters, embeddings, and most retrieval.
-2. Oxigraph/pyoxigraph (new): derived graph index for graph traversal and graph-pattern queries.
-3. Python simulation layer (new): FCM state propagation, scenario runs, and decision scoring.
-
-Why this is the best fit:
-
-1. Permissive licensing profile.
-2. Works locally with low footprint.
-3. Cloud-compatible as sidecar service while keeping Neon Postgres as core cloud store.
-4. Clear boundary between graph query and numeric simulation concerns.
-
-### Candidate Trade-Offs
-
-#### Oxigraph/pyoxigraph
-
-Pros:
-
-1. Lightweight local embedding.
-2. Good fit for derived-index strategy.
-3. Strong path for standards-based graph representation.
-
-Cons:
-
-1. SPARQL fluency is less common than SQL/Cypher.
-2. Requires a translation layer so product features are not query-language-coupled.
-
-#### Apache AGE (Postgres extension)
-
-Pros:
-
-1. SQL + graph in one engine.
-2. Attractive for cloud-side graph operations.
-
-Cons:
-
-1. Neon support is uncertain for this extension.
-2. Local/cloud parity is harder if local uses SQLite.
-
-#### SurrealDB / FalkorDB
-
-Pros:
-
-1. Strong graph-oriented developer experience.
-
-Cons:
-
-1. License posture is misaligned with a future hosted/open-source roadmap unless commercial terms are accepted.
-
-Decision:
-Do not make these core dependencies for v1 of Local+ Graph Intelligence.
-
-## Two-Graph Model
-
-### A) Knowledge Graph (Descriptive)
-
-Node examples:
-
-1. Note
-2. Decision
-3. Spec
-4. Person
-5. Project
-6. Concept
-
-Edge examples:
-
-1. `depends_on`
-2. `informed_by`
-3. `contradicts`
-4. `supports`
-5. `implements`
-6. `derived_from`
-
-Purpose:
-Power navigation, lineage, path explanation, impact radius, and health checks.
-
-### B) FCM Graph (Causal, Signed, Weighted)
-
-Node examples:
-
-1. Goal: "Reduce regressions"
-2. Driver: "Test coverage"
-3. Risk: "Scope creep"
-4. Intervention: "Add review gate"
-5. Context variable: "Team bandwidth"
-
-Edge attributes:
-
-1. `weight` in [-1.0, 1.0]
-2. `confidence` in [0.0, 1.0]
-3. `evidence_refs` (links to notes/specs)
-4. `time_decay` (optional)
-
-Purpose:
-Power scenario simulation and action ranking, not generic retrieval.
-
-## Premium Feature Mapping to Architecture
-
-### Decision Lineage
-
-Backed by:
-
-1. Knowledge graph path queries.
-2. Evidence references stored on edges.
-
-### Impact Radius
-
-Backed by:
-
-1. Multi-hop neighborhood expansion with relation-type weights.
-2. Risk ranking using centrality + recency + confidence.
-
-### Contradiction Watch
-
-Backed by:
-
-1. Candidate contradiction edges.
-2. Confidence-scored reconciliation queue.
-
-### Priority Briefs
-
-Backed by:
-
-1. Health metrics (orphan rate, stale-central nodes, unresolved contradictions).
-2. Optional FCM "top leverage actions" summary.
-
-### New Premium Feature: Action Simulator
-
-Backed by:
-
-1. FCM scenario runs over selected action nodes.
-2. Ranked interventions with expected positive/negative downstream effects.
-3. Explicit rationale graph for every recommendation.
-
-## Mental Modeler Interop Plan
-
-Goal:
-Make Basic Memory the AI-enabled operating layer around existing researcher workflows, not a replacement for their tools.
-
-Interoperability phases:
-
-1. Import/export edge lists and node tables via CSV as the baseline interchange.
-2. Preserve concept IDs and metadata so round-trips remain stable.
-3. Add translator support for native model files if/when schema contracts are validated with partner data.
-
-Validation requirement:
-
-1. Round-trip tests must preserve node count, edge count, and signed weights.
-2. Confidence/evidence metadata may be Basic Memory extensions and should degrade gracefully when exported.
-
-## Suggested Tool/API Surface (Product-Facing)
-
-1. `graph_lineage(start, goal?)`  
-   Returns explainable evidence paths.
-2. `graph_impact(target, horizon=2..4)`  
-   Returns ranked affected nodes with reasons.
-3. `graph_health()`  
-   Returns actionable graph quality issues.
-4. `fcm_simulate(actions, scenario?)`  
-   Returns projected effects and uncertainty.
-5. `fcm_rank_actions(goal, constraints?)`  
-   Returns top candidate actions with trade-offs.
-6. `fcm_import_model(source)` / `fcm_export_model(format)`  
-   Handles interop with external cognitive mapping workflows.
-
-## Local and Cloud Deployment Shape
-
-### Local (Primary)
-
-1. SQLite + local embeddings.
-2. Oxigraph as local sidecar/index library.
-3. FCM simulation in process.
-
-### Cloud (Future-Compatible)
-
-1. Neon Postgres remains system of record in hosted mode.
-2. Graph index service runs per tenant or shared multi-tenant with strict tenancy boundaries.
-3. FCM simulation service can run stateless workers reading graph snapshots.
-
-Principle:
-Do not require cloud to run premium local features.
-
-## Rollout Plan With Go/No-Go Gates
-
-### Phase 0: Proof of Utility (4-6 weeks)
-
-Deliver:
-
-1. Decision Lineage
-2. Impact Radius
-3. CSV FCM import + `fcm_simulate` prototype
-
-Gate to continue:
-
-1. Repeated weekly usage by pilot users.
-2. Users report changed decisions, not just curiosity clicks.
-
-### Phase 1: Productized Local+ Beta
-
-Deliver:
-
-1. Graph health workflow
-2. Contradiction Watch
-3. Action ranking with explicit rationale
-
-Gate to continue:
-
-1. Retention of graph features after first month.
-2. Measured reduction in "surprise side effects" after edits.
-
-### Phase 2: Hosted Expansion
-
-Deliver:
-
-1. Optional cloud execution for heavy simulations.
-2. Team-shared model governance.
-
-Gate to continue:
-
-1. Clear willingness to pay for hosted collaboration.
-
-## Risks and Mitigations
-
-Risk: FCM outputs feel "made up."  
-Mitigation: Require evidence links and confidence scoring in every recommendation.
-
-Risk: Research-heavy feature alienates casual users.  
-Mitigation: Keep FCM features in an advanced mode; default to concise guidance workflows.
-
-Risk: Overengineering early graph stack.  
-Mitigation: Keep derived-index architecture and strict phase gates tied to behavior change.
-
-Risk: Interop friction with external tooling.  
-Mitigation: Start with transparent CSV contract and strict round-trip validation.
-
-## Candid Recommendation
-
-Pursue this. It is a high-upside differentiation path for Local+ if executed with staged validation.
-
-The key is to sell outcomes:
-
-1. "Safer decisions"
-2. "Explainable recommendations"
-3. "Faster synthesis for complex research"
-
-Avoid selling "graph DB" as the product. That is implementation detail.

docs/specs/SPEC-LOCAL-GRAPH-INTELLIGENCE.md

@@ -1,262 +0,0 @@
-# SPEC-LOCAL-GRAPH-INTELLIGENCE: Premium Local Graph Intelligence
-
-**Status:** Draft  
-**Date:** 2026-03-05  
-**Owner:** Basic Memory
-**Current Phase (2026-03-05):** Phase 1 contract foundation shipped; engineering is now executing SQL-backed Phase 2 graph logic.
-
-Companion technical addendum:
-`/docs/specs/SPEC-LOCAL-GRAPH-INTELLIGENCE-TECHNICAL-ADDENDUM.md`
-
-## Summary
-
-Add a premium local feature that turns Basic Memory from "search and recall" into "explain and guide."
-
-The value is not a new database. The value is better decisions for local users:
-
-1. Understand why something matters.
-2. See what will be affected before making a change.
-3. Detect weak spots in the knowledge base early.
-4. Navigate complex knowledge intentionally instead of loading everything.
-
-This feature is additive. Existing local workflows remain intact.
-
-## Positioning
-
-Core message:
-"Your notes do more than store knowledge. They reveal consequences, lineage, and blind spots."
-
-Local user promise:
-
-1. Keep files local.
-2. Keep markdown as source of truth.
-3. Get advanced graph intelligence as an opt-in premium capability.
-
-## Problem
-
-Today, deep graph navigation is possible but often expensive in context size and hard to steer for complex questions.
-Users can find information, but they still do manual synthesis to answer:
-
-1. What changed because of this note?
-2. Why did we decide this?
-3. What might break if I update this?
-4. Which parts of the graph are stale, isolated, or contradictory?
-
-The cost is time, cognitive load, and missed risk.
-
-## Goals
-
-1. Provide clear, explainable graph insights that users can act on.
-2. Make deep navigation feel guided, not overwhelming.
-3. Help users prevent mistakes before they happen.
-4. Create premium local value that is easy to understand and justify.
-5. Keep feature behavior transparent and trustworthy.
-
-## Non-Goals
-
-1. Replacing SQLite as the primary operational store.
-2. Changing markdown as source of truth.
-3. Forcing users to learn graph query languages.
-4. Building a cloud-only feature set.
-5. Turning Basic Memory into an enterprise BI product.
-
-## Product Frame: From Retrieval to Reasoning
-
-The feature should be framed as a shift in user outcome:
-
-1. Retrieval: "Find me the note."
-2. Reasoning: "Show me the path, impact, and confidence around this note."
-
-This is the main narrative upgrade for premium local users.
-
-## Premium Value Pillars
-
-### 1) Decision Confidence
-
-Users can see decision lineage:
-
-1. What evidence supported a decision.
-2. Which notes/specs informed it.
-3. How that decision evolved over time.
-
-### 2) Change Safety
-
-Users can run impact-aware workflows:
-
-1. Estimate blast radius before editing.
-2. Surface downstream dependencies.
-3. Prioritize what to review first.
-
-### 3) Knowledge Quality
-
-Users can maintain graph health:
-
-1. Detect orphaned notes.
-2. Detect overloaded hub notes.
-3. Detect stale but high-centrality notes.
-4. Detect likely contradictions.
-
-### 4) Guided Navigation
-
-Users can explore deeper relationships without context explosion:
-
-1. Follow promising branches.
-2. Stop when confidence is sufficient.
-3. Avoid "load everything and hope."
-
-## Feature Catalog (Value-First)
-
-### A. Decision Lineage
-
-What users get:
-
-1. A clear "why chain" for important conclusions.
-2. Traceable connections to supporting notes.
-3. Better handoffs and historical understanding.
-
-### B. Impact Radius
-
-What users get:
-
-1. A ranked list of likely affected notes before edits.
-2. Safer refactors for docs, plans, and architecture.
-3. Reduced accidental drift and inconsistency.
-
-### C. Knowledge Health Dashboard
-
-What users get:
-
-1. Weekly health signals for the graph.
-2. Actionable cleanup targets.
-3. Better long-term memory quality with less manual auditing.
-
-### D. Path Explorer
-
-What users get:
-
-1. "Show me how A connects to B" style explanations.
-2. Multiple candidate paths with confidence cues.
-3. Better discovery across large note collections.
-
-### E. Contradiction Watch
-
-What users get:
-
-1. Early warnings for conflicting statements.
-2. Suggested reconciliation workflow.
-3. Higher trust in the knowledge base.
-
-### F. Priority Briefs
-
-What users get:
-
-1. Periodic "what matters now" graph summaries.
-2. Focused recommendations, not noisy activity dumps.
-3. Better focus for solo builders and small teams.
-
-## User Personas and Why They Pay
-
-### Solo Technical Founder
-
-Pain:
-Cannot hold full architecture and decision history in working memory.
-
-Premium value:
-Impact Radius + Decision Lineage prevent rework and regressions.
-
-### Product/Research Lead
-
-Pain:
-Knowledge is fragmented across specs, notes, and decisions.
-
-Premium value:
-Path Explorer + Priority Briefs compress synthesis time.
-
-### Consultant/Fractional Operator
-
-Pain:
-Frequent context switching across domains and clients.
-
-Premium value:
-Knowledge Health + Decision Lineage speed onboarding and reporting.
-
-## Packaging Direction
-
-Suggested packaging:
-
-1. OSS Local: existing search + context tools.
-2. Local+ Graph Intelligence: advanced graph insight features listed above.
-3. Future Team Add-On: shared policies, shared graph health views, shared lineage views.
-
-Core upsell line:
-"Keep your local workflow. Add graph intelligence when complexity grows."
-
-## Experience Principles
-
-1. Explainability first.
-Every advanced result should show "why this was suggested."
-
-2. Actionability over novelty.
-Insights should lead to concrete next steps, not abstract charts.
-
-3. Progressive disclosure.
-Start with concise summaries, expand on demand.
-
-4. Deterministic where possible.
-Users should trust repeated runs of the same workflow.
-
-5. Respect local-first expectations.
-No surprise cloud dependency in premium local mode.
-
-## Success Criteria (Product)
-
-1. Users can describe the benefit in one sentence:
-"It shows me what matters and what breaks before I change things."
-2. Premium users report lower time-to-understanding for complex topics.
-3. Premium users report fewer "surprise side effects" after edits.
-4. Premium users keep larger knowledge graphs healthy with less manual effort.
-5. Feature adoption is driven by outcomes, not by curiosity-only usage.
-
-## Risks and Mitigations
-
-Risk: Feature sounds like "just better search."  
-Mitigation: Lead messaging with decision confidence and change safety, not traversal depth.
-
-Risk: Feature feels too advanced for normal users.  
-Mitigation: Package as guided insights and reports, not as a query language.
-
-Risk: Insight quality feels noisy.  
-Mitigation: Focus launch scope on high-precision insight types and transparent rationale.
-
-Risk: Value is hard to prove.  
-Mitigation: Track user-facing outcomes (time saved, risk avoided, cleanup completed).
-
-## Rollout Narrative
-
-Phase 1: "Safer Changes"
-
-1. Impact Radius
-2. Decision Lineage
-
-Phase 2: "Health and Clarity"
-
-1. Knowledge Health Dashboard
-2. Contradiction Watch
-
-Phase 3: "Strategic Navigation"
-
-1. Path Explorer
-2. Priority Briefs
-
-## One-Line Positioning Options
-
-1. "Local notes, strategic intelligence."
-2. "Know what changed, why it matters, and what it affects."
-3. "From note-taking to decision support."
-
-## Open Questions
-
-1. Which two features best define the paid tier at launch?
-2. Which insight types should be guaranteed deterministic in v1?
-3. Should Priority Briefs be bundled or separate as an add-on?
-4. What is the simplest in-product education flow for first-time premium users?

glama.json

@@ -0,0 +1,7 @@
+{
+  "$schema": "https://glama.ai/mcp/schemas/server.json",
+  "maintainers": [
+    "phernandez",
+    "groksrc"
+  ]
+}

integrations/hermes/.coveragerc

@@ -0,0 +1,9 @@
+[run]
+source = .
+omit =
+    tests/*
+    tests/**/*
+
+[report]
+fail_under = 85
+show_missing = True

integrations/hermes/.github/workflows/integration.yml

@@ -0,0 +1,48 @@
+name: integration
+
+# Heavier than test.yml — installs the real `basic-memory` CLI via uv, runs
+# every bm_* tool against a live `bm mcp` subprocess. Catches BM-API drift
+# (e.g., a bm release renaming a tool argument) before our users see it.
+concurrency:
+  group: hbm-integration-${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+on:
+  push:
+    branches: [main]
+  workflow_dispatch:
+
+jobs:
+  integration:
+    name: Integration tests (real bm + mcp)
+    runs-on: ubuntu-latest
+    timeout-minutes: 15
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install uv
+        # No cache (no uv.lock to key off — see test.yml comment).
+        uses: astral-sh/setup-uv@v3
+
+      - name: Set up Python 3.12
+        # basic-memory itself requires 3.12+; the bm install needs that.
+        # Hermes-runtime compatibility (3.11) is covered by test.yml.
+        run: uv python install 3.12
+
+      - name: Install basic-memory CLI via uv
+        run: |
+          uv tool install basic-memory
+          # uv puts entry-point shims under ~/.local/bin
+          echo "$HOME/.local/bin" >> "$GITHUB_PATH"
+
+      - name: Verify bm is on PATH
+        run: |
+          which bm
+          bm --version
+
+      - name: Run integration tests
+        env:
+          BM_INTEGRATION: "1"
+        run: |
+          uv run --with pytest --with mcp --python 3.12 pytest tests/test_integration.py -v

integrations/hermes/.github/workflows/pr-title.yml

@@ -0,0 +1,36 @@
+name: pr-title
+
+on:
+  pull_request:
+    types: [opened, edited, synchronize]
+
+jobs:
+  semantic-pr-title:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: amannn/action-semantic-pull-request@v5
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+        with:
+          # Conventional-commit types we accept in PR titles + commit subjects.
+          types: |
+            feat
+            fix
+            chore
+            docs
+            style
+            refactor
+            perf
+            test
+            build
+            ci
+          # Single-file plugin — no real submodule structure. We don't require
+          # a scope, but if a contributor uses one we accept these:
+          scopes: |
+            core
+            tests
+            ci
+            docs
+            deps
+          requireScope: false
+          requireScopeForBreakingChange: true

integrations/hermes/.github/workflows/release.yml

@@ -0,0 +1,194 @@
+name: release
+
+# Manual trigger: Actions → release → Run workflow. Always runs against main
+# (the workflow validates GITHUB_REF). Steps:
+#   1. Compute the new version from the `version` input (patch/minor/major
+#      or explicit semver). The *current* version is read from the latest
+#      git tag (`v*.*.*`) — NOT from __init__.py. This is robust to PRs that
+#      pre-bump __init__.py: the bump always runs from the last released
+#      version, not from whatever the working files happen to say.
+#   2. Update __version__ in __init__.py and version in plugin.yaml to match
+#      the new tag — bringing the files in sync if a PR pre-bumped them.
+#   3. Commit as `chore(release): vX.Y.Z`, tag, push to main + push the tag.
+#   4. Publish a GitHub Release. Body is the matching `## [X.Y.Z]` block from
+#      CHANGELOG.md when present; otherwise auto-generated release notes.
+#
+# Recommended flow: land a PR that adds a `## [X.Y.Z]` section to CHANGELOG.md
+# first, then run this workflow with the matching version so the release notes
+# are the hand-written changelog instead of commit-message-derived notes.
+on:
+  workflow_dispatch:
+    inputs:
+      version:
+        description: "Version bump (`patch`, `minor`, `major`) or explicit semver (`0.3.0`)"
+        required: true
+        default: "patch"
+
+permissions:
+  contents: write
+
+concurrency:
+  group: release-${{ github.ref }}
+  cancel-in-progress: false
+
+jobs:
+  release:
+    name: Tag and Publish GitHub Release
+    runs-on: ubuntu-latest
+    steps:
+      - name: Validate trigger is main
+        run: |
+          if [ "$GITHUB_REF" != "refs/heads/main" ]; then
+            echo "::error::release must run against main. Got $GITHUB_REF"
+            exit 1
+          fi
+
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          token: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Compute new version
+        id: bump
+        run: |
+          set -euo pipefail
+          VERSION_INPUT="${{ github.event.inputs.version }}"
+
+          # Current = latest released tag (sort by version, descending; pick
+          # the first v*.*.* tag). NOT __init__.py — a PR may have pre-bumped
+          # the version files, and we don't want to double-bump on top of
+          # that. The bump is computed from the last *released* version.
+          LAST_TAG=$(git tag --list 'v*.*.*' --sort=-v:refname | head -1 || true)
+          if [ -z "$LAST_TAG" ]; then
+            # First release in the repo. Seed with 0.0.0 so a `patch` bump
+            # yields v0.0.1, `minor` yields v0.1.0, `major` yields v1.0.0.
+            # Explicit semver inputs bypass the seed entirely.
+            CURRENT="0.0.0"
+            echo "No prior v*.*.* tag found; seeding current=0.0.0"
+          else
+            CURRENT="${LAST_TAG#v}"
+            echo "Latest released tag: $LAST_TAG (current=$CURRENT)"
+          fi
+
+          if [[ "$VERSION_INPUT" =~ ^(patch|minor|major)$ ]]; then
+            IFS=. read -r MAJ MIN PAT <<< "$CURRENT"
+            case "$VERSION_INPUT" in
+              major) MAJ=$((MAJ + 1)); MIN=0; PAT=0 ;;
+              minor) MIN=$((MIN + 1)); PAT=0 ;;
+              patch) PAT=$((PAT + 1)) ;;
+            esac
+            NEW="${MAJ}.${MIN}.${PAT}"
+          elif [[ "$VERSION_INPUT" =~ ^[0-9]+\.[0-9]+\.[0-9]+$ ]]; then
+            NEW="$VERSION_INPUT"
+          else
+            echo "::error::Invalid version input: '$VERSION_INPUT'"
+            echo "::error::Use patch|minor|major or explicit X.Y.Z"
+            exit 1
+          fi
+
+          if [ "$NEW" = "$CURRENT" ]; then
+            echo "::error::New version equals last released ($NEW). Pick a different version."
+            exit 1
+          fi
+
+          # Sanity check: refuse if __init__.py is already ahead of the
+          # version we're about to ship. Catches the "PR bumped to 0.5.0 but
+          # workflow was asked for a patch that would land 0.1.8" foot-gun
+          # before it overwrites the files.
+          FILE_VERSION=$(grep -E '^__version__ = ' __init__.py \
+            | sed -E 's/^__version__ = "([^"]+)".*/\1/')
+          if [ -n "$FILE_VERSION" ] && [ "$FILE_VERSION" != "$CURRENT" ] && [ "$FILE_VERSION" != "$NEW" ]; then
+            echo "::error::__init__.py reports version $FILE_VERSION, but the release would ship $NEW (last tag: $CURRENT)."
+            echo "::error::Reconcile by picking a version input that matches __init__.py, or roll __init__.py back to $CURRENT."
+            exit 1
+          fi
+
+          echo "New version: $NEW"
+          echo "current=$CURRENT" >> "$GITHUB_OUTPUT"
+          echo "version=$NEW" >> "$GITHUB_OUTPUT"
+          echo "tag=v$NEW" >> "$GITHUB_OUTPUT"
+
+      - name: Refuse if tag already exists
+        run: |
+          set -euo pipefail
+          TAG="${{ steps.bump.outputs.tag }}"
+          if git rev-parse --verify "refs/tags/$TAG" >/dev/null 2>&1; then
+            echo "::error::Tag $TAG already exists locally. Pick a different version."
+            exit 1
+          fi
+          if git ls-remote --tags origin "$TAG" | grep -q "refs/tags/$TAG$"; then
+            echo "::error::Tag $TAG already exists on origin. Pick a different version."
+            exit 1
+          fi
+
+      - name: Update version files
+        run: |
+          set -euo pipefail
+          NEW="${{ steps.bump.outputs.version }}"
+          sed -i -E "s/^__version__ = \"[^\"]+\"/__version__ = \"${NEW}\"/" __init__.py
+          sed -i -E "s/^version: .*/version: ${NEW}/" plugin.yaml
+
+          # Verify both files changed and that the new version is present.
+          grep -q "^__version__ = \"${NEW}\"" __init__.py
+          grep -q "^version: ${NEW}$" plugin.yaml
+
+          echo "--- diff ---"
+          git --no-pager diff -- __init__.py plugin.yaml
+
+      - name: Configure Git identity
+        run: |
+          git config user.name "github-actions[bot]"
+          git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
+
+      - name: Commit and tag
+        run: |
+          set -euo pipefail
+          TAG="${{ steps.bump.outputs.tag }}"
+          git add __init__.py plugin.yaml
+          if git diff --cached --quiet; then
+            # PR already bumped the files to the target version. Tag the
+            # existing HEAD rather than creating an empty release commit.
+            echo "Version files already at ${TAG}; tagging current HEAD."
+          else
+            git commit -m "chore(release): ${TAG}"
+          fi
+          git tag -a "${TAG}" -m "${TAG}"
+
+      - name: Push commit and tag
+        run: |
+          set -euo pipefail
+          # HEAD push is a no-op when nothing was committed in this run.
+          git push origin HEAD:main
+          git push origin "${{ steps.bump.outputs.tag }}"
+
+      - name: Extract CHANGELOG section for this version
+        id: changelog
+        run: |
+          set -euo pipefail
+          VERSION="${{ steps.bump.outputs.version }}"
+          # Pull lines between `## [VERSION]` and the next `## [` heading.
+          SECTION=$(awk -v ver="$VERSION" '
+            $0 ~ "^## \\[" ver "\\]" { found=1; next }
+            found && /^## \[/ { exit }
+            found { print }
+          ' CHANGELOG.md)
+          if [ -z "$SECTION" ]; then
+            echo "::warning::No CHANGELOG.md section found for v${VERSION} — falling back to auto-generated release notes."
+            echo "has_section=false" >> "$GITHUB_OUTPUT"
+          else
+            echo "has_section=true" >> "$GITHUB_OUTPUT"
+            {
+              echo 'body<<EOF_CHANGELOG'
+              echo "$SECTION"
+              echo 'EOF_CHANGELOG'
+            } >> "$GITHUB_OUTPUT"
+          fi
+
+      - name: Create GitHub Release
+        uses: softprops/action-gh-release@v2
+        with:
+          tag_name: ${{ steps.bump.outputs.tag }}
+          name: ${{ steps.bump.outputs.tag }}
+          body: ${{ steps.changelog.outputs.body }}
+          generate_release_notes: ${{ steps.changelog.outputs.has_section == 'false' }}
+          token: ${{ secrets.GITHUB_TOKEN }}

integrations/hermes/.github/workflows/test.yml

@@ -0,0 +1,39 @@
+name: tests
+
+# Cancel an in-progress run when a new commit lands on the same branch — the
+# latest result is the one we care about.
+concurrency:
+  group: hbm-tests-${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+on:
+  # Branch pushes already cover PRs (the PR branch tip is what's being tested),
+  # so we don't run the matrix twice for the same commit on push + pull_request.
+  push:
+
+jobs:
+  unit:
+    name: Unit tests (Python ${{ matrix.python-version }})
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    strategy:
+      fail-fast: false
+      matrix:
+        # 3.11 is the runtime Hermes itself ships on today; 3.12-3.14 cover
+        # forward-compat for whenever Hermes upgrades.
+        python-version: ["3.11", "3.12", "3.13", "3.14"]
+
+    steps:
+      - uses: actions/checkout@v4
+
+      - name: Install uv
+        # No `enable-cache: true` — the action's default cache key globs for
+        # `uv.lock`, which we don't ship (we use `uv run --with` instead of
+        # `uv sync`). Without a lock file the cache step errors out.
+        uses: astral-sh/setup-uv@v3
+
+      - name: Set up Python ${{ matrix.python-version }}
+        run: uv python install ${{ matrix.python-version }}
+
+      - name: Run unit tests
+        run: uv run --with pytest --python ${{ matrix.python-version }} pytest -q

integrations/hermes/.gitignore

@@ -0,0 +1,6 @@
+__pycache__/
+*.pyc
+.venv/
+.DS_Store
+.pytest_cache/
+*.egg-info/

integrations/hermes/CHANGELOG.md

@@ -0,0 +1,133 @@
+# Changelog
+
+All notable changes to this project will be documented in this file.
+
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/), and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+
+## [0.3.2] — 2026-05-23
+
+### Fixed
+- **Let Basic Memory v0.21.3 self-route workspace-qualified identifiers and URLs.** Hermes no longer injects its configured default project into `bm_read`, `bm_edit`, `bm_delete`, `bm_move`, or `bm_context` calls when the identifier/URL is already workspace-qualified, such as `personal/main/...`, `memory://personal/main/...`, or an organization workspace slug with a 32-character hash suffix. This preserves Basic Memory Cloud's workspace-aware routing while keeping the existing default-project behavior for short/local identifiers.
+
+## [0.3.1] — 2026-05-16
+
+### Changed
+- **Documented the Hermes Agent v0.14.0-compatible `/bm-*` slash-command monkeypatch.** `MONKEYPATCH.md` now distinguishes the plugin's runtime version from the Hermes Agent-side compatibility patch: plugin `v0.3.0` remains the correct runtime release for Hermes Agent `v0.13.x`, while Hermes Agent `v0.14.0` still needs the updated two-part core patch so gateway startup command discovery loads the active exclusive memory provider and the memory-provider collector delegates `register_command` / `register_skill`.
+- **Clarified install guidance for users and agents.** The README known-issue section now points Hermes `v0.13.x` and `v0.14.0` users at the compatibility matrix in `MONKEYPATCH.md`, so agents do not mistake a plugin update for the required Hermes Agent core patch.
+
+### Notes
+- This is a documentation/compatibility-instructions release only. It does not change the plugin runtime code or Basic Memory data behavior. The plugin remains backward-compatible with Hermes Agent `v0.13.x`; the new documentation explains how to patch Hermes Agent `v0.14.0` until the upstream Hermes fix ships.
+
+## [0.3.0] — 2026-05-12
+
+### Added
+- **Per-call project routing on every `bm_*` tool.** All eight tools now accept optional `project` (name) and `project_id` (UUID from `bm_projects`) parameters. The agent can write or read against a project other than the Hermes-configured one — useful when the user asks to write into a different cloud project (e.g. a personal `main` project) without reconfiguring the plugin. `project_id` takes precedence over `project`; both fall back to the configured default when omitted. Workspace routing is handled transparently by BM via `project_id` — no separate workspace parameter is needed.
+- **`bm_projects` and `bm_workspaces` agent tools.** Promotes the discovery logic previously available only as `/bm-project` and `/bm-workspace` slash commands to agent-facing tools. `bm_projects` returns JSON with `name` and `external_id` (UUID) per project so the agent can hand the UUID to `bm_write` / `bm_read` / etc. via `project_id` — the unambiguous form across cloud workspaces. `bm_workspaces` lists BM Cloud workspaces (name, type, role, default flag). Together with per-call routing, these unblock the workflow Drew's friction note flagged: agent picks the right project + workspace before writing, instead of silently operating against the active Hermes memory project.
+- **SKILL.md cross-project workflow** documenting the discovery → route → write → verify recipe end-to-end. Adds a "Permalinks" section covering the three canonical shapes (short, project-qualified, workspace-qualified) and the round-trip property where `bm_write`'s returned permalink self-routes for follow-up reads. A "Cross-project routing" section explains `project` (including workspace-qualified syntax like `"personal/main"`) vs `project_id` and when to use each. Also backfills `bm_recent` documentation (the tool shipped in 0.2.0 but the skill hadn't been updated).
+- **SKILL.md "Further reading" section** linking to the official docs at [docs.basicmemory.com](https://docs.basicmemory.com), with raw-markdown URLs (`/raw/<path>.md`) the agent can `WebFetch` on demand for deeper material — knowledge format, observations & relations, memory URL wildcards, semantic search, cloud routing, BM's full MCP tool surface, and the `llms.txt` sitemap.
+
+### Notes
+- Addresses the routing, discovery, and documentation gaps in the real-world note "Hermes Basic Memory Cloud Task Experience." A proposed `bm_import` tool was evaluated and dropped — `read_file` + `bm_write` already composes the same operation with no new capability, at the cost of one more tool in the surface.
+- The slash commands `/bm-project` and `/bm-workspace` still exist and behave identically — they continue to call `list_memory_projects` / `list_workspaces` directly via the actor. No behavior change for human use.
+
+## [0.2.0] — 2026-05-11
+
+### Added
+- **Plugin-owned `/bm-*` slash commands** for CLI/gateway sessions. Eight commands give humans direct memory-graph access without going through the agent: `/bm-search`, `/bm-read`, `/bm-context`, `/bm-recent`, `/bm-status`, `/bm-remember`, `/bm-project`, `/bm-workspace`. Closes #2.
+- **`bm_recent` tool** wrapping BM's `recent_activity`. Surfaces notes updated within a timeframe (`7d` default, accepts natural language like `"2 weeks"` or `"yesterday"`). Agent-facing and reused by `/bm-recent`.
+- **`remember_folder` config key** (default `"bm-remember"`). Separate from `capture_folder` so manual captures via `/bm-remember` don't intermix with auto-generated session transcripts. Notes are tagged `manual-capture` for further disambiguation.
+
+### Fixed
+- **`ctx.register_skill(...)` was silently no-opping since 0.1.5** in real Hermes installs. Hermes loads memory-provider plugins through a stripped-down `_ProviderCollector` context (`plugins/memory/__init__.py`) that captures only `register_memory_provider`; `register_skill` and `register_command` are not delegated. The plugin now writes directly to `PluginManager._plugin_commands` and `_plugin_skills`, matching the entry shape and name normalization `PluginContext.register_command` / `register_skill` produce. This makes both the new slash commands and the bundled SKILL.md work in current Hermes installs. The clean fix lives upstream — a small patch to teach `_ProviderCollector` to delegate — and once that lands, the reach-in becomes a redundant double-write of identical entries. Forward-compat `ctx.register_command` / `ctx.register_skill` calls remain in place for the future code path.
+
+### Notes
+- `/bm-remember` derives the title from the first non-empty line of the input, trimmed to 80 chars; falls back to `Note YYYY-MM-DD HHMM UTC`.
+- `/bm-workspace` short-circuits in local mode with a one-line explanation. Workspaces are a BM Cloud concept.
+- Mid-session project/workspace switching is intentionally not supported in 0.2.0 — auto-capture would land in unexpected places. Tracked as a follow-up.
+
+## [0.1.7] — 2026-05-10
+
+### Changed
+- **Stronger nudge in `system_prompt_block()`** to steer agents toward the `bm_*` tools instead of shelling out to `bm` CLI. Pre-v0.1.7 the prompt listed the tools neutrally; given Claude/Hermes models' heavy training-data exposure to `bm tool ...` CLI patterns, neutral language wasn't enough — agents reached for the shell by reflex, paying 1-2s of cold-start per call instead of ~0.1s through our persistent MCP connection. New prompt is explicit (**"Use the `bm_*` tools below directly — do not shell out to the `bm` CLI"**) and gives a one-line latency rationale so the model has a reason to follow it.
+- `SKILL.md` mirrors the directive with a "Use `bm_*`, not the `bm` CLI" section + a tool-vs-CLI table.
+
+### Added
+- Regression test `test_system_prompt_block_steers_away_from_cli` locks in the directive language so future prompt edits don't accidentally weaken it.
+
+## [0.1.6] — 2026-05-10
+
+### Fixed
+- **`bm_*` tools were never registered with Hermes's `MemoryManager._tool_to_provider`.** `get_tool_schemas()` was gated on `self._initialized`, but Hermes captures the schema list at *register* time — before `initialize()` runs. The gate caused every session to start with zero tools registered for our provider, so every LLM-issued `bm_search` (and friends) returned `"Unknown tool: bm_search"` from MemoryManager's dispatch. Symptoms were asymmetric: prefetch (recall injection) worked because it's invoked per-turn after init, but tool calls didn't. Schemas are static — they now return unconditionally, with `handle_tool_call()` doing the runtime "is the actor ready?" gate.
+- Regression test pins this so we don't reintroduce it: `test_get_tool_schemas_unconditional` asserts `get_tool_schemas()` returns all 7 schemas on a fresh, uninitialized provider.
+
+## [0.1.5] — 2026-05-10
+
+### Added
+- Bundled `SKILL.md` is now auto-registered via `ctx.register_skill("basic-memory", ...)` during plugin load. No more manual symlink to `~/.hermes/skills/`. The skill is opt-in (resolvable via `skill:view basic-memory:basic-memory`); always-on agent guidance still flows through `system_prompt_block()`.
+
+### Changed
+- README rewritten for community install. Lead command is now `hermes plugins install basicmachines-co/hermes-basic-memory`. Clone-and-symlink instructions moved to the Development section.
+- Added GitHub Actions CI: unit tests on push and PR.
+- Added this CHANGELOG.
+
+## [0.1.4] — 2026-05-10
+
+### Added
+- `_uv_binary_path()` and `_install_bm_via_uv()`. When `bm` is missing from the host, the plugin runs `uv tool install basic-memory --quiet` once at first `initialize()`. The bm binary lands at `~/.local/bin/bm` — the same canonical path a manual `uv tool install basic-memory` produces, so subsequent manual installs are no-ops rather than creating a second install.
+- 8 new unit tests covering `is_available()` with bm/uv combinations, the install subprocess (success / non-zero exit / OSError / no-uv), and `initialize()` install-or-not branching.
+
+### Changed
+- `is_available()` now returns `True` when **either** `bm` is on disk **or** `uv` is on disk (we can install the missing CLI ourselves).
+- README's prerequisites section: dropped manual basic-memory install requirement; added the one-time ~10s cold-start note.
+
+## [0.1.3] — 2026-05-10
+
+### Fixed
+- README's cloud-mode section described the wrong setup (`bm project add ... --cloud --local-path` + `bm cloud bisync`), which gives a local-mode project with file-level cloud sync rather than true cloud routing. Replaced with `bm project set-cloud <name> --workspace <name>`, which flips the project to `ProjectMode.CLOUD` so tool calls route over HTTPS to `<cloud_host>/proxy` directly. No local files involved.
+- Documented OAuth / API-key auth options, and the `--workspace` requirement when the user belongs to multiple BM Cloud workspaces.
+
+## [0.1.2] — 2026-05-10
+
+### Changed
+- `_default_project()`: `"hermes-memory"` (was `"hermes-{hostname}"`).
+- `_default_project_path()`: `~/hermes-memory/` (was `~/.basic-memory/hermes/`). The previous path violated the principle that `~/.basic-memory/` is reserved for BM's app state, not project storage.
+
+### Added
+- `_bm_known_projects()` reads bm's `~/.basic-memory/config.json`. `BasicMemoryProvider._verify_project_registered()` uses it to refuse initialization when `mode: cloud` is set against a project that isn't registered with bm. Local mode still auto-creates as before.
+- 13 new unit tests for the introspection + bail-out paths.
+
+## [0.1.1] — 2026-05-10
+
+### Added
+- `tests/test_actor.py` — 15 tests covering `_BmMcpActor` lifecycle, call dispatch, timeout-with-cancellation, idempotent shutdown.
+- `tests/test_capture.py` — 25 tests for `sync_turn` (first-write + append paths), `on_session_end` summary shape, and gating.
+- `tests/test_prefetch.py` — 25 tests for `prefetch` / `queue_prefetch` / `_format_prefetch` including forward-compat with unknown response fields.
+- `tests/test_integration.py` — 12 gated tests exercising every tool against a real `bm` MCP server (`BM_INTEGRATION=1` + `bm` + `mcp`). Each session uses a throwaway BM project that's torn down on completion.
+
+### Changed
+- `_BmMcpActor.call` now refuses calls after `shutdown()` (sets `_running=False`) and cancels the underlying coroutine on timeout instead of leaking it.
+- `_format_prefetch` defensively coerces non-string fields and skips non-dict entries.
+- Added module-level `__version__`, kept in sync with `plugin.yaml` (verified by a test).
+
+## [0.1.0] — 2026-05-10
+
+### Added
+- Initial release of the Hermes Memory Provider plugin for Basic Memory.
+- Seven `bm_*` agent tools: `bm_search`, `bm_read`, `bm_write`, `bm_edit`, `bm_context`, `bm_delete`, `bm_move`.
+- Per-turn capture (`sync_turn`) and end-of-session summary (`on_session_end`).
+- Local mode (default) with auto-created BM project; cloud mode with project-name-based routing.
+- Single-file plugin at `__init__.py`, AGPL-3.0-or-later.
+- 84-test pytest suite.
+
+[0.3.2]: https://github.com/basicmachines-co/hermes-basic-memory/releases/tag/v0.3.2
+[0.3.1]: https://github.com/basicmachines-co/hermes-basic-memory/releases/tag/v0.3.1
+[0.3.0]: https://github.com/basicmachines-co/hermes-basic-memory/releases/tag/v0.3.0
+[0.2.0]: https://github.com/basicmachines-co/hermes-basic-memory/releases/tag/v0.2.0
+[0.1.7]: https://github.com/basicmachines-co/hermes-basic-memory/releases/tag/v0.1.7
+[0.1.6]: https://github.com/basicmachines-co/hermes-basic-memory/releases/tag/v0.1.6
+[0.1.5]: https://github.com/basicmachines-co/hermes-basic-memory/releases/tag/v0.1.5
+[0.1.4]: https://github.com/basicmachines-co/hermes-basic-memory/releases/tag/v0.1.4
+[0.1.3]: https://github.com/basicmachines-co/hermes-basic-memory/releases/tag/v0.1.3
+[0.1.2]: https://github.com/basicmachines-co/hermes-basic-memory/releases/tag/v0.1.2
+[0.1.1]: https://github.com/basicmachines-co/hermes-basic-memory/releases/tag/v0.1.1
+[0.1.0]: https://github.com/basicmachines-co/hermes-basic-memory/releases/tag/v0.1.0

integrations/hermes/LICENSE

@@ -0,0 +1,661 @@
+                    GNU AFFERO GENERAL PUBLIC LICENSE
+                       Version 3, 19 November 2007
+
+ Copyright (C) 2007 Free Software Foundation, Inc. <https://fsf.org/>
+ Everyone is permitted to copy and distribute verbatim copies
+ of this license document, but changing it is not allowed.
+
+                            Preamble
+
+  The GNU Affero General Public License is a free, copyleft license for
+software and other kinds of works, specifically designed to ensure
+cooperation with the community in the case of network server software.
+
+  The licenses for most software and other practical works are designed
+to take away your freedom to share and change the works.  By contrast,
+our General Public Licenses are intended to guarantee your freedom to
+share and change all versions of a program--to make sure it remains free
+software for all its users.
+
+  When we speak of free software, we are referring to freedom, not
+price.  Our General Public Licenses are designed to make sure that you
+have the freedom to distribute copies of free software (and charge for
+them if you wish), that you receive source code or can get it if you
+want it, that you can change the software or use pieces of it in new
+free programs, and that you know you can do these things.
+
+  Developers that use our General Public Licenses protect your rights
+with two steps: (1) assert copyright on the software, and (2) offer
+you this License which gives you legal permission to copy, distribute
+and/or modify the software.
+
+  A secondary benefit of defending all users' freedom is that
+improvements made in alternate versions of the program, if they
+receive widespread use, become available for other developers to
+incorporate.  Many developers of free software are heartened and
+encouraged by the resulting cooperation.  However, in the case of
+software used on network servers, this result may fail to come about.
+The GNU General Public License permits making a modified version and
+letting the public access it on a server without ever releasing its
+source code to the public.
+
+  The GNU Affero General Public License is designed specifically to
+ensure that, in such cases, the modified source code becomes available
+to the community.  It requires the operator of a network server to
+provide the source code of the modified version running there to the
+users of that server.  Therefore, public use of a modified version, on
+a publicly accessible server, gives the public access to the source
+code of the modified version.
+
+  An older license, called the Affero General Public License and
+published by Affero, was designed to accomplish similar goals.  This is
+a different license, not a version of the Affero GPL, but Affero has
+released a new version of the Affero GPL which permits relicensing under
+this license.
+
+  The precise terms and conditions for copying, distribution and
+modification follow.
+
+                       TERMS AND CONDITIONS
+
+  0. Definitions.
+
+  "This License" refers to version 3 of the GNU Affero General Public License.
+
+  "Copyright" also means copyright-like laws that apply to other kinds of
+works, such as semiconductor masks.
+
+  "The Program" refers to any copyrightable work licensed under this
+License.  Each licensee is addressed as "you".  "Licensees" and
+"recipients" may be individuals or organizations.
+
+  To "modify" a work means to copy from or adapt all or part of the work
+in a fashion requiring copyright permission, other than the making of an
+exact copy.  The resulting work is called a "modified version" of the
+earlier work or a work "based on" the earlier work.
+
+  A "covered work" means either the unmodified Program or a work based
+on the Program.
+
+  To "propagate" a work means to do anything with it that, without
+permission, would make you directly or secondarily liable for
+infringement under applicable copyright law, except executing it on a
+computer or modifying a private copy.  Propagation includes copying,
+distribution (with or without modification), making available to the
+public, and in some countries other activities as well.
+
+  To "convey" a work means any kind of propagation that enables other
+parties to make or receive copies.  Mere interaction with a user through
+a computer network, with no transfer of a copy, is not conveying.
+
+  An interactive user interface displays "Appropriate Legal Notices"
+to the extent that it includes a convenient and prominently visible
+feature that (1) displays an appropriate copyright notice, and (2)
+tells the user that there is no warranty for the work (except to the
+extent that warranties are provided), that licensees may convey the
+work under this License, and how to view a copy of this License.  If
+the interface presents a list of user commands or options, such as a
+menu, a prominent item in the list meets this criterion.
+
+  1. Source Code.
+
+  The "source code" for a work means the preferred form of the work
+for making modifications to it.  "Object code" means any non-source
+form of a work.
+
+  A "Standard Interface" means an interface that either is an official
+standard defined by a recognized standards body, or, in the case of
+interfaces specified for a particular programming language, one that
+is widely used among developers working in that language.
+
+  The "System Libraries" of an executable work include anything, other
+than the work as a whole, that (a) is included in the normal form of
+packaging a Major Component, but which is not part of that Major
+Component, and (b) serves only to enable use of the work with that
+Major Component, or to implement a Standard Interface for which an
+implementation is available to the public in source code form.  A
+"Major Component", in this context, means a major essential component
+(kernel, window system, and so on) of the specific operating system
+(if any) on which the executable work runs, or a compiler used to
+produce the work, or an object code interpreter used to run it.
+
+  The "Corresponding Source" for a work in object code form means all
+the source code needed to generate, install, and (for an executable
+work) run the object code and to modify the work, including scripts to
+control those activities.  However, it does not include the work's
+System Libraries, or general-purpose tools or generally available free
+programs which are used unmodified in performing those activities but
+which are not part of the work.  For example, Corresponding Source
+includes interface definition files associated with source files for
+the work, and the source code for shared libraries and dynamically
+linked subprograms that the work is specifically designed to require,
+such as by intimate data communication or control flow between those
+subprograms and other parts of the work.
+
+  The Corresponding Source need not include anything that users
+can regenerate automatically from other parts of the Corresponding
+Source.
+
+  The Corresponding Source for a work in source code form is that
+same work.
+
+  2. Basic Permissions.
+
+  All rights granted under this License are granted for the term of
+copyright on the Program, and are irrevocable provided the stated
+conditions are met.  This License explicitly affirms your unlimited
+permission to run the unmodified Program.  The output from running a
+covered work is covered by this License only if the output, given its
+content, constitutes a covered work.  This License acknowledges your
+rights of fair use or other equivalent, as provided by copyright law.
+
+  You may make, run and propagate covered works that you do not
+convey, without conditions so long as your license otherwise remains
+in force.  You may convey covered works to others for the sole purpose
+of having them make modifications exclusively for you, or provide you
+with facilities for running those works, provided that you comply with
+the terms of this License in conveying all material for which you do
+not control copyright.  Those thus making or running the covered works
+for you must do so exclusively on your behalf, under your direction
+and control, on terms that prohibit them from making any copies of
+your copyrighted material outside their relationship with you.
+
+  Conveying under any other circumstances is permitted solely under
+the conditions stated below.  Sublicensing is not allowed; section 10
+makes it unnecessary.
+
+  3. Protecting Users' Legal Rights From Anti-Circumvention Law.
+
+  No covered work shall be deemed part of an effective technological
+measure under any applicable law fulfilling obligations under article
+11 of the WIPO copyright treaty adopted on 20 December 1996, or
+similar laws prohibiting or restricting circumvention of such
+measures.
+
+  When you convey a covered work, you waive any legal power to forbid
+circumvention of technological measures to the extent such circumvention
+is effected by exercising rights under this License with respect to
+the covered work, and you disclaim any intention to limit operation or
+modification of the work as a means of enforcing, against the work's
+users, your or third parties' legal rights to forbid circumvention of
+technological measures.
+
+  4. Conveying Verbatim Copies.
+
+  You may convey verbatim copies of the Program's source code as you
+receive it, in any medium, provided that you conspicuously and
+appropriately publish on each copy an appropriate copyright notice;
+keep intact all notices stating that this License and any
+non-permissive terms added in accord with section 7 apply to the code;
+keep intact all notices of the absence of any warranty; and give all
+recipients a copy of this License along with the Program.
+
+  You may charge any price or no price for each copy that you convey,
+and you may offer support or warranty protection for a fee.
+
+  5. Conveying Modified Source Versions.
+
+  You may convey a work based on the Program, or the modifications to
+produce it from the Program, in the form of source code under the
+terms of section 4, provided that you also meet all of these conditions:
+
+    a) The work must carry prominent notices stating that you modified
+    it, and giving a relevant date.
+
+    b) The work must carry prominent notices stating that it is
+    released under this License and any conditions added under section
+    7.  This requirement modifies the requirement in section 4 to
+    "keep intact all notices".
+
+    c) You must license the entire work, as a whole, under this
+    License to anyone who comes into possession of a copy.  This
+    License will therefore apply, along with any applicable section 7
+    additional terms, to the whole of the work, and all its parts,
+    regardless of how they are packaged.  This License gives no
+    permission to license the work in any other way, but it does not
+    invalidate such permission if you have separately received it.
+
+    d) If the work has interactive user interfaces, each must display
+    Appropriate Legal Notices; however, if the Program has interactive
+    interfaces that do not display Appropriate Legal Notices, your
+    work need not make them do so.
+
+  A compilation of a covered work with other separate and independent
+works, which are not by their nature extensions of the covered work,
+and which are not combined with it such as to form a larger program,
+in or on a volume of a storage or distribution medium, is called an
+"aggregate" if the compilation and its resulting copyright are not
+used to limit the access or legal rights of the compilation's users
+beyond what the individual works permit.  Inclusion of a covered work
+in an aggregate does not cause this License to apply to the other
+parts of the aggregate.
+
+  6. Conveying Non-Source Forms.
+
+  You may convey a covered work in object code form under the terms
+of sections 4 and 5, provided that you also convey the
+machine-readable Corresponding Source under the terms of this License,
+in one of these ways:
+
+    a) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by the
+    Corresponding Source fixed on a durable physical medium
+    customarily used for software interchange.
+
+    b) Convey the object code in, or embodied in, a physical product
+    (including a physical distribution medium), accompanied by a
+    written offer, valid for at least three years and valid for as
+    long as you offer spare parts or customer support for that product
+    model, to give anyone who possesses the object code either (1) a
+    copy of the Corresponding Source for all the software in the
+    product that is covered by this License, on a durable physical
+    medium customarily used for software interchange, for a price no
+    more than your reasonable cost of physically performing this
+    conveying of source, or (2) access to copy the
+    Corresponding Source from a network server at no charge.
+
+    c) Convey individual copies of the object code with a copy of the
+    written offer to provide the Corresponding Source.  This
+    alternative is allowed only occasionally and noncommercially, and
+    only if you received the object code with such an offer, in accord
+    with subsection 6b.
+
+    d) Convey the object code by offering access from a designated
+    place (gratis or for a charge), and offer equivalent access to the
+    Corresponding Source in the same way through the same place at no
+    further charge.  You need not require recipients to copy the
+    Corresponding Source along with the object code.  If the place to
+    copy the object code is a network server, the Corresponding Source
+    may be on a different server (operated by you or a third party)
+    that supports equivalent copying facilities, provided you maintain
+    clear directions next to the object code saying where to find the
+    Corresponding Source.  Regardless of what server hosts the
+    Corresponding Source, you remain obligated to ensure that it is
+    available for as long as needed to satisfy these requirements.
+
+    e) Convey the object code using peer-to-peer transmission, provided
+    you inform other peers where the object code and Corresponding
+    Source of the work are being offered to the general public at no
+    charge under subsection 6d.
+
+  A separable portion of the object code, whose source code is excluded
+from the Corresponding Source as a System Library, need not be
+included in conveying the object code work.
+
+  A "User Product" is either (1) a "consumer product", which means any
+tangible personal property which is normally used for personal, family,
+or household purposes, or (2) anything designed or sold for incorporation
+into a dwelling.  In determining whether a product is a consumer product,
+doubtful cases shall be resolved in favor of coverage.  For a particular
+product received by a particular user, "normally used" refers to a
+typical or common use of that class of product, regardless of the status
+of the particular user or of the way in which the particular user
+actually uses, or expects or is expected to use, the product.  A product
+is a consumer product regardless of whether the product has substantial
+commercial, industrial or non-consumer uses, unless such uses represent
+the only significant mode of use of the product.
+
+  "Installation Information" for a User Product means any methods,
+procedures, authorization keys, or other information required to install
+and execute modified versions of a covered work in that User Product from
+a modified version of its Corresponding Source.  The information must
+suffice to ensure that the continued functioning of the modified object
+code is in no case prevented or interfered with solely because
+modification has been made.
+
+  If you convey an object code work under this section in, or with, or
+specifically for use in, a User Product, and the conveying occurs as
+part of a transaction in which the right of possession and use of the
+User Product is transferred to the recipient in perpetuity or for a
+fixed term (regardless of how the transaction is characterized), the
+Corresponding Source conveyed under this section must be accompanied
+by the Installation Information.  But this requirement does not apply
+if neither you nor any third party retains the ability to install
+modified object code on the User Product (for example, the work has
+been installed in ROM).
+
+  The requirement to provide Installation Information does not include a
+requirement to continue to provide support service, warranty, or updates
+for a work that has been modified or installed by the recipient, or for
+the User Product in which it has been modified or installed.  Access to a
+network may be denied when the modification itself materially and
+adversely affects the operation of the network or violates the rules and
+protocols for communication across the network.
+
+  Corresponding Source conveyed, and Installation Information provided,
+in accord with this section must be in a format that is publicly
+documented (and with an implementation available to the public in
+source code form), and must require no special password or key for
+unpacking, reading or copying.
+
+  7. Additional Terms.
+
+  "Additional permissions" are terms that supplement the terms of this
+License by making exceptions from one or more of its conditions.
+Additional permissions that are applicable to the entire Program shall
+be treated as though they were included in this License, to the extent
+that they are valid under applicable law.  If additional permissions
+apply only to part of the Program, that part may be used separately
+under those permissions, but the entire Program remains governed by
+this License without regard to the additional permissions.
+
+  When you convey a copy of a covered work, you may at your option
+remove any additional permissions from that copy, or from any part of
+it.  (Additional permissions may be written to require their own
+removal in certain cases when you modify the work.)  You may place
+additional permissions on material, added by you to a covered work,
+for which you have or can give appropriate copyright permission.
+
+  Notwithstanding any other provision of this License, for material you
+add to a covered work, you may (if authorized by the copyright holders of
+that material) supplement the terms of this License with terms:
+
+    a) Disclaiming warranty or limiting liability differently from the
+    terms of sections 15 and 16 of this License; or
+
+    b) Requiring preservation of specified reasonable legal notices or
+    author attributions in that material or in the Appropriate Legal
+    Notices displayed by works containing it; or
+
+    c) Prohibiting misrepresentation of the origin of that material, or
+    requiring that modified versions of such material be marked in
+    reasonable ways as different from the original version; or
+
+    d) Limiting the use for publicity purposes of names of licensors or
+    authors of the material; or
+
+    e) Declining to grant rights under trademark law for use of some
+    trade names, trademarks, or service marks; or
+
+    f) Requiring indemnification of licensors and authors of that
+    material by anyone who conveys the material (or modified versions of
+    it) with contractual assumptions of liability to the recipient, for
+    any liability that these contractual assumptions directly impose on
+    those licensors and authors.
+
+  All other non-permissive additional terms are considered "further
+restrictions" within the meaning of section 10.  If the Program as you
+received it, or any part of it, contains a notice stating that it is
+governed by this License along with a term that is a further
+restriction, you may remove that term.  If a license document contains
+a further restriction but permits relicensing or conveying under this
+License, you may add to a covered work material governed by the terms
+of that license document, provided that the further restriction does
+not survive such relicensing or conveying.
+
+  If you add terms to a covered work in accord with this section, you
+must place, in the relevant source files, a statement of the
+additional terms that apply to those files, or a notice indicating
+where to find the applicable terms.
+
+  Additional terms, permissive or non-permissive, may be stated in the
+form of a separately written license, or stated as exceptions;
+the above requirements apply either way.
+
+  8. Termination.
+
+  You may not propagate or modify a covered work except as expressly
+provided under this License.  Any attempt otherwise to propagate or
+modify it is void, and will automatically terminate your rights under
+this License (including any patent licenses granted under the third
+paragraph of section 11).
+
+  However, if you cease all violation of this License, then your
+license from a particular copyright holder is reinstated (a)
+provisionally, unless and until the copyright holder explicitly and
+finally terminates your license, and (b) permanently, if the copyright
+holder fails to notify you of the violation by some reasonable means
+prior to 60 days after the cessation.
+
+  Moreover, your license from a particular copyright holder is
+reinstated permanently if the copyright holder notifies you of the
+violation by some reasonable means, this is the first time you have
+received notice of violation of this License (for any work) from that
+copyright holder, and you cure the violation prior to 30 days after
+your receipt of the notice.
+
+  Termination of your rights under this section does not terminate the
+licenses of parties who have received copies or rights from you under
+this License.  If your rights have been terminated and not permanently
+reinstated, you do not qualify to receive new licenses for the same
+material under section 10.
+
+  9. Acceptance Not Required for Having Copies.
+
+  You are not required to accept this License in order to receive or
+run a copy of the Program.  Ancillary propagation of a covered work
+occurring solely as a consequence of using peer-to-peer transmission
+to receive a copy likewise does not require acceptance.  However,
+nothing other than this License grants you permission to propagate or
+modify any covered work.  These actions infringe copyright if you do
+not accept this License.  Therefore, by modifying or propagating a
+covered work, you indicate your acceptance of this License to do so.
+
+  10. Automatic Licensing of Downstream Recipients.
+
+  Each time you convey a covered work, the recipient automatically
+receives a license from the original licensors, to run, modify and
+propagate that work, subject to this License.  You are not responsible
+for enforcing compliance by third parties with this License.
+
+  An "entity transaction" is a transaction transferring control of an
+organization, or substantially all assets of one, or subdividing an
+organization, or merging organizations.  If propagation of a covered
+work results from an entity transaction, each party to that
+transaction who receives a copy of the work also receives whatever
+licenses to the work the party's predecessor in interest had or could
+give under the previous paragraph, plus a right to possession of the
+Corresponding Source of the work from the predecessor in interest, if
+the predecessor has it or can get it with reasonable efforts.
+
+  You may not impose any further restrictions on the exercise of the
+rights granted or affirmed under this License.  For example, you may
+not impose a license fee, royalty, or other charge for exercise of
+rights granted under this License, and you may not initiate litigation
+(including a cross-claim or counterclaim in a lawsuit) alleging that
+any patent claim is infringed by making, using, selling, offering for
+sale, or importing the Program or any portion of it.
+
+  11. Patents.
+
+  A "contributor" is a copyright holder who authorizes use under this
+License of the Program or a work on which the Program is based.  The
+work thus licensed is called the contributor's "contributor version".
+
+  A contributor's "essential patent claims" are all patent claims
+owned or controlled by the contributor, whether already acquired or
+hereafter acquired, that would be infringed by some manner, permitted
+by this License, of making, using, or selling its contributor version,
+but do not include claims that would be infringed only as a
+consequence of further modification of the contributor version.  For
+purposes of this definition, "control" includes the right to grant
+patent sublicenses in a manner consistent with the requirements of
+this License.
+
+  Each contributor grants you a non-exclusive, worldwide, royalty-free
+patent license under the contributor's essential patent claims, to
+make, use, sell, offer for sale, import and otherwise run, modify and
+propagate the contents of its contributor version.
+
+  In the following three paragraphs, a "patent license" is any express
+agreement or commitment, however denominated, not to enforce a patent
+(such as an express permission to practice a patent or covenant not to
+sue for patent infringement).  To "grant" such a patent license to a
+party means to make such an agreement or commitment not to enforce a
+patent against the party.
+
+  If you convey a covered work, knowingly relying on a patent license,
+and the Corresponding Source of the work is not available for anyone
+to copy, free of charge and under the terms of this License, through a
+publicly available network server or other readily accessible means,
+then you must either (1) cause the Corresponding Source to be so
+available, or (2) arrange to deprive yourself of the benefit of the
+patent license for this particular work, or (3) arrange, in a manner
+consistent with the requirements of this License, to extend the patent
+license to downstream recipients.  "Knowingly relying" means you have
+actual knowledge that, but for the patent license, your conveying the
+covered work in a country, or your recipient's use of the covered work
+in a country, would infringe one or more identifiable patents in that
+country that you have reason to believe are valid.
+
+  If, pursuant to or in connection with a single transaction or
+arrangement, you convey, or propagate by procuring conveyance of, a
+covered work, and grant a patent license to some of the parties
+receiving the covered work authorizing them to use, propagate, modify
+or convey a specific copy of the covered work, then the patent license
+you grant is automatically extended to all recipients of the covered
+work and works based on it.
+
+  A patent license is "discriminatory" if it does not include within
+the scope of its coverage, prohibits the exercise of, or is
+conditioned on the non-exercise of one or more of the rights that are
+specifically granted under this License.  You may not convey a covered
+work if you are a party to an arrangement with a third party that is
+in the business of distributing software, under which you make payment
+to the third party based on the extent of your activity of conveying
+the work, and under which the third party grants, to any of the
+parties who would receive the covered work from you, a discriminatory
+patent license (a) in connection with copies of the covered work
+conveyed by you (or copies made from those copies), or (b) primarily
+for and in connection with specific products or compilations that
+contain the covered work, unless you entered into that arrangement,
+or that patent license was granted, prior to 28 March 2007.
+
+  Nothing in this License shall be construed as excluding or limiting
+any implied license or other defenses to infringement that may
+otherwise be available to you under applicable patent law.
+
+  12. No Surrender of Others' Freedom.
+
+  If conditions are imposed on you (whether by court order, agreement or
+otherwise) that contradict the conditions of this License, they do not
+excuse you from the conditions of this License.  If you cannot convey a
+covered work so as to satisfy simultaneously your obligations under this
+License and any other pertinent obligations, then as a consequence you may
+not convey it at all.  For example, if you agree to terms that obligate you
+to collect a royalty for further conveying from those to whom you convey
+the Program, the only way you could satisfy both those terms and this
+License would be to refrain entirely from conveying the Program.
+
+  13. Remote Network Interaction; Use with the GNU General Public License.
+
+  Notwithstanding any other provision of this License, if you modify the
+Program, your modified version must prominently offer all users
+interacting with it remotely through a computer network (if your version
+supports such interaction) an opportunity to receive the Corresponding
+Source of your version by providing access to the Corresponding Source
+from a network server at no charge, through some standard or customary
+means of facilitating copying of software.  This Corresponding Source
+shall include the Corresponding Source for any work covered by version 3
+of the GNU General Public License that is incorporated pursuant to the
+following paragraph.
+
+  Notwithstanding any other provision of this License, you have
+permission to link or combine any covered work with a work licensed
+under version 3 of the GNU General Public License into a single
+combined work, and to convey the resulting work.  The terms of this
+License will continue to apply to the part which is the covered work,
+but the work with which it is combined will remain governed by version
+3 of the GNU General Public License.
+
+  14. Revised Versions of this License.
+
+  The Free Software Foundation may publish revised and/or new versions of
+the GNU Affero General Public License from time to time.  Such new versions
+will be similar in spirit to the present version, but may differ in detail to
+address new problems or concerns.
+
+  Each version is given a distinguishing version number.  If the
+Program specifies that a certain numbered version of the GNU Affero General
+Public License "or any later version" applies to it, you have the
+option of following the terms and conditions either of that numbered
+version or of any later version published by the Free Software
+Foundation.  If the Program does not specify a version number of the
+GNU Affero General Public License, you may choose any version ever published
+by the Free Software Foundation.
+
+  If the Program specifies that a proxy can decide which future
+versions of the GNU Affero General Public License can be used, that proxy's
+public statement of acceptance of a version permanently authorizes you
+to choose that version for the Program.
+
+  Later license versions may give you additional or different
+permissions.  However, no additional obligations are imposed on any
+author or copyright holder as a result of your choosing to follow a
+later version.
+
+  15. Disclaimer of Warranty.
+
+  THERE IS NO WARRANTY FOR THE PROGRAM, TO THE EXTENT PERMITTED BY
+APPLICABLE LAW.  EXCEPT WHEN OTHERWISE STATED IN WRITING THE COPYRIGHT
+HOLDERS AND/OR OTHER PARTIES PROVIDE THE PROGRAM "AS IS" WITHOUT WARRANTY
+OF ANY KIND, EITHER EXPRESSED OR IMPLIED, INCLUDING, BUT NOT LIMITED TO,
+THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
+PURPOSE.  THE ENTIRE RISK AS TO THE QUALITY AND PERFORMANCE OF THE PROGRAM
+IS WITH YOU.  SHOULD THE PROGRAM PROVE DEFECTIVE, YOU ASSUME THE COST OF
+ALL NECESSARY SERVICING, REPAIR OR CORRECTION.
+
+  16. Limitation of Liability.
+
+  IN NO EVENT UNLESS REQUIRED BY APPLICABLE LAW OR AGREED TO IN WRITING
+WILL ANY COPYRIGHT HOLDER, OR ANY OTHER PARTY WHO MODIFIES AND/OR CONVEYS
+THE PROGRAM AS PERMITTED ABOVE, BE LIABLE TO YOU FOR DAMAGES, INCLUDING ANY
+GENERAL, SPECIAL, INCIDENTAL OR CONSEQUENTIAL DAMAGES ARISING OUT OF THE
+USE OR INABILITY TO USE THE PROGRAM (INCLUDING BUT NOT LIMITED TO LOSS OF
+DATA OR DATA BEING RENDERED INACCURATE OR LOSSES SUSTAINED BY YOU OR THIRD
+PARTIES OR A FAILURE OF THE PROGRAM TO OPERATE WITH ANY OTHER PROGRAMS),
+EVEN IF SUCH HOLDER OR OTHER PARTY HAS BEEN ADVISED OF THE POSSIBILITY OF
+SUCH DAMAGES.
+
+  17. Interpretation of Sections 15 and 16.
+
+  If the disclaimer of warranty and limitation of liability provided
+above cannot be given local legal effect according to their terms,
+reviewing courts shall apply local law that most closely approximates
+an absolute waiver of all civil liability in connection with the
+Program, unless a warranty or assumption of liability accompanies a
+copy of the Program in return for a fee.
+
+                     END OF TERMS AND CONDITIONS
+
+            How to Apply These Terms to Your New Programs
+
+  If you develop a new program, and you want it to be of the greatest
+possible use to the public, the best way to achieve this is to make it
+free software which everyone can redistribute and change under these terms.
+
+  To do so, attach the following notices to the program.  It is safest
+to attach them to the start of each source file to most effectively
+state the exclusion of warranty; and each file should have at least
+the "copyright" line and a pointer to where the full notice is found.
+
+    <one line to give the program's name and a brief idea of what it does.>
+    Copyright (C) <year>  <name of author>
+
+    This program is free software: you can redistribute it and/or modify
+    it under the terms of the GNU Affero General Public License as published
+    by the Free Software Foundation, either version 3 of the License, or
+    (at your option) any later version.
+
+    This program is distributed in the hope that it will be useful,
+    but WITHOUT ANY WARRANTY; without even the implied warranty of
+    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+    GNU Affero General Public License for more details.
+
+    You should have received a copy of the GNU Affero General Public License
+    along with this program.  If not, see <https://www.gnu.org/licenses/>.
+
+Also add information on how to contact you by electronic and paper mail.
+
+  If your software can interact with users remotely through a computer
+network, you should also make sure that it provides a way for users to
+get its source.  For example, if your program is a web application, its
+interface could display a "Source" link that leads users to an archive
+of the code.  There are many ways you could offer source, and different
+solutions will be better for different programs; see section 13 for the
+specific requirements.
+
+  You should also get your employer (if you work as a programmer) or school,
+if any, to sign a "copyright disclaimer" for the program, if necessary.
+For more information on this, and how to apply and follow the GNU AGPL, see
+<https://www.gnu.org/licenses/>.

integrations/hermes/MONKEYPATCH.md

@@ -0,0 +1,269 @@
+# Hermes slash-command monkeypatch
+
+This document describes the Hermes Agent-side patch needed when `/bm-*` slash commands do not appear in the gateway even though the Basic Memory provider itself is installed and working.
+
+This issue is tracked upstream in https://github.com/NousResearch/hermes-agent/issues/23603
+
+Agents may use this file as a guide to add support for `/bm-*` slash commands in Hermes Agent builds that do not yet include the upstream fix.
+
+## Current status and compatibility
+
+This is a **Hermes Agent-side** compatibility patch, not a Basic Memory plugin runtime patch.
+
+| Hermes Agent version | Basic Memory plugin version | Guidance |
+|---|---:|---|
+| `v0.13.x` | `v0.3.0` | Plugin `v0.3.0` remains the right runtime release. If `/bm-*` commands are missing, use the Hermes Agent-side monkeypatch below or wait for the upstream Hermes fix. |
+| `v0.14.0` / `v2026.5.16` | `v0.3.1` docs, runtime still equivalent to `v0.3.0` | The plugin runtime still works, but Hermes Agent `v0.14.0` still does **not** include the upstream slash-command discovery fix. Use the v0.14.0-compatible Hermes Agent-side patch below. |
+| Future Hermes release with upstream fix | Latest plugin | Do **not** apply this monkeypatch unless `/bm-*` commands are still absent; the fix should be redundant once Hermes loads active exclusive memory-provider commands during command discovery. |
+
+Checked against Hermes Agent `v2026.5.16` / `v0.14.0` on 2026-05-16: the upstream Hermes release still does **not** include this fix. After applying the Hermes Agent-side patch below locally, `get_plugin_commands()` returns the expected `/bm-*` commands.
+
+Important nuance: recent `hermes-basic-memory` versions include a best-effort PluginManager reach-in that registers commands when the provider is loaded. That workaround alone is not enough for gateway startup discovery in affected Hermes builds, because `get_plugin_commands()` does not load the active exclusive memory provider. The Hermes Agent-side patch is still needed until upstream command discovery loads the active memory provider and the memory-provider collector delegates command/skill registration.
+
+Release/tagging note for agents: `v0.3.1` is a documentation release that clarifies Hermes Agent `v0.14.0` compatibility instructions. It does not require users on Hermes Agent `v0.13.x` to change plugin runtime behavior, and it should not be interpreted as a Basic Memory data/schema migration.
+
+## Problem
+
+`hermes-basic-memory` is an **exclusive memory-provider plugin**. Hermes loads exclusive memory providers through `plugins.memory`, not through the normal `PluginManager` discovery path.
+
+Gateway adapters register native slash commands during startup by calling Hermes's plugin command discovery APIs. In affected Hermes builds, that startup path only sees commands registered by normal plugins. The active memory provider has not been loaded yet, and the memory-provider loader uses a collector that captures only `register_memory_provider(...)`. As a result, commands registered by this plugin with `ctx.register_command(...)` never reach the central plugin command registry before Discord/native slash-command sync.
+
+Symptoms:
+
+- `hermes memory status` shows `Provider: basic-memory` and `Status: available`.
+- Agent tools such as `bm_search`, `bm_read`, and `bm_recent` work.
+- Native slash commands such as `/bm-search`, `/bm-read`, and `/bm-context` are missing after `hermes gateway restart`.
+
+## Target behavior
+
+When Hermes builds its plugin command list, it should also load the configured active memory provider once, allowing that provider to register commands and skills into the same central registries used by ordinary plugins.
+
+After the patch, `get_plugin_commands()` should include commands such as:
+
+```text
+bm-context
+bm-project
+bm-read
+bm-recent
+bm-remember
+bm-search
+bm-status
+bm-workspace
+```
+
+## Files to patch in Hermes Agent
+
+Patch these files in the Hermes Agent repository, not in this plugin repository:
+
+```text
+hermes_cli/plugins.py
+plugins/memory/__init__.py
+```
+
+Recommended tests to add/update in Hermes Agent:
+
+```text
+tests/hermes_cli/test_plugin_cli_registration.py
+tests/hermes_cli/test_plugins.py
+```
+
+## Implementation outline
+
+### 1. Load active memory-provider commands from `get_plugin_commands()`
+
+In `hermes_cli/plugins.py`, add module-level idempotency/recursion guards near the global plugin manager:
+
+```python
+_plugin_manager: Optional[PluginManager] = None
+_memory_provider_command_loads: set[str] = set()
+_memory_provider_command_loading = False
+```
+
+Update `get_plugin_commands()` so it first ensures normal plugin discovery, then best-effort loads the active memory provider before returning the command registry:
+
+```python
+def get_plugin_commands() -> Dict[str, dict]:
+    """Return the full plugin commands dict (name → {handler, description, plugin}).
+
+    Triggers idempotent plugin discovery so callers can use plugin commands
+    before any explicit discover_plugins() call. Also initializes the active
+    memory provider once so exclusive memory-provider plugins can contribute
+    gateway slash commands during startup discovery.
+    """
+    manager = _ensure_plugins_discovered()
+    _ensure_active_memory_provider_commands_loaded()
+    return manager._plugin_commands
+```
+
+Add the helper:
+
+```python
+def _ensure_active_memory_provider_commands_loaded() -> None:
+    """Best-effort load of the active memory provider's slash commands."""
+    global _memory_provider_command_loading
+    if _memory_provider_command_loading:
+        return
+    try:
+        from plugins import memory as memory_plugins
+
+        active = memory_plugins._get_active_memory_provider()
+        if not active or active in _memory_provider_command_loads:
+            return
+        _memory_provider_command_loading = True
+        try:
+            memory_plugins.load_memory_provider(active)
+            _memory_provider_command_loads.add(active)
+        finally:
+            _memory_provider_command_loading = False
+    except Exception as exc:
+        logger.debug(
+            "Failed to load active memory-provider plugin commands: %s",
+            exc,
+            exc_info=_PLUGINS_DEBUG,
+        )
+```
+
+Notes:
+
+- This must be best-effort; command discovery should not break Hermes startup if a memory provider is misconfigured.
+- The recursion guard prevents `load_memory_provider(...)` → provider `register(...)` → `ctx.register_command(...)` → plugin manager access from re-entering endlessly.
+- The load set prevents duplicate provider command registration work.
+
+### 2. Make the memory-provider collector delegate commands and skills
+
+In `plugins/memory/__init__.py`, import `Callable`:
+
+```python
+from typing import Callable, List, Optional, Tuple
+```
+
+When loading a provider directory, pass the plugin/provider name to the collector:
+
+```python
+collector = _ProviderCollector(plugin_name=name)
+```
+
+Replace the collector that only captures `register_memory_provider(...)` with a plugin-context shim that also delegates `register_command(...)` and `register_skill(...)` into the central `PluginManager` registries:
+
+```python
+class _ProviderCollector:
+    """Plugin-context shim used while loading memory providers.
+
+    Memory providers are exclusive plugins and are loaded by this module
+    instead of the general PluginManager. They still need access to the same
+    slash-command and skill registries as normal plugins, otherwise active
+    memory-provider commands are invisible during gateway startup discovery.
+    """
+
+    def __init__(self, plugin_name: str = "memory-provider"):
+        self.provider = None
+        self.plugin_name = plugin_name
+
+    def register_memory_provider(self, provider):
+        self.provider = provider
+
+    def register_command(
+        self,
+        name: str,
+        handler: Callable,
+        description: str = "",
+        args_hint: str = "",
+    ) -> None:
+        """Register a memory-provider slash command with PluginManager."""
+        try:
+            from hermes_cli.plugins import _ensure_plugins_discovered
+        except Exception:
+            return
+
+        clean = name.lower().strip().lstrip("/").replace(" ", "-")
+        if not clean:
+            return
+
+        try:
+            manager = _ensure_plugins_discovered()
+        except Exception:
+            return
+
+        plugin_commands = getattr(manager, "_plugin_commands", None)
+        if plugin_commands is None:
+            return
+        plugin_commands[clean] = {
+            "handler": handler,
+            "description": description or "Plugin command",
+            "plugin": self.plugin_name,
+            "args_hint": (args_hint or "").strip(),
+        }
+
+    def register_skill(
+        self,
+        name: str,
+        path: Path,
+        description: str = "",
+    ) -> None:
+        """Register a memory-provider skill with PluginManager."""
+        try:
+            from agent.skill_utils import _NAMESPACE_RE
+            from hermes_cli.plugins import _ensure_plugins_discovered
+        except Exception:
+            return
+
+        if ":" in name or not name or not _NAMESPACE_RE.match(name):
+            raise ValueError(f"Invalid skill name '{name}'.")
+        if not path.exists():
+            raise FileNotFoundError(f"SKILL.md not found at {path}")
+
+        try:
+            manager = _ensure_plugins_discovered()
+        except Exception:
+            return
+
+        plugin_skills = getattr(manager, "_plugin_skills", None)
+        if plugin_skills is None:
+            return
+        plugin_skills[f"{self.plugin_name}:{name}"] = {
+            "path": path,
+            "plugin": self.plugin_name,
+            "bare_name": name,
+            "description": description,
+        }
+```
+
+Keep existing no-op methods such as `register_tool(...)` and `register_cli_command(...)` as no-ops unless the target Hermes version expects otherwise.
+
+## Verification
+
+From the Hermes Agent repository, run focused compile/tests:
+
+```bash
+python -m py_compile hermes_cli/plugins.py plugins/memory/__init__.py
+python -m pytest \
+  tests/hermes_cli/test_plugins.py::TestPluginCommands::test_get_plugin_commands_loads_active_memory_provider_commands \
+  tests/hermes_cli/test_plugin_cli_registration.py::TestProviderCollectorRegistration \
+  -q -o 'addopts='
+```
+
+Then verify the active config sees the Basic Memory commands:
+
+```bash
+python - <<'PY'
+import hermes_cli.plugins as p
+p._plugin_manager = None
+p._memory_provider_command_loads.clear()
+cmds = p.get_plugin_commands()
+print(sorted(k for k in cmds if k.startswith('bm-')))
+PY
+```
+
+Expected output:
+
+```text
+['bm-context', 'bm-project', 'bm-read', 'bm-recent', 'bm-remember', 'bm-search', 'bm-status', 'bm-workspace']
+```
+
+Finally restart the gateway so native slash commands are synced:
+
+```bash
+hermes gateway restart
+```
+
+For Discord, global command propagation can lag briefly. If the commands do not show immediately, type `/bm` directly or reload the Discord client.

integrations/hermes/README.md

@@ -0,0 +1,235 @@
+# hermes-basic-memory
+
+[![License: AGPL v3](https://img.shields.io/badge/License-AGPL_v3-blue.svg)](https://www.gnu.org/licenses/agpl-3.0)
+
+Hermes Memory Provider plugin that gives [Hermes Agent](https://github.com/NousResearch/hermes-agent) a persistent knowledge graph backed by [Basic Memory](https://github.com/basicmachines-co/basic-memory).
+
+The plugin replaces Hermes's "no external memory provider" with a real graph: search-before-answer recall, per-turn capture, end-of-session summaries, and ten `bm_*` tools the agent can call directly. Local mode by default; one CLI flip switches to true cloud routing through Basic Memory Cloud.
+
+## Install
+
+```bash
+hermes plugins install basicmachines-co/basic-memory --path integrations/hermes
+```
+
+Then activate it in `~/.hermes/config.yaml`:
+
+```yaml
+memory:
+  provider: basic-memory
+```
+
+If you run the gateway, restart it (`hermes gateway restart`). Done.
+
+If your installed Hermes build does not support `--path`, use the final deprecated `basicmachines-co/hermes-basic-memory` pointer release until Hermes subpath installs are available. Ongoing development now lives in [`basic-memory/integrations/hermes`](https://github.com/basicmachines-co/basic-memory/tree/main/integrations/hermes).
+
+The plugin self-installs the `basic-memory` CLI on first init via `uv tool install basic-memory` (one-time ~10s pause if it isn't already present). The bm binary lands at `~/.local/bin/bm` — the same location a manual `uv tool install basic-memory` would produce, so a later manual install or upgrade is a no-op rather than a second install.
+
+### Prerequisites
+
+- [Hermes Agent](https://github.com/NousResearch/hermes-agent)
+- [`uv`](https://docs.astral.sh/uv/) on PATH (used for the bootstrap install)
+- The `mcp` Python package in the Hermes venv. If `hermes plugins install` doesn't auto-install it (it follows `pip_dependencies` in `plugin.yaml`), run:
+  ```bash
+  uv pip install --python ~/.hermes/hermes-agent/venv/bin/python mcp
+  ```
+
+### Verify
+
+```bash
+hermes memory status
+```
+
+Expected:
+```
+  Provider:  basic-memory
+  Plugin:    installed ✓
+  Status:    available ✓
+```
+
+## What the agent gets
+
+Ten tools (curated subset of Basic Memory's MCP surface):
+
+| Tool | Use |
+|---|---|
+| `bm_search` | Semantic + full-text search; **call this before answering** |
+| `bm_read` | Fetch a note by title, permalink, or `memory://` URL |
+| `bm_write` | Create a new note (capture decisions, meeting notes, insights) |
+| `bm_edit` | Append, prepend, find/replace, replace-section |
+| `bm_context` | Navigate via `memory://` URLs to find related notes |
+| `bm_delete` | Delete a note |
+| `bm_move` | Move a note to a different folder |
+| `bm_recent` | List notes updated recently (default `7d`; accepts natural-language timeframes) |
+| `bm_projects` | List available projects with their UUIDs (for cross-project routing) |
+| `bm_workspaces` | List Basic Memory Cloud workspaces |
+
+Every read/write tool also accepts optional `project` / `project_id` for per-call routing — write or read against a project other than the configured one without reconfiguring the plugin.
+
+Plus automatic capture:
+- **Per turn**: every user/assistant exchange appends to a running session-transcript note
+- **End of session**: a separate summary note is written, linked back to the transcript via a `summary_of` relation
+
+A bundled skill (`skill:view basic-memory:basic-memory`) gives the agent a longer reference doc on top of the always-on `system_prompt_block`.
+
+## Slash commands
+
+For direct, in-session use without going through the agent (requires Hermes ≥ v0.11.0):
+
+| Command | Use |
+|---|---|
+| `/bm-search <query>` | Search the knowledge graph; returns compact title/permalink/preview rows. |
+| `/bm-read <identifier>` | Read a note by title, permalink, or `memory://` URL. |
+| `/bm-context <identifier>` | Build context for a note (target + related). |
+| `/bm-recent [timeframe]` | Recently updated notes. Default `7d`; accepts `"2 weeks"`, `"yesterday"`, etc. |
+| `/bm-status` | Plugin/provider state: mode, project, capture flags, bm CLI path. |
+| `/bm-remember <text>` | Capture a quick note. Title = first line (≤80 chars), folder = `remember_folder` (default `bm-remember`), tagged `manual-capture`. |
+| `/bm-project` | List all known projects; the active one is marked. |
+| `/bm-workspace` | List BM Cloud workspaces. Cloud mode only — prints an explanatory line in local mode. |
+
+Examples:
+
+```text
+/bm-search Q3 OKRs
+/bm-read decisions/auth-rewrite
+/bm-recent yesterday
+/bm-remember Reminder: switch the staging job to the new image after the rebase lands.
+```
+
+`/bm-project` and `/bm-workspace` are read-only in 0.2.0 — mid-session switching is intentionally not supported because auto-capture would otherwise land in the wrong place. Tracked as a follow-up.
+
+### Known issue: `/bm-*` commands may not appear in some Hermes gateway builds
+
+Plugin v0.2.0 and later register the commands above, but some Hermes Agent gateway builds do not discover slash commands contributed by an **exclusive memory-provider plugin** during startup. The symptoms are:
+
+- the memory tools work for the agent (`bm_search`, `bm_read`, etc.);
+- `hermes memory status` shows `Provider: basic-memory` and `Status: available`; but
+- Discord/native slash command pickers do not show `/bm-search`, `/bm-read`, `/bm-context`, and the other `/bm-*` commands after `hermes gateway restart`.
+
+This is a Hermes Agent plugin-discovery issue, not a Basic Memory runtime issue. It is tracked upstream in [NousResearch/hermes-agent#23603](https://github.com/NousResearch/hermes-agent/issues/23603). Updating the Basic Memory plugin alone cannot fix affected gateway startup discovery; Hermes Agent itself must include or receive the compatibility patch. Until the upstream Hermes fix is available in your installed Hermes version, use one of these workarounds:
+
+1. apply the Hermes Agent-side patch described in [MONKEYPATCH.md](MONKEYPATCH.md), which includes compatibility notes for Hermes Agent v0.13.x and v0.14.0; or
+2. use the agent tools directly (`bm_search`, `bm_read`, `bm_recent`, etc.) instead of native slash commands.
+
+After applying an updated or patched Hermes build, restart the gateway so Discord/native slash commands are re-synced:
+
+```bash
+hermes gateway restart
+```
+
+If Discord still does not show the commands immediately, type `/bm` directly or reload the Discord client; global command propagation can lag briefly.
+
+## Configuration
+
+Defaults are reasonable for local use:
+
+| Key | Default | Notes |
+|---|---|---|
+| `mode` | `local` | `local` (in-process) or `cloud` (route through BM Cloud API) |
+| `project` | `hermes-memory` | BM project name |
+| `project_path` | `~/hermes-memory/` | Local mode only — where session notes land |
+| `capture_folder` | `hermes-sessions` | Folder within the project for session notes |
+| `capture_per_turn` | `true` | Append every turn to a session transcript |
+| `capture_session_end` | `true` | Write a summary note when the session ends |
+| `remember_folder` | `bm-remember` | Folder where `/bm-remember` captures land (kept separate from session transcripts) |
+
+To override, write `~/.hermes/basic-memory.json` or run `hermes memory setup basic-memory`:
+
+```json
+{
+  "mode": "local",
+  "project": "hermes-memory",
+  "project_path": "~/hermes-memory/",
+  "capture_per_turn": true,
+  "capture_session_end": true,
+  "capture_folder": "hermes-sessions",
+  "remember_folder": "bm-remember"
+}
+```
+
+In local mode the plugin auto-creates the BM project on first init via `bm project add`. In cloud mode it doesn't — you create the cloud-routed project yourself (see below) and the plugin verifies it's registered before initializing.
+
+### Cloud mode
+
+When `mode: cloud`, tool calls route directly through the BM cloud API — no local file mirror, no bisync. You set this up once with the BM CLI:
+
+```bash
+# Authenticate (OAuth) or save an API key
+bm cloud login                     # OAuth — interactive
+# OR for headless/automation:
+bm cloud create-key "hermes"
+bm cloud set-key bmc_...
+
+# Create the project, then flip it to cloud routing.
+# --workspace is required if you belong to more than one workspace
+# (otherwise BM auto-resolves the only one available).
+bm project add hermes-memory-cloud
+bm project set-cloud hermes-memory-cloud --workspace Personal
+
+# Point the plugin at it
+cat > ~/.hermes/basic-memory.json <<EOF
+{
+  "mode": "cloud",
+  "project": "hermes-memory-cloud",
+  "capture_per_turn": true,
+  "capture_session_end": true,
+  "capture_folder": "hermes-sessions"
+}
+EOF
+
+hermes gateway restart
+```
+
+Tool calls now route from `bm mcp` → `<cloud_host>/proxy` over HTTPS using your OAuth token (or API key). Notes never touch local disk.
+
+**Don't confuse cloud mode with `bm cloud bisync`.** Bisync is rclone-style two-way file sync between a *local* project and cloud storage, intended for keeping local working copies. For agent-driven capture you want true cloud routing (`set-cloud`), not bisync.
+
+## Updating / removing
+
+```bash
+hermes plugins update basic-memory
+hermes plugins remove basic-memory     # then revert memory.provider in config.yaml
+```
+
+## Foot-guns
+
+- **`<memory-context>` tags in notes**: Hermes's streaming output scrubber strips literal `<memory-context>...</memory-context>` blocks from assistant text. If a note contains those tags and the assistant echoes the body verbatim, the echoed copy gets eaten mid-stream. Tool results inbound are unaffected. Avoid those tags in BM notes; if you must include them, fence in a code block.
+- **Single external provider**: Hermes accepts only one external memory provider at a time. Activating basic-memory displaces any other.
+- **CLI cold start**: `hermes -z ...` invocations spawn `bm mcp` per run (~2-5s). Long-running gateway sessions amortize this.
+- **Multiple cloud workspaces**: if your BM Cloud account belongs to more than one workspace, `bm project set-cloud` must be invoked with `--workspace <name>`. Otherwise tool calls fail with "Multiple workspaces are available".
+
+## Development
+
+The plugin is a single-file Python module at `__init__.py`. The Hermes plugin loader expects `register(ctx)` and grep-detects either `register_memory_provider` or `MemoryProvider` in the file.
+
+For local development (point Hermes at your working tree instead of going through `hermes plugins install`):
+
+```bash
+git clone https://github.com/basicmachines-co/basic-memory ~/code/basic-memory
+mkdir -p ~/.hermes/plugins
+ln -snf ~/code/basic-memory/integrations/hermes ~/.hermes/plugins/basic-memory
+```
+
+### Running tests
+
+```bash
+# From the monorepo root
+just package-check-hermes
+
+# Or from integrations/hermes
+just check
+
+# Unit tests (fast, hermetic — no Hermes or bm required)
+uv run --with pytest pytest
+
+# Integration tests (gated — exercise every tool against a real bm MCP server)
+BM_INTEGRATION=1 uv run --with pytest --with mcp pytest tests/test_integration.py
+```
+
+The unit suite stubs out Hermes-internal imports (`agent.memory_provider`, `tools.registry`) so it runs without a Hermes install. `mcp` is optional at unit-test time — its absence just makes `is_available()` return False, which the tests verify.
+
+Integration tests require `BM_INTEGRATION=1`, `bm` CLI on PATH, and `mcp` Python package importable. Each session creates a unique throwaway BM project (under `tempfile.mkdtemp`) and removes it on teardown, so they never touch your real BM projects.
+
+## License
+
+AGPL-3.0-or-later, matching [basic-memory](https://github.com/basicmachines-co/basic-memory). See [LICENSE](LICENSE).

integrations/hermes/justfile

@@ -0,0 +1,22 @@
+# Basic Memory Hermes plugin checks
+
+repo_root := "../.."
+
+# Validate plugin.yaml, module entrypoint, bundled skill, and test layout.
+manifest-check:
+    python3 {{repo_root}}/scripts/validate_hermes_plugin.py .
+
+# Unit tests are hermetic and do not require a Hermes install.
+test:
+    uv run --no-project --with pytest --with pytest-cov --python 3.12 pytest -q --cov=. --cov-report=term-missing
+
+# Gated integration test against a real bm MCP server.
+test-int:
+    BM_INTEGRATION=1 uv run --no-project --with pytest --with mcp --python 3.12 pytest tests/test_integration.py -q
+
+# Full local check.
+check: manifest-check test
+
+# Show available recipes
+default:
+    @just --list

integrations/hermes/plugin.yaml

@@ -0,0 +1,11 @@
+name: basic-memory
+version: 0.22.0
+description: "Basic Memory — persistent knowledge graph backed by the basic-memory MCP server"
+pip_dependencies:
+  - mcp
+hooks:
+  - prefetch
+  - queue_prefetch
+  - sync_turn
+  - on_session_end
+  - shutdown

integrations/hermes/pytest.ini

@@ -0,0 +1,8 @@
+[pytest]
+testpaths = tests
+addopts = -ra
+pythonpath =
+    .
+    tests/stubs
+filterwarnings =
+    ignore::DeprecationWarning

integrations/hermes/requirements-dev.txt

@@ -0,0 +1 @@
+pytest>=7.0

integrations/hermes/skill/SKILL.md

@@ -0,0 +1,242 @@
+---
+name: basic-memory
+description: Use the Basic Memory knowledge graph for persistent memory across sessions. Search before answering; capture decisions, meetings, and insights as notes.
+category: memory
+---
+
+# Basic Memory Knowledge Graph
+
+You have access to a persistent knowledge graph backed by Basic Memory. The graph survives across sessions and is shared with other tools (Claude Desktop, Obsidian, the `bm` CLI). Use the `bm_*` tools below to recall and capture information.
+
+## Use `bm_*`, not the `bm` CLI
+
+**Always invoke the `bm_*` tools directly. Do not shell out to the `bm` CLI for note operations.**
+
+The `bm_*` tools route through a persistent MCP connection — roughly 0.1 seconds per call. Running `bm` from the shell spawns a fresh Python process per call (1-2 seconds of cold-start every time) and bypasses Hermes's automatic per-turn capture, so the session-transcript and summary notes won't reflect what you did.
+
+The CLI is fine when you genuinely need a feature these wrappers don't expose (rare). Otherwise, prefer:
+
+| Use case | Tool (not CLI) |
+|---|---|
+| Search the graph | `bm_search` |
+| Read a note | `bm_read` |
+| Create / update a note | `bm_write` / `bm_edit` |
+| Navigate relations | `bm_context` |
+| Move / delete | `bm_move` / `bm_delete` |
+| What's been touched lately | `bm_recent` |
+| List available projects | `bm_projects` |
+| List cloud workspaces | `bm_workspaces` |
+
+## Tool reference
+
+### `bm_search` — search the graph
+Use **before** answering questions about prior decisions, projects, meetings, or anything that might already be documented.
+
+```
+bm_search({ query: "auth strategy decision", limit: 5 })
+```
+
+### `bm_read` — fetch a note's full content
+After search shows a relevant note, read it for context.
+
+```
+bm_read({ identifier: "decisions/auth-strategy" })
+bm_read({ identifier: "memory://projects/api-redesign" })
+```
+
+### `bm_context` — navigate via memory:// URLs
+Returns the target note plus related notes via traversed relations.
+
+```
+bm_context({ url: "memory://projects/api-redesign", depth: 1 })
+```
+
+### `bm_write` — capture new knowledge
+When the user shares a decision, meeting outcome, or insight worth keeping, capture it. Use clear titles and a folder.
+
+```
+bm_write({
+  title: "API Authentication Decision",
+  folder: "decisions",
+  content: "# API Authentication\n\n## Context\n...\n\n## Decision\n..."
+})
+```
+
+Recommended folders: `projects/`, `decisions/`, `meetings/`, `concepts/`, `weekly/`.
+
+### `bm_edit` — incremental updates
+Operations: `append`, `prepend`, `find_replace` (requires `find_text`), `replace_section` (requires `section`).
+
+```
+bm_edit({
+  identifier: "projects/api-redesign",
+  operation: "append",
+  content: "\n## Update 2026-05-09\nDeployed to staging."
+})
+```
+
+### `bm_delete` / `bm_move` — maintenance
+Use sparingly. `bm_move` takes `new_folder`.
+
+### `bm_recent` — what's been touched lately
+Returns notes updated within a window. Use when there's no specific query yet — e.g. "what was I working on yesterday?"
+
+```
+bm_recent({ timeframe: "7d" })
+bm_recent({ timeframe: "yesterday", limit: 20 })
+bm_recent({ timeframe: "2 weeks", type: "entity" })
+```
+
+`timeframe` accepts natural language (`"yesterday"`, `"2 weeks"`, `"last month"`) or compact forms (`"7d"`, `"24h"`). Default is `7d`.
+
+### `bm_projects` — list available projects
+Returns name, workspace slug, and `external_id` (UUID) per project across local and cloud. Call this when the user names a project that isn't the active one. Route follow-up tool calls either by workspace-qualified name (`project: "personal/main"`) or by UUID (`project_id: "bf2a4c1e-d77f-..."`) — see Cross-project routing below.
+
+```
+bm_projects()
+```
+
+### `bm_workspaces` — list BM Cloud workspaces
+Workspaces are a BM Cloud concept. Returns name, type, role, and default flag. Pair with `bm_projects` when the same project name might exist in more than one workspace and you need to disambiguate.
+
+```
+bm_workspaces()
+```
+
+## Permalinks
+
+A permalink is the canonical, URL-friendly identifier for a note. Three shapes exist; the read/write tools accept all of them:
+
+| Shape | Example | When |
+|---|---|---|
+| **Short** | `decisions/auth-strategy` | Bare `folder/note-slug`. Tools need a `project` (or `project_id`) arg to route — the permalink alone isn't enough. |
+| **Project-qualified** | `main/decisions/auth-strategy` | `project-name/folder/note-slug`. Carries enough context to route without a separate `project` arg. |
+| **Workspace-qualified** | `personal/main/decisions/auth-strategy` | `workspace-slug/project-name/folder/note-slug`. Fully routes, including across cloud workspaces with same-named projects. |
+
+**Important: the permalink returned by `bm_write` already encodes the routing it needs for follow-up reads.** If you wrote with `project="personal/main"`, you get back `personal/main/folder/note-slug` and can call `bm_read({ identifier: <that permalink> })` with no `project` arg. The permalink self-routes.
+
+`memory://` URLs follow the same shapes: `memory://personal/main/decisions/auth-strategy` is valid. The `memory://` prefix is optional for `bm_read` (any of the three permalink shapes works directly); `bm_context` expects the prefix.
+
+## Cross-project routing
+
+Every read/write tool (`bm_search`, `bm_read`, `bm_write`, `bm_edit`, `bm_context`, `bm_delete`, `bm_move`, `bm_recent`) accepts optional `project` and `project_id`:
+
+- `project` — project name, optionally workspace-qualified. Plain (`"main"`) when the name is globally unique; qualified (`"personal/main"`, `"team-paul/research"`) when you need to pick a specific cloud workspace by slug.
+- `project_id` — UUID from `bm_projects` (`external_id` field). The most stable identifier — survives project renames and works across workspaces without qualification. Wins over `project` if both are passed.
+
+Omit both and the call uses the Hermes-configured active project.
+
+```
+# Plain project name (unique)
+bm_write({ title: "...", folder: "...", content: "...", project: "main" })
+
+# Workspace-qualified name (disambiguates same-named projects across workspaces)
+bm_write({ title: "...", folder: "...", content: "...", project: "personal/main" })
+
+# UUID (most stable, survives renames)
+bm_write({ title: "...", folder: "...", content: "...", project_id: "bf2a4c1e-d77f-..." })
+```
+
+`bm_projects` and `bm_workspaces` themselves do **not** take routing — they list across everything.
+
+## Recipe: writing an existing file into a specific project
+
+When the user asks something like *"save this markdown file to my personal `main` project, return the permalink"*:
+
+1. **Discover the project.** Call `bm_projects()` and find the entry matching the user's described project + workspace. You can route by either the workspace-qualified name (`personal/main`) or the UUID (`external_id`).
+
+   ```
+   bm_projects()
+   # → [{name: "main", external_id: "bf2a4c1e-d77f-4b7a-9c3e-5d8a1f0e2b6d", workspace: "Personal", ...}, ...]
+   ```
+
+   If a project name appears in multiple workspaces, use `bm_workspaces()` to confirm which slug you want.
+
+2. **Read the file from disk.** Use Hermes's filesystem tool (not a `bm_*` tool — local files aren't in the graph yet).
+
+3. **Write the note with explicit routing.** Either form works; the workspace-qualified name reads cleaner in logs, the UUID is more durable.
+
+   ```
+   bm_write({
+     title: "StartWithDrew Level 9 Task Queue",
+     folder: "startwithdrew",
+     content: <file body>,
+     project: "personal/main"
+   })
+   # → returns "personal/main/startwithdrew/start-with-drew-level-9-task-queue"
+   # (the returned permalink is workspace-qualified — carries its own routing)
+   ```
+
+4. **Verify by reading back.** No `project` arg needed — the workspace-qualified permalink routes itself.
+
+   ```
+   bm_read({ identifier: "personal/main/startwithdrew/start-with-drew-level-9-task-queue" })
+   ```
+
+Return the permalink (and the project name for clarity) to the user.
+
+## When to use each tool
+
+| Situation | Tool |
+|---|---|
+| User asks about a topic that might already be documented | `bm_search` first, then `bm_read` |
+| User exposes a decision, plan, or meeting outcome | offer to `bm_write` |
+| Updating prior work | `bm_edit` (append for time-ordered logs, replace_section for living docs) |
+| Exploring related concepts | `bm_context` |
+| "What was I working on yesterday?" / no specific query yet | `bm_recent` |
+| User names a project that isn't the active one | `bm_projects` → call read/write tool with `project: "workspace/name"` or `project_id: "<uuid>"` |
+| Same project name might exist in multiple workspaces | `bm_projects` (+ `bm_workspaces` if needed) → route with workspace-qualified `project` or `project_id` |
+| Following up on a freshly-written note | Use the returned permalink directly — it already encodes the routing |
+
+## Note structure
+
+BM treats `- [category]` lines as **observations** and WikiLink lines under `## Relations` as **relations**. Categories (`[decision]`, `[insight]`, `[risk]`, `[fact]`, `[todo]`, …) and relation types (`relates_to`, `implements`, `depends_on`, `blocks`, …) are open-ended — use what fits the content. YAML frontmatter is supported with `title`, `type`, `tags`, and `permalink` as standard fields; any custom fields are allowed. See the [knowledge format docs](https://docs.basicmemory.com/raw/concepts/knowledge-format.md) for the full convention.
+
+```markdown
+# Clear Title
+
+## Context
+Background and current situation.
+
+## Key Points
+- Main insights
+- Important details
+
+## Observations
+- [decision] We chose PostgreSQL for ACID guarantees
+- [insight] Users prefer social login
+- [risk] Deployment lacks rollback path
+
+## Relations
+- relates_to [[Other Note Title]]
+- depends_on [[Database Choice]]
+
+## Next Steps
+- [ ] Implement
+- [ ] Document
+```
+
+## Behavior guidelines
+
+1. **Search before answering.** If the user asks "what did we decide about X?", run `bm_search` first.
+2. **Offer to capture.** When the user shares decisions or meeting outcomes, ask: "Should I save this as a note?"
+3. **Suggest connections.** When a search returns related notes, surface them so the user knows what already exists.
+4. **Don't over-capture.** Auto-capture is already running per turn. Don't create a `bm_write` for every response — only for substantive content the user wants preserved.
+5. **Sensitive info.** Don't capture credentials or personal data without confirmation.
+
+## Footgun
+
+If a note's body contains literal `<memory-context>...</memory-context>` tags, Hermes's streaming output scrubber will eat those tags (and the text between paired ones) when you echo the note verbatim back to the user. Tool *inputs* are unaffected. If you must include such content, fence it in a code block.
+
+## Further reading
+
+Official docs live at [docs.basicmemory.com](https://docs.basicmemory.com). Every page has an AI-friendly raw markdown view at `/raw/<path>.md` (or send `Accept: text/markdown` to the canonical URL). `WebFetch` any of these when you need detail beyond what this skill covers:
+
+- **[Knowledge format](https://docs.basicmemory.com/raw/concepts/knowledge-format.md)** — observation categories, relation types, frontmatter conventions.
+- **[Observations & relations](https://docs.basicmemory.com/raw/concepts/observations-and-relations.md)** — how notes form a graph that's searchable and traversable.
+- **[Memory URLs](https://docs.basicmemory.com/raw/concepts/memory-urls.md)** — title-based addressing, wildcards (`memory://docs/*`), and routing resolution order.
+- **[Projects & folders](https://docs.basicmemory.com/raw/concepts/projects-and-folders.md)** — multi-project layout, folder organization, cloud routing behavior.
+- **[Semantic search](https://docs.basicmemory.com/raw/concepts/semantic-search.md)** — how `bm_search` resolves queries (semantic + full-text).
+- **[MCP tools reference](https://docs.basicmemory.com/raw/reference/mcp-tools-reference.md)** — Basic Memory's full MCP surface (the `bm_*` tools here are a curated subset).
+- **[Cloud routing](https://docs.basicmemory.com/raw/cloud/routing.md)** — local vs cloud project modes, per-project routing setup.
+- **[llms.txt index](https://docs.basicmemory.com/llms.txt)** — full sitemap of raw markdown pages, useful when you need to look up a page not listed above.

integrations/hermes/tests/conftest.py

@@ -0,0 +1,167 @@
+"""
+Pytest configuration: stub Hermes-internal imports so the plugin loads
+without a Hermes install, and expose the loaded plugin module as a fixture.
+"""
+
+from __future__ import annotations
+
+import importlib.util
+import json
+import os
+import sys
+import types
+
+import pytest
+
+
+def _stub_hermes_modules() -> None:
+    """
+    The plugin imports `agent.memory_provider.MemoryProvider` and
+    `tools.registry.tool_error`. These are provided by Hermes at runtime,
+    not as a pip-installable package. Stub them before module load so unit
+    tests don't need Hermes installed.
+    """
+    agent_mod = types.ModuleType("agent")
+    agent_mp_mod = types.ModuleType("agent.memory_provider")
+
+    class _StubMemoryProvider:
+        """Stand-in for `agent.memory_provider.MemoryProvider`."""
+
+    agent_mp_mod.MemoryProvider = _StubMemoryProvider
+    sys.modules.setdefault("agent", agent_mod)
+    sys.modules.setdefault("agent.memory_provider", agent_mp_mod)
+
+    tools_mod = types.ModuleType("tools")
+    tools_registry_mod = types.ModuleType("tools.registry")
+
+    def _tool_error(msg: str) -> str:
+        return json.dumps({"error": str(msg)})
+
+    tools_registry_mod.tool_error = _tool_error
+    sys.modules.setdefault("tools", tools_mod)
+    sys.modules.setdefault("tools.registry", tools_registry_mod)
+
+
+_stub_hermes_modules()
+
+
+_PLUGIN_PATH = os.path.join(os.path.dirname(os.path.dirname(__file__)), "__init__.py")
+_spec = importlib.util.spec_from_file_location("hermes_basic_memory_plugin", _PLUGIN_PATH)
+assert _spec is not None and _spec.loader is not None
+_plugin = importlib.util.module_from_spec(_spec)
+_spec.loader.exec_module(_plugin)
+
+
+@pytest.fixture
+def bm():
+    """The loaded plugin module."""
+    return _plugin
+
+
+# ---- Synthetic MCP CallToolResult objects (no MCP SDK needed at test time) ----
+
+
+class FakeContent:
+    """Stand-in for `mcp.types.TextContent`."""
+
+    def __init__(self, text: str):
+        self.text = text
+
+
+class FakeCallToolResult:
+    """Stand-in for `mcp.types.CallToolResult`."""
+
+    def __init__(self, content_texts, is_error: bool = False):
+        self.content = [FakeContent(t) for t in content_texts]
+        self.isError = is_error
+
+
+@pytest.fixture
+def fake_result():
+    return FakeCallToolResult
+
+
+# ---- Actor test helpers ----
+
+
+class FakeSession:
+    """
+    Stand-in for `mcp.ClientSession`. `call_tool` is an async coroutine
+    that records every call and returns a configurable CallToolResult.
+
+    - default_response: dict serialized to JSON in the result text
+    - hang: when True, call_tool sleeps long enough to force a client-side timeout
+    - is_error: when True, return value has isError=True
+    """
+
+    def __init__(self, default_response=None, hang: bool = False, is_error: bool = False):
+        self.default_response = default_response if default_response is not None else {"ok": True}
+        self.hang = hang
+        self.is_error = is_error
+        self.calls: list = []
+        self.was_cancelled = False
+        self._stub_handlers: dict = {}
+
+    def stub(self, tool_name: str, handler):
+        """Register a per-tool handler. Handler receives (args) and returns a response dict (or raises)."""
+        self._stub_handlers[tool_name] = handler
+
+    async def call_tool(self, name: str, args: dict):
+        self.calls.append((name, dict(args)))
+        if self.hang:
+            try:
+                import asyncio as _a
+
+                await _a.sleep(60)
+            except BaseException as e:
+                # Track cooperative cancellation so tests can verify cleanup
+                if type(e).__name__ in ("CancelledError",):
+                    self.was_cancelled = True
+                raise
+        if name in self._stub_handlers:
+            response = self._stub_handlers[name](args)
+        else:
+            response = self.default_response
+        return FakeCallToolResult([json.dumps(response)], is_error=self.is_error)
+
+
+def make_scripted_actor(
+    bm, session=None, raise_at_init: BaseException | None = None, fake_tools: list | None = None
+):
+    """
+    Build an actor whose `_main()` is replaced with a deterministic version.
+    The fake `_main` mirrors the production happy path (sets _session, _stop_future,
+    fills _tools_cache, signals ready, awaits stop) but skips the stdio subprocess.
+
+    raise_at_init: if set, raised inside _main so we can test failure paths.
+    """
+    import asyncio
+
+    actor = bm._BmMcpActor(["fake-bm", "mcp"])
+    sess = session or FakeSession()
+
+    async def _fake_main():
+        if raise_at_init is not None:
+            actor._init_error = raise_at_init
+            actor._ready.set()
+            raise raise_at_init
+        actor._session = sess
+        actor._stop_future = asyncio.get_running_loop().create_future()
+        actor._tools_cache = fake_tools or [
+            {"name": n, "description": ""}
+            for n in [
+                "search_notes",
+                "read_note",
+                "write_note",
+                "edit_note",
+                "build_context",
+                "delete_note",
+                "move_note",
+            ]
+        ]
+        actor._ready.set()
+        await actor._stop_future
+
+    actor._main = _fake_main  # type: ignore[assignment]
+    actor._test_session = sess  # convenience handle for assertions
+    return actor

integrations/hermes/tests/stubs/agent/__init__.py

@@ -0,0 +1 @@
+"""Test-only Hermes agent package stub."""

integrations/hermes/tests/stubs/agent/memory_provider.py

@@ -0,0 +1,5 @@
+"""Test-only stand-in for Hermes's memory provider interface."""
+
+
+class MemoryProvider:
+    """Minimal base class used by the plugin during unit tests."""

integrations/hermes/tests/stubs/tools/__init__.py

@@ -0,0 +1 @@
+"""Test-only Hermes tools package stub."""

integrations/hermes/tests/stubs/tools/registry.py

@@ -0,0 +1,9 @@
+"""Test-only stand-in for Hermes's tool registry helpers."""
+
+from __future__ import annotations
+
+import json
+
+
+def tool_error(msg: str) -> str:
+    return json.dumps({"error": str(msg)})

integrations/hermes/tests/test_actor.py

@@ -0,0 +1,183 @@
+"""Tests for _BmMcpActor: lifecycle, call dispatch, timeout, shutdown."""
+
+from __future__ import annotations
+
+import concurrent.futures
+import json
+import time
+
+import pytest
+
+from tests.conftest import FakeSession, make_scripted_actor
+
+
+# ---- Lifecycle ----
+
+
+def test_new_actor_is_not_running(bm):
+    actor = bm._BmMcpActor(["fake-bm", "mcp"])
+    assert actor._running is False
+
+
+def test_start_brings_actor_up(bm):
+    actor = make_scripted_actor(bm)
+    actor.start(timeout=5.0)
+    try:
+        assert actor._running is True
+        assert actor._session is not None
+        assert actor._stop_future is not None
+        assert actor._thread is not None and actor._thread.is_alive()
+        # Tools cache populated
+        names = {t["name"] for t in actor.list_tools()}
+        assert "search_notes" in names
+    finally:
+        actor.shutdown(timeout=2.0)
+
+
+def test_start_is_idempotent_when_thread_alive(bm):
+    actor = make_scripted_actor(bm)
+    actor.start(timeout=5.0)
+    try:
+        first_thread = actor._thread
+        actor.start(timeout=5.0)  # second call
+        assert actor._thread is first_thread  # same thread, no new one spawned
+    finally:
+        actor.shutdown(timeout=2.0)
+
+
+def test_start_init_error_raises(bm):
+    boom = ValueError("BM unreachable")
+    actor = make_scripted_actor(bm, raise_at_init=boom)
+    with pytest.raises(RuntimeError, match="BM unreachable"):
+        actor.start(timeout=5.0)
+    assert actor._running is False
+
+
+def test_start_timeout_raises(bm):
+    """If _ready never gets set within the timeout, start() raises TimeoutError."""
+    import asyncio
+
+    actor = bm._BmMcpActor(["fake-bm", "mcp"])
+
+    async def _hang_forever():
+        # Never call self._ready.set(); start() should time out waiting.
+        await asyncio.sleep(60)
+
+    actor._main = _hang_forever  # type: ignore[assignment]
+
+    with pytest.raises(TimeoutError):
+        actor.start(timeout=0.5)
+    assert actor._running is False
+
+
+# ---- Shutdown ----
+
+
+def test_shutdown_before_start_is_noop(bm):
+    actor = bm._BmMcpActor(["fake-bm", "mcp"])
+    # Should not raise even though nothing is running
+    actor.shutdown(timeout=1.0)
+    assert actor._running is False
+
+
+def test_shutdown_stops_running_actor(bm):
+    actor = make_scripted_actor(bm)
+    actor.start(timeout=5.0)
+    assert actor._thread is not None and actor._thread.is_alive()
+    actor.shutdown(timeout=5.0)
+    assert actor._running is False
+    # Thread should exit shortly after stop_future resolves
+    actor._thread.join(timeout=2.0)
+    assert not actor._thread.is_alive()
+
+
+def test_shutdown_is_idempotent(bm):
+    actor = make_scripted_actor(bm)
+    actor.start(timeout=5.0)
+    actor.shutdown(timeout=2.0)
+    # Second call should not raise
+    actor.shutdown(timeout=2.0)
+    assert actor._running is False
+
+
+# ---- call() dispatch ----
+
+
+def test_call_before_start_raises(bm):
+    actor = bm._BmMcpActor(["fake-bm", "mcp"])
+    with pytest.raises(RuntimeError):
+        actor.call("search_notes", {})
+
+
+def test_call_after_shutdown_raises(bm):
+    actor = make_scripted_actor(bm)
+    actor.start(timeout=5.0)
+    actor.shutdown(timeout=2.0)
+    with pytest.raises(RuntimeError, match="not running"):
+        actor.call("search_notes", {})
+
+
+def test_call_dispatches_through_actor_loop(bm):
+    session = FakeSession(default_response={"results": [], "ok": True})
+    actor = make_scripted_actor(bm, session=session)
+    actor.start(timeout=5.0)
+    try:
+        out = actor.call("search_notes", {"query": "hi"}, timeout=5.0)
+        # Output is whatever _extract_mcp_text produces; we just verify it
+        # contains the response payload we configured.
+        assert "ok" in out
+        assert session.calls == [("search_notes", {"query": "hi"})]
+    finally:
+        actor.shutdown(timeout=2.0)
+
+
+def test_call_returns_mcp_extracted_text(bm):
+    session = FakeSession(default_response={"permalink": "p/q/r", "title": "T"})
+    actor = make_scripted_actor(bm, session=session)
+    actor.start(timeout=5.0)
+    try:
+        out = actor.call("write_note", {"title": "T"}, timeout=5.0)
+        parsed = json.loads(out)
+        assert parsed["permalink"] == "p/q/r"
+    finally:
+        actor.shutdown(timeout=2.0)
+
+
+def test_call_timeout_raises_and_cancels_coroutine(bm):
+    session = FakeSession(hang=True)
+    actor = make_scripted_actor(bm, session=session)
+    actor.start(timeout=5.0)
+    try:
+        t0 = time.monotonic()
+        with pytest.raises(concurrent.futures.TimeoutError):
+            actor.call("search_notes", {}, timeout=0.3)
+        elapsed = time.monotonic() - t0
+        # Sanity: we didn't accidentally wait the full session-side sleep
+        assert elapsed < 2.0
+        # Give the actor loop a moment to propagate the cancellation
+        deadline = time.monotonic() + 2.0
+        while time.monotonic() < deadline and not session.was_cancelled:
+            time.sleep(0.05)
+        assert session.was_cancelled is True, (
+            "Expected the underlying coroutine to be cancelled when call() times out"
+        )
+    finally:
+        actor.shutdown(timeout=2.0)
+
+
+def test_list_tools_returns_independent_copy(bm):
+    actor = make_scripted_actor(bm)
+    actor.start(timeout=5.0)
+    try:
+        snapshot = actor.list_tools()
+        snapshot.append({"name": "tampered"})
+        assert "tampered" not in {t["name"] for t in actor.list_tools()}
+    finally:
+        actor.shutdown(timeout=2.0)
+
+
+def test_actor_init_error_logs_and_marks_not_running(bm, caplog):
+    actor = make_scripted_actor(bm, raise_at_init=RuntimeError("server crashed"))
+    with pytest.raises(RuntimeError):
+        actor.start(timeout=5.0)
+    assert actor._running is False

integrations/hermes/tests/test_capture.py

@@ -0,0 +1,337 @@
+"""
+Tests for the capture pipeline: sync_turn (per-turn) and on_session_end (summary).
+
+These tests run the *real* sync_turn code path with a mocked actor, so threading
+and argument-shape regressions are caught.
+"""
+
+from __future__ import annotations
+
+import json
+from datetime import datetime, timezone
+from unittest.mock import MagicMock
+
+import pytest
+
+
+def _provider_with_mock_actor(bm, *, project="test-proj", capture_folder="hermes-sessions"):
+    p = bm.BasicMemoryProvider()
+    p._initialized = True
+    p._project = project
+    p._capture_folder = capture_folder
+    p._session_id = "20260510_123456_abcdef"
+    p._session_started_at = datetime(2026, 5, 10, 12, 34, 56, tzinfo=timezone.utc)
+    actor = MagicMock()
+    p._actor = actor
+    return p, actor
+
+
+def _wait_for_thread(p, attr="_sync_thread", timeout=5.0):
+    t = getattr(p, attr)
+    if t is not None:
+        t.join(timeout=timeout)
+        assert not t.is_alive(), f"{attr} did not finish within {timeout}s"
+
+
+# ---- sync_turn first-turn path: write_note ----
+
+
+def test_sync_turn_first_turn_calls_write_note(bm):
+    p, actor = _provider_with_mock_actor(bm)
+    actor.call.return_value = json.dumps(
+        {
+            "permalink": "test-proj/hermes-sessions/hermes-session-2026-05-10-1234-abcdef",
+            "title": "Hermes Session 2026-05-10 1234 abcdef",
+        }
+    )
+
+    p.sync_turn("hello", "hi back")
+    _wait_for_thread(p, "_sync_thread")
+
+    actor.call.assert_called_once()
+    bm_tool, bm_args = actor.call.call_args[0][:2]
+    assert bm_tool == "write_note"
+    assert bm_args["project"] == "test-proj"
+    assert bm_args["directory"] == "hermes-sessions"
+    assert "Hermes Session" in bm_args["title"]
+    assert "hello" in bm_args["content"]
+    assert "hi back" in bm_args["content"]
+    assert "## Turns" in bm_args["content"]
+    assert bm_args["output_format"] == "json"
+    assert "hermes-session" in bm_args["tags"]
+
+
+def test_sync_turn_first_turn_stores_extracted_permalink(bm):
+    p, actor = _provider_with_mock_actor(bm)
+    actor.call.return_value = json.dumps(
+        {
+            "permalink": "test-proj/hermes-sessions/hermes-session-foo",
+            "title": "T",
+        }
+    )
+
+    p.sync_turn("u", "a")
+    _wait_for_thread(p, "_sync_thread")
+
+    assert p._session_note_id == "test-proj/hermes-sessions/hermes-session-foo"
+
+
+def test_sync_turn_records_first_user_message(bm):
+    p, actor = _provider_with_mock_actor(bm)
+    actor.call.return_value = json.dumps({"permalink": "x"})
+
+    p.sync_turn("the very first user message", "reply")
+    _wait_for_thread(p, "_sync_thread")
+
+    assert p._first_user_msg == "the very first user message"
+
+    # A subsequent first-user-msg call should NOT overwrite the original
+    p.sync_turn("a much later user message", "another reply")
+    _wait_for_thread(p, "_sync_thread")
+    assert p._first_user_msg == "the very first user message"
+
+
+# ---- sync_turn append path: edit_note ----
+
+
+def test_sync_turn_subsequent_turn_calls_edit_note_append(bm):
+    p, actor = _provider_with_mock_actor(bm)
+    p._session_note_id = "test-proj/hermes-sessions/already-exists"
+
+    p.sync_turn("turn 2 user", "turn 2 assistant")
+    _wait_for_thread(p, "_sync_thread")
+
+    actor.call.assert_called_once()
+    bm_tool, bm_args = actor.call.call_args[0][:2]
+    assert bm_tool == "edit_note"
+    assert bm_args["identifier"] == "test-proj/hermes-sessions/already-exists"
+    assert bm_args["operation"] == "append"
+    assert "turn 2 user" in bm_args["content"]
+    assert "turn 2 assistant" in bm_args["content"]
+
+
+def test_sync_turn_session_note_id_stable_across_turns(bm):
+    p, actor = _provider_with_mock_actor(bm)
+    actor.call.return_value = json.dumps({"permalink": "test-proj/folder/note-perma"})
+
+    p.sync_turn("u1", "a1")
+    _wait_for_thread(p, "_sync_thread")
+    first_id = p._session_note_id
+
+    # Reconfigure mock to return something different — should be IGNORED for
+    # the second turn since we're now using the existing permalink to append.
+    actor.call.return_value = json.dumps({"permalink": "wrong-id"})
+
+    p.sync_turn("u2", "a2")
+    _wait_for_thread(p, "_sync_thread")
+
+    assert p._session_note_id == first_id, "session_note_id mutated on second turn"
+    # And the second call was edit_note, not write_note
+    second_call_tool = actor.call.call_args_list[1][0][0]
+    assert second_call_tool == "edit_note"
+
+
+# ---- sync_turn gating ----
+
+
+def test_sync_turn_skipped_when_capture_per_turn_off(bm):
+    p, actor = _provider_with_mock_actor(bm)
+    p._capture_per_turn = False
+    p.sync_turn("u", "a")
+    actor.call.assert_not_called()
+
+
+def test_sync_turn_skipped_when_uninitialized(bm):
+    p = bm.BasicMemoryProvider()
+    p._actor = MagicMock()
+    p.sync_turn("u", "a")
+    p._actor.call.assert_not_called()
+
+
+def test_sync_turn_skipped_when_actor_none(bm):
+    p, _ = _provider_with_mock_actor(bm)
+    p._actor = None
+    p.sync_turn("u", "a")  # should not raise
+
+
+def test_sync_turn_skipped_when_circuit_open(bm):
+    import time as _time
+
+    p, actor = _provider_with_mock_actor(bm)
+    p._failure_pause_until = _time.monotonic() + 60.0
+    p.sync_turn("u", "a")
+    actor.call.assert_not_called()
+
+
+def test_sync_turn_records_failure_on_actor_exception(bm):
+    p, actor = _provider_with_mock_actor(bm)
+    actor.call.side_effect = RuntimeError("boom")
+
+    p.sync_turn("u", "a")
+    _wait_for_thread(p, "_sync_thread")
+
+    assert p._failure_count >= 1
+
+
+def test_sync_turn_thread_is_daemonic(bm):
+    p, actor = _provider_with_mock_actor(bm)
+    actor.call.return_value = json.dumps({"permalink": "x"})
+    p.sync_turn("u", "a")
+    assert p._sync_thread is not None
+    assert p._sync_thread.daemon is True
+    _wait_for_thread(p, "_sync_thread")
+
+
+# ---- _capture_turn directly (no thread) ----
+
+
+def test_capture_turn_first_call_writes_note_with_session_metadata(bm):
+    p, actor = _provider_with_mock_actor(bm)
+    actor.call.return_value = json.dumps({"permalink": "p/folder/note"})
+    p._capture_turn("u msg", "a msg")
+    bm_tool, bm_args = actor.call.call_args[0][:2]
+    assert bm_tool == "write_note"
+    assert "20260510_123456_abcdef" in bm_args["content"]
+    # Auto-captured banner present
+    assert "Auto-captured" in bm_args["content"]
+
+
+def test_capture_turn_truncates_huge_messages(bm):
+    p, actor = _provider_with_mock_actor(bm)
+    actor.call.return_value = json.dumps({"permalink": "p/x"})
+    huge = "X" * 10000
+    p._capture_turn(huge, huge)
+    bm_args = actor.call.call_args[0][1]
+    # Body should not contain 10k Xs verbatim — _truncate caps at 4000
+    assert bm_args["content"].count("X") < 9000
+    assert "..." in bm_args["content"]
+
+
+# ---- on_session_end summary ----
+
+
+def test_on_session_end_writes_summary_note(bm):
+    p, actor = _provider_with_mock_actor(bm)
+    p._session_note_id = "p/folder/transcript"
+    actor.call.return_value = json.dumps({"permalink": "p/folder/summary"})
+
+    messages = [
+        {"role": "user", "content": "first user message"},
+        {"role": "assistant", "content": "first assistant"},
+        {"role": "user", "content": "second user"},
+        {"role": "assistant", "content": "last assistant message"},
+    ]
+    p.on_session_end(messages)
+
+    actor.call.assert_called_once()
+    bm_tool, bm_args = actor.call.call_args[0][:2]
+    assert bm_tool == "write_note"
+    assert "Hermes Session Summary" in bm_args["title"]
+    assert bm_args["directory"] == "hermes-sessions"
+    assert "first user message" in bm_args["content"]
+    assert "last assistant message" in bm_args["content"]
+    # Summary should link back to the transcript via Relations when
+    # session_note_id is known
+    assert "summary_of [[p/folder/transcript]]" in bm_args["content"]
+
+
+def test_on_session_end_omits_relations_when_no_transcript_id(bm):
+    p, actor = _provider_with_mock_actor(bm)
+    p._session_note_id = None
+    actor.call.return_value = json.dumps({"permalink": "p/folder/summary"})
+
+    p.on_session_end([{"role": "user", "content": "u"}, {"role": "assistant", "content": "a"}])
+
+    bm_args = actor.call.call_args[0][1]
+    assert "## Relations" not in bm_args["content"]
+
+
+def test_on_session_end_handles_list_of_dicts_content(bm):
+    """OpenAI-style content blocks: list of {type: text, text: ...}."""
+    p, actor = _provider_with_mock_actor(bm)
+    actor.call.return_value = json.dumps({"permalink": "p/x"})
+
+    messages = [
+        {"role": "user", "content": [{"type": "text", "text": "hello world"}]},
+        {"role": "assistant", "content": [{"type": "text", "text": "goodbye"}]},
+    ]
+    p.on_session_end(messages)
+    bm_args = actor.call.call_args[0][1]
+    assert "hello world" in bm_args["content"]
+    assert "goodbye" in bm_args["content"]
+
+
+def test_on_session_end_uses_first_user_msg_when_set(bm):
+    p, actor = _provider_with_mock_actor(bm)
+    p._first_user_msg = "captured-via-sync_turn"
+    actor.call.return_value = json.dumps({"permalink": "p/x"})
+
+    messages = [
+        {"role": "user", "content": "in-the-messages-list"},
+        {"role": "assistant", "content": "ok"},
+    ]
+    p.on_session_end(messages)
+    bm_args = actor.call.call_args[0][1]
+    # _first_user_msg takes priority over messages list
+    assert "captured-via-sync_turn" in bm_args["content"]
+
+
+def test_on_session_end_handles_empty_messages(bm):
+    p, actor = _provider_with_mock_actor(bm)
+    actor.call.return_value = json.dumps({"permalink": "p/x"})
+    p.on_session_end([])
+    bm_args = actor.call.call_args[0][1]
+    assert "(no user message)" in bm_args["content"]
+    assert "(no assistant message)" in bm_args["content"]
+    assert "0 user / 0 assistant" in bm_args["content"]
+
+
+def test_on_session_end_skipped_when_disabled(bm):
+    p, actor = _provider_with_mock_actor(bm)
+    p._capture_session_end = False
+    p.on_session_end([{"role": "user", "content": "u"}])
+    actor.call.assert_not_called()
+
+
+def test_on_session_end_skipped_when_uninitialized(bm):
+    p = bm.BasicMemoryProvider()
+    p._actor = MagicMock()
+    p.on_session_end([{"role": "user", "content": "u"}])
+    p._actor.call.assert_not_called()
+
+
+def test_on_session_end_logs_and_swallows_errors(bm, caplog):
+    p, actor = _provider_with_mock_actor(bm)
+    actor.call.side_effect = RuntimeError("BM down")
+    # Should not raise — the summary is best-effort
+    p.on_session_end([{"role": "user", "content": "u"}])
+
+
+# ---- shutdown lifecycle ----
+
+
+def test_shutdown_clears_initialized_flag(bm):
+    p, actor = _provider_with_mock_actor(bm)
+    p.shutdown()
+    assert p._initialized is False
+    assert p._actor is None
+    actor.shutdown.assert_called_once()
+
+
+def test_shutdown_swallows_actor_errors(bm):
+    p, actor = _provider_with_mock_actor(bm)
+    actor.shutdown.side_effect = RuntimeError("bad")
+    # Must not raise
+    p.shutdown()
+    assert p._initialized is False
+
+
+def test_shutdown_idempotent(bm):
+    p, _ = _provider_with_mock_actor(bm)
+    p.shutdown()
+    p.shutdown()  # should not raise
+
+
+def test_shutdown_before_initialize_is_noop(bm):
+    p = bm.BasicMemoryProvider()
+    p.shutdown()  # _actor is None, shouldn't error

integrations/hermes/tests/test_commands.py

@@ -0,0 +1,725 @@
+"""
+Tests for the plugin-owned /bm-* slash commands.
+
+Covers:
+- registration through ctx.register_command (forward-compat path)
+- PluginManager reach-in (production path with current Hermes collector)
+- per-handler behavior: usage text, uninitialized provider, happy path,
+  and exception → plain-text error.
+"""
+
+from __future__ import annotations
+
+import json
+import sys
+import types
+from pathlib import Path
+from unittest.mock import MagicMock
+
+import pytest
+
+from .conftest import FakeSession, make_scripted_actor
+
+
+# ---------------------------------------------------------------------------
+# Helpers for the reach-in tests
+# ---------------------------------------------------------------------------
+
+
+class _ProviderCollectorLike:
+    """
+    Mirror of Hermes's real `_ProviderCollector` shape: captures
+    `register_memory_provider` and no-ops everything else. NOT a MagicMock —
+    `hasattr(collector, "register_command")` must return False, matching the
+    real collector.
+    """
+
+    def __init__(self):
+        self.provider = None
+
+    def register_memory_provider(self, provider):
+        self.provider = provider
+
+
+class _FakePluginManager:
+    """Stand-in for hermes_cli.plugins.PluginManager — just the bits we touch."""
+
+    def __init__(self):
+        self._plugin_commands: dict = {}
+        self._plugin_skills: dict = {}
+
+
+def _install_fake_hermes_cli(monkeypatch, *, resolve_returns=None):
+    """
+    Insert a fake `hermes_cli.plugins` (with `_ensure_plugins_discovered`) and
+    `hermes_cli.commands` (with `resolve_command`) into sys.modules so the
+    reach-in's lazy imports resolve. Returns the FakePluginManager instance
+    so tests can assert against its registries.
+
+    resolve_returns: optional mapping of command name → truthy/falsy value
+    the fake resolve_command should return. Use a truthy value to simulate a
+    built-in conflict for that name.
+    """
+    fake_mgr = _FakePluginManager()
+
+    plugins_mod = types.ModuleType("hermes_cli.plugins")
+
+    def _ensure_plugins_discovered(force: bool = False):
+        return fake_mgr
+
+    plugins_mod._ensure_plugins_discovered = _ensure_plugins_discovered  # type: ignore[attr-defined]
+
+    commands_mod = types.ModuleType("hermes_cli.commands")
+
+    def _resolve_command(name: str):
+        if resolve_returns and name in resolve_returns:
+            return resolve_returns[name]
+        return None
+
+    commands_mod.resolve_command = _resolve_command  # type: ignore[attr-defined]
+
+    hermes_cli = types.ModuleType("hermes_cli")
+    monkeypatch.setitem(sys.modules, "hermes_cli", hermes_cli)
+    monkeypatch.setitem(sys.modules, "hermes_cli.plugins", plugins_mod)
+    monkeypatch.setitem(sys.modules, "hermes_cli.commands", commands_mod)
+
+    return fake_mgr
+
+
+# ---------------------------------------------------------------------------
+# Registration
+# ---------------------------------------------------------------------------
+
+_EXPECTED_COMMANDS = {
+    "bm-search",
+    "bm-read",
+    "bm-context",
+    "bm-recent",
+    "bm-status",
+    "bm-remember",
+    "bm-project",
+    "bm-workspace",
+}
+
+
+def test_register_wires_up_all_slash_commands_on_modern_ctx(bm):
+    """Forward-compat path: when ctx supports register_command (e.g. after the
+    upstream collector patch lands, or for plugins loaded via PluginContext),
+    every /bm-* command is registered through that path."""
+    ctx = MagicMock()
+    bm._active_providers.clear()
+    bm.register(ctx)
+    names = {call.args[0] for call in ctx.register_command.call_args_list}
+    assert names == _EXPECTED_COMMANDS
+    bm._active_providers.clear()
+
+
+def test_register_command_calls_include_description_and_args_hint(bm):
+    ctx = MagicMock()
+    bm._active_providers.clear()
+    bm.register(ctx)
+    for call in ctx.register_command.call_args_list:
+        name, handler = call.args[0], call.args[1]
+        kwargs = call.kwargs
+        assert callable(handler)
+        assert "description" in kwargs and kwargs["description"]
+        assert "args_hint" in kwargs  # may be empty string for no-arg commands
+    bm._active_providers.clear()
+
+
+def test_register_tolerates_old_hermes_without_register_command(bm):
+    """Plugins must not crash on Hermes < v0.11.0 (no register_command)."""
+
+    class _OldCtx:
+        def __init__(self):
+            self.memory_calls = []
+
+        def register_memory_provider(self, provider):
+            self.memory_calls.append(provider)
+
+    ctx = _OldCtx()
+    bm._active_providers.clear()
+    bm.register(ctx)  # must not raise
+    assert len(ctx.memory_calls) == 1
+    bm._active_providers.clear()
+
+
+def test_register_swallows_register_command_errors(bm, caplog):
+    """If one register_command call fails, the others — and provider
+    registration — still proceed."""
+    ctx = MagicMock()
+    ctx.register_command.side_effect = ValueError("name collision with builtin")
+    bm._active_providers.clear()
+    with caplog.at_level("WARNING"):
+        bm.register(ctx)
+    ctx.register_memory_provider.assert_called_once()
+    assert ctx.register_command.call_count == len(_EXPECTED_COMMANDS)
+    assert "register_command" in caplog.text
+    bm._active_providers.clear()
+
+
+# ---------------------------------------------------------------------------
+# Reach-in (production path): ctx is the no-op _ProviderCollector
+# ---------------------------------------------------------------------------
+
+
+def test_reach_in_writes_all_commands_to_plugin_manager(bm, monkeypatch):
+    """Regression for Codex P1: with the real collector shape (no
+    register_command method), reach into PluginManager and write commands
+    directly. The unit suite previously used MagicMock, which masked this
+    silent-skip by making every attribute exist."""
+    fake_mgr = _install_fake_hermes_cli(monkeypatch)
+    ctx = _ProviderCollectorLike()
+    assert not hasattr(ctx, "register_command"), (
+        "test collector must mirror real _ProviderCollector — no register_command"
+    )
+
+    bm._active_providers.clear()
+    bm.register(ctx)
+    try:
+        assert ctx.provider is not None  # memory provider still registered
+        assert set(fake_mgr._plugin_commands.keys()) == _EXPECTED_COMMANDS
+        for name, entry in fake_mgr._plugin_commands.items():
+            assert callable(entry["handler"])
+            assert entry["plugin"] == "basic-memory"
+            assert "description" in entry
+            assert "args_hint" in entry
+    finally:
+        bm._active_providers.clear()
+
+
+def test_reach_in_writes_skill_to_plugin_manager(bm, monkeypatch):
+    """Same silent-skip applies to register_skill — the bundled skill never
+    landed in real installs prior to this fix. Reach-in writes the namespaced
+    entry directly."""
+    fake_mgr = _install_fake_hermes_cli(monkeypatch)
+    ctx = _ProviderCollectorLike()
+    bm._active_providers.clear()
+    bm.register(ctx)
+    try:
+        assert "basic-memory:basic-memory" in fake_mgr._plugin_skills
+        skill = fake_mgr._plugin_skills["basic-memory:basic-memory"]
+        assert skill["plugin"] == "basic-memory"
+        assert skill["bare_name"] == "basic-memory"
+        assert isinstance(skill["path"], Path)
+        assert skill["path"].name == "SKILL.md"
+    finally:
+        bm._active_providers.clear()
+
+
+def test_reach_in_skips_command_conflicting_with_builtin(bm, monkeypatch, caplog):
+    """Mirror Hermes's PluginContext.register_command guard — when
+    resolve_command(name) returns a truthy value, skip that command and
+    log a warning rather than overwriting a built-in."""
+    # Simulate /bm-search colliding with a built-in.
+    fake_mgr = _install_fake_hermes_cli(monkeypatch, resolve_returns={"bm-search": object()})
+    ctx = _ProviderCollectorLike()
+    bm._active_providers.clear()
+    with caplog.at_level("WARNING"):
+        bm.register(ctx)
+    try:
+        assert "bm-search" not in fake_mgr._plugin_commands
+        # Other commands still landed
+        assert "bm-read" in fake_mgr._plugin_commands
+        assert "conflicts with a built-in" in caplog.text
+    finally:
+        bm._active_providers.clear()
+
+
+def test_reach_in_degrades_when_hermes_cli_missing(bm, monkeypatch, caplog):
+    """If hermes_cli.plugins isn't importable (e.g. running outside a Hermes
+    install), the reach-in must log and continue — never crash the plugin's
+    memory-provider registration."""
+    # Don't install fake modules; force import to fail.
+    monkeypatch.setitem(sys.modules, "hermes_cli.plugins", None)  # type: ignore[arg-type]
+    ctx = _ProviderCollectorLike()
+    bm._active_providers.clear()
+    with caplog.at_level("DEBUG"):
+        bm.register(ctx)  # must not raise
+    try:
+        assert ctx.provider is not None
+        # Either DEBUG message logged or nothing — both acceptable degrade modes.
+    finally:
+        bm._active_providers.clear()
+
+
+def test_reach_in_degrades_when_plugin_manager_missing_attrs(bm, monkeypatch):
+    """Forward-compat: if Hermes ever refactors _plugin_commands /
+    _plugin_skills away, the reach-in must not crash."""
+    fake_mgr = _FakePluginManager()
+    # Strip the attrs to simulate the rename/refactor
+    del fake_mgr._plugin_commands
+    del fake_mgr._plugin_skills
+
+    plugins_mod = types.ModuleType("hermes_cli.plugins")
+    plugins_mod._ensure_plugins_discovered = lambda force=False: fake_mgr  # type: ignore[attr-defined]
+    monkeypatch.setitem(sys.modules, "hermes_cli.plugins", plugins_mod)
+
+    ctx = _ProviderCollectorLike()
+    bm._active_providers.clear()
+    bm.register(ctx)  # must not raise
+    bm._active_providers.clear()
+
+
+def test_reach_in_normalizes_command_names(bm, monkeypatch):
+    """Reach-in must mirror Hermes's name normalization (lowercase, strip,
+    leading slash removed, spaces → hyphens). All our names are already
+    canonical, so this is a defensive check on the transform itself."""
+    fake_mgr = _install_fake_hermes_cli(monkeypatch)
+    ctx = _ProviderCollectorLike()
+    bm._active_providers.clear()
+    bm.register(ctx)
+    try:
+        for name in fake_mgr._plugin_commands:
+            assert name == name.lower()
+            assert not name.startswith("/")
+            assert " " not in name
+    finally:
+        bm._active_providers.clear()
+
+
+def test_reach_in_entries_match_hermes_internal_shape(bm, monkeypatch):
+    """The dict shape PluginContext.register_command writes
+    (plugins.py:447-452) is the contract Hermes's dispatch reads. Our
+    reach-in must produce byte-identical entries."""
+    fake_mgr = _install_fake_hermes_cli(monkeypatch)
+    ctx = _ProviderCollectorLike()
+    bm._active_providers.clear()
+    bm.register(ctx)
+    try:
+        entry = fake_mgr._plugin_commands["bm-search"]
+        assert set(entry.keys()) == {"handler", "description", "plugin", "args_hint"}
+        assert callable(entry["handler"])
+        assert entry["description"]  # non-empty string
+        assert entry["plugin"] == "basic-memory"
+        assert isinstance(entry["args_hint"], str)
+    finally:
+        bm._active_providers.clear()
+
+
+# ---------------------------------------------------------------------------
+# Per-handler tests
+# ---------------------------------------------------------------------------
+
+
+def _ready_provider(bm, session: FakeSession | None = None):
+    """Build a provider in 'initialized' state with a scripted actor."""
+    provider = bm.BasicMemoryProvider()
+    actor = make_scripted_actor(bm, session=session)
+    actor.start()
+    provider._actor = actor
+    provider._initialized = True
+    provider._project = "test-proj"
+    return provider, actor
+
+
+def _handlers_by_name(bm, provider):
+    return {name: handler for name, handler, _, _ in bm._build_slash_commands(provider)}
+
+
+# ---- Usage strings ----
+
+
+@pytest.mark.parametrize(
+    "name,args",
+    [
+        ("bm-search", ""),
+        ("bm-search", "help"),
+        ("bm-read", ""),
+        ("bm-read", "-h"),
+        ("bm-context", ""),
+        ("bm-remember", ""),
+        ("bm-remember", "--help"),
+        # Commands that take no args use 'help' to surface their usage line
+        ("bm-recent", "help"),
+        ("bm-status", "help"),
+        ("bm-project", "help"),
+        ("bm-workspace", "help"),
+    ],
+)
+def test_usage_returned_for_empty_or_help_args(bm, name, args):
+    provider = bm.BasicMemoryProvider()  # not initialized — usage path shouldn't need it
+    handlers = _handlers_by_name(bm, provider)
+    out = handlers[name](args)
+    assert isinstance(out, str)
+    assert out.lower().startswith("usage:")
+
+
+# ---- Uninitialized provider ----
+
+
+@pytest.mark.parametrize(
+    "name,args",
+    [
+        ("bm-search", "hello"),
+        ("bm-read", "some/note"),
+        ("bm-context", "memory://x"),
+        ("bm-recent", ""),
+        ("bm-remember", "a thought"),
+        ("bm-project", ""),
+    ],
+)
+def test_handler_init_failure_returns_message(bm, monkeypatch, name, args):
+    provider = bm.BasicMemoryProvider()
+    monkeypatch.setattr(provider, "initialize", lambda *a, **kw: None)
+    handlers = _handlers_by_name(bm, provider)
+    out = handlers[name](args)
+    assert "not initialized" in out
+    assert name in out  # message includes command name
+
+
+def test_handler_lazily_initializes_provider_for_slash_command(bm, monkeypatch):
+    provider = bm.BasicMemoryProvider()
+    calls = []
+
+    def fake_initialize(*args, **kwargs):
+        calls.append((args, kwargs))
+        session = FakeSession(default_response={"results": []})
+        actor = make_scripted_actor(bm, session=session)
+        actor.start()
+        provider._actor = actor
+        provider._initialized = True
+
+    monkeypatch.setattr(provider, "initialize", fake_initialize)
+    out = _handlers_by_name(bm, provider)["bm-search"]("widgets")
+    assert "No results" in out
+    assert len(calls) == 1
+    assert calls[0][1]["session_id"].startswith("slash:bm-search:")
+
+
+# ---- /bm-status ----
+
+
+def test_bm_status_renders_provider_state(bm, monkeypatch):
+    provider = bm.BasicMemoryProvider()
+    provider._mode = "local"
+    provider._project = "demo"
+    provider._project_path = "/tmp/demo"
+    provider._capture_per_turn = True
+    provider._capture_session_end = False
+    provider._capture_folder = "transcripts"
+    provider._remember_folder = "inbox"
+    monkeypatch.setattr(bm, "_bm_binary_path", lambda: "/fake/bin/bm")
+    out = _handlers_by_name(bm, provider)["bm-status"]("")
+    assert "demo" in out
+    assert "/tmp/demo" in out
+    assert "/fake/bin/bm" in out
+    assert "Initialized: no" in out
+    assert "transcripts" in out and "inbox" in out
+
+
+# ---- /bm-search ----
+
+
+def test_bm_search_happy_path(bm):
+    session = FakeSession()
+    session.stub(
+        "search_notes",
+        lambda args: {
+            "results": [
+                {"title": "Decisions", "permalink": "decisions/foo", "content": "we chose X"},
+                {"title": "Plan", "permalink": "plans/p1", "preview": "next quarter"},
+            ]
+        },
+    )
+    provider, actor = _ready_provider(bm, session)
+    try:
+        out = _handlers_by_name(bm, provider)["bm-search"]("widgets")
+        assert "Decisions" in out
+        assert "decisions/foo" in out
+        assert "we chose X" in out
+        # Args sent to BM
+        call = session.calls[-1]
+        assert call[0] == "search_notes"
+        assert call[1]["query"] == "widgets"
+        assert call[1]["project"] == "test-proj"
+        assert call[1]["output_format"] == "json"
+    finally:
+        actor.shutdown()
+
+
+def test_bm_search_empty_results(bm):
+    session = FakeSession(default_response={"results": []})
+    provider, actor = _ready_provider(bm, session)
+    try:
+        out = _handlers_by_name(bm, provider)["bm-search"]("missing")
+        assert "No results" in out
+        assert "missing" in out
+    finally:
+        actor.shutdown()
+
+
+def test_bm_search_actor_exception_returns_plain_string(bm):
+    session = FakeSession()
+
+    def _boom(_args):
+        raise RuntimeError("MCP transport closed")
+
+    session.stub("search_notes", _boom)
+    provider, actor = _ready_provider(bm, session)
+    try:
+        out = _handlers_by_name(bm, provider)["bm-search"]("anything")
+        assert isinstance(out, str)
+        assert out.startswith("bm-search:")
+        assert "MCP transport closed" in out
+    finally:
+        actor.shutdown()
+
+
+# ---- /bm-read ----
+
+
+def test_bm_read_returns_text_body(bm):
+    """BM's read_note returns markdown wrapped in {"text": "..."} once our
+    extractor wraps the non-JSON response. The handler should unwrap and
+    return the bare markdown."""
+    session = FakeSession()
+    session.stub(
+        "read_note",
+        # FakeSession serializes whatever the handler returns; emit the JSON
+        # the wrapper would produce for a markdown response.
+        lambda args: {"text": "# Foo\n\nbody text"},
+    )
+    provider, actor = _ready_provider(bm, session)
+    try:
+        out = _handlers_by_name(bm, provider)["bm-read"]("foo")
+        assert out == "# Foo\n\nbody text"
+        assert session.calls[-1][1]["identifier"] == "foo"
+    finally:
+        actor.shutdown()
+
+
+# ---- /bm-recent ----
+
+
+def test_bm_recent_default_timeframe(bm):
+    session = FakeSession(default_response={"results": []})
+    provider, actor = _ready_provider(bm, session)
+    try:
+        out = _handlers_by_name(bm, provider)["bm-recent"]("")
+        assert "7d" in out
+        assert session.calls[-1][1]["timeframe"] == "7d"
+    finally:
+        actor.shutdown()
+
+
+def test_bm_recent_custom_timeframe(bm):
+    session = FakeSession()
+    session.stub(
+        "recent_activity",
+        lambda args: {"results": [{"title": "Recent thing", "permalink": "x/y"}]},
+    )
+    provider, actor = _ready_provider(bm, session)
+    try:
+        out = _handlers_by_name(bm, provider)["bm-recent"]("2 weeks")
+        assert "2 weeks" in out
+        assert "Recent thing" in out
+        assert session.calls[-1][1]["timeframe"] == "2 weeks"
+    finally:
+        actor.shutdown()
+
+
+def test_bm_recent_bare_list_shape(bm):
+    """Regression: BM's `recent_activity(output_format="json")` returns a bare
+    `list[dict]` (signature: `-> str | list[dict]`), not a dict-with-results.
+    The handler must surface those rows, not report "no activity"."""
+    session = FakeSession()
+    session.stub(
+        "recent_activity",
+        lambda args: [
+            {"title": "Edited yesterday", "permalink": "notes/a", "content": "blob"},
+            {"title": "Edited 3d ago", "permalink": "notes/b"},
+        ],
+    )
+    provider, actor = _ready_provider(bm, session)
+    try:
+        out = _handlers_by_name(bm, provider)["bm-recent"]("")
+        assert "Edited yesterday" in out
+        assert "Edited 3d ago" in out
+        assert "No activity" not in out
+    finally:
+        actor.shutdown()
+
+
+# ---- /bm-remember ----
+
+
+def test_bm_remember_derives_title_from_first_line(bm):
+    captured = {}
+
+    def _write(args):
+        captured.update(args)
+        return {"permalink": "bm-remember/note-perm"}
+
+    session = FakeSession()
+    session.stub("write_note", _write)
+    provider, actor = _ready_provider(bm, session)
+    provider._remember_folder = "bm-remember"
+    try:
+        out = _handlers_by_name(bm, provider)["bm-remember"](
+            "Quarterly OKR review notes\n\nWe agreed to ship X."
+        )
+        assert "Saved:" in out
+        assert "bm-remember/note-perm" in out
+        assert captured["title"] == "Quarterly OKR review notes"
+        assert captured["directory"] == "bm-remember"
+        assert "manual-capture" in captured["tags"]
+    finally:
+        actor.shutdown()
+
+
+def test_bm_remember_long_first_line_truncated_to_80(bm):
+    session = FakeSession(default_response={"permalink": "x"})
+    provider, actor = _ready_provider(bm, session)
+    try:
+        long_line = "A" * 200
+        _handlers_by_name(bm, provider)["bm-remember"](long_line)
+        title = session.calls[-1][1]["title"]
+        assert len(title) == 80
+    finally:
+        actor.shutdown()
+
+
+def test_bm_remember_uses_configured_folder(bm):
+    session = FakeSession(default_response={"permalink": "x"})
+    provider, actor = _ready_provider(bm, session)
+    provider._remember_folder = "scratch"
+    try:
+        _handlers_by_name(bm, provider)["bm-remember"]("hello")
+        assert session.calls[-1][1]["directory"] == "scratch"
+    finally:
+        actor.shutdown()
+
+
+# ---- /bm-project ----
+
+
+def test_bm_project_lists_and_marks_active(bm):
+    session = FakeSession()
+    session.stub(
+        "list_memory_projects",
+        lambda args: {
+            "projects": [
+                {"name": "other-proj"},
+                {"name": "test-proj"},
+            ]
+        },
+    )
+    provider, actor = _ready_provider(bm, session)
+    try:
+        out = _handlers_by_name(bm, provider)["bm-project"]("")
+        assert "other-proj" in out
+        assert "test-proj" in out
+        # Active project line includes the marker
+        active_line = next(line for line in out.splitlines() if "test-proj" in line)
+        assert "active" in active_line
+    finally:
+        actor.shutdown()
+
+
+# ---- /bm-workspace ----
+
+
+def test_bm_workspace_local_mode_message(bm):
+    provider, actor = _ready_provider(bm)
+    provider._mode = "local"
+    try:
+        out = _handlers_by_name(bm, provider)["bm-workspace"]("")
+        assert "Cloud" in out
+        assert "local" in out
+    finally:
+        actor.shutdown()
+
+
+def test_bm_workspace_cloud_mode_lists(bm):
+    session = FakeSession()
+    session.stub(
+        "list_workspaces",
+        lambda args: {
+            "workspaces": [
+                {
+                    "name": "Personal",
+                    "workspace_type": "personal",
+                    "role": "owner",
+                    "is_default": True,
+                },
+                {"name": "Acme", "workspace_type": "team", "role": "member"},
+            ]
+        },
+    )
+    provider, actor = _ready_provider(bm, session)
+    provider._mode = "cloud"
+    try:
+        out = _handlers_by_name(bm, provider)["bm-workspace"]("")
+        assert "Personal" in out
+        assert "Acme" in out
+        assert "default" in out
+    finally:
+        actor.shutdown()
+
+
+def test_bm_workspace_lazily_initializes_before_mode_check(bm, monkeypatch):
+    session = FakeSession()
+    session.stub(
+        "list_workspaces",
+        lambda args: {"workspaces": [{"name": "Personal", "workspace_type": "personal"}]},
+    )
+    provider = bm.BasicMemoryProvider()
+    calls = []
+
+    def fake_initialize(*args, **kwargs):
+        calls.append((args, kwargs))
+        actor = make_scripted_actor(bm, session=session)
+        actor.start()
+        provider._actor = actor
+        provider._initialized = True
+        provider._mode = "cloud"
+
+    monkeypatch.setattr(provider, "initialize", fake_initialize)
+    try:
+        out = _handlers_by_name(bm, provider)["bm-workspace"]("")
+        assert "Personal" in out
+        assert "no workspaces to list" not in out
+        assert calls
+        assert session.calls[-1][0] == "list_workspaces"
+    finally:
+        if provider._actor is not None:
+            provider._actor.shutdown()
+
+
+# ---- _unwrap_json_or_text helper ----
+
+
+def test_unwrap_passes_through_raw_string(bm):
+    assert bm._unwrap_json_or_text("plain text") == "plain text"
+
+
+def test_unwrap_returns_inner_json_when_text_wraps_json(bm):
+    outer = json.dumps({"text": json.dumps({"a": 1})})
+    assert bm._unwrap_json_or_text(outer) == {"a": 1}
+
+
+def test_unwrap_returns_text_value_when_inner_is_markdown(bm):
+    outer = json.dumps({"text": "# Heading\n\nbody"})
+    assert bm._unwrap_json_or_text(outer) == "# Heading\n\nbody"
+
+
+def test_unwrap_returns_dict_when_top_level_json(bm):
+    outer = json.dumps({"results": [1, 2]})
+    assert bm._unwrap_json_or_text(outer) == {"results": [1, 2]}
+
+
+# ---- _remember_title ----
+
+
+def test_remember_title_strips_markdown_heading(bm):
+    assert bm._remember_title("# Decisions\n\nbody") == "Decisions"
+
+
+def test_remember_title_skips_blank_lines(bm):
+    assert bm._remember_title("\n\nFirst real line\nrest") == "First real line"
+
+
+def test_remember_title_falls_back_to_timestamp(bm):
+    title = bm._remember_title("   \n\n")
+    assert title.startswith("Note ")

integrations/hermes/tests/test_helpers.py

@@ -0,0 +1,531 @@
+"""Unit tests for the pure helpers in __init__.py."""
+
+import json
+
+import pytest
+
+
+# ---- _truncate ----
+
+
+def test_truncate_short_passes_through(bm):
+    assert bm._truncate("hello", 10) == "hello"
+
+
+def test_truncate_long_gets_ellipsis(bm):
+    out = bm._truncate("a" * 100, 10)
+    assert out.endswith("...")
+    assert len(out) == 10
+
+
+def test_truncate_non_string_coerced(bm):
+    assert bm._truncate(42, 10) == "42"
+
+
+def test_truncate_none(bm):
+    assert bm._truncate(None, 10) == ""
+
+
+# ---- _join_message_content ----
+
+
+def test_join_string_content(bm):
+    assert bm._join_message_content("hello") == "hello"
+
+
+def test_join_list_of_dicts(bm):
+    parts = [{"text": "a"}, {"text": "b"}, {"content": "c"}]
+    assert bm._join_message_content(parts) == "a\nb\nc"
+
+
+def test_join_list_of_strings(bm):
+    parts = ["a", "b"]
+    assert bm._join_message_content(parts) == "a\nb"
+
+
+def test_join_mixed(bm):
+    parts = ["a", {"text": "b"}, {"foo": "bar"}, "c"]
+    assert bm._join_message_content(parts) == "a\nb\nc"
+
+
+def test_join_none(bm):
+    assert bm._join_message_content(None) == ""
+
+
+# ---- _coerce_bool ----
+
+
+@pytest.mark.parametrize(
+    "value,expected",
+    [
+        (True, True),
+        (False, False),
+        ("true", True),
+        ("True", True),
+        ("YES", True),
+        ("1", True),
+        ("y", True),
+        ("false", False),
+        ("False", False),
+        ("NO", False),
+        ("0", False),
+        ("n", False),
+    ],
+)
+def test_coerce_bool(bm, value, expected):
+    assert bm._coerce_bool(value) is expected
+
+
+def test_coerce_bool_non_bool_passes_through(bm):
+    assert bm._coerce_bool(42) == 42
+    assert bm._coerce_bool("hello") == "hello"
+
+
+# ---- _extract_mcp_text ----
+
+
+def test_extract_mcp_text_passes_json_through(bm, fake_result):
+    payload = json.dumps({"permalink": "foo/bar", "title": "T"})
+    out = bm._extract_mcp_text(fake_result([payload]))
+    assert json.loads(out)["permalink"] == "foo/bar"
+
+
+def test_extract_mcp_text_wraps_markdown(bm, fake_result):
+    md = "# Created note\npermalink: foo/bar"
+    out = bm._extract_mcp_text(fake_result([md]))
+    parsed = json.loads(out)
+    assert parsed["text"] == md
+
+
+def test_extract_mcp_text_joins_multiple_blocks(bm, fake_result):
+    out = bm._extract_mcp_text(fake_result(["a", "b"]))
+    parsed = json.loads(out)
+    assert parsed["text"] == "a\nb"
+
+
+def test_extract_mcp_text_empty(bm, fake_result):
+    out = bm._extract_mcp_text(fake_result([]))
+    assert json.loads(out) == {"ok": True}
+
+
+def test_extract_mcp_text_error(bm, fake_result):
+    out = bm._extract_mcp_text(fake_result(["something broke"], is_error=True))
+    assert "error" in json.loads(out)
+
+
+# ---- _extract_permalink ----
+
+
+def test_extract_permalink_from_bare_json(bm):
+    text = json.dumps({"permalink": "proj/folder/note", "title": "T"})
+    assert bm._extract_permalink(text, "fb") == "proj/folder/note"
+
+
+def test_extract_permalink_from_wrapped_json(bm):
+    text = json.dumps({"text": json.dumps({"permalink": "proj/folder/note"})})
+    assert bm._extract_permalink(text, "fb") == "proj/folder/note"
+
+
+def test_extract_permalink_from_wrapped_markdown(bm):
+    md = (
+        "# Created note\n"
+        "project: hermes-jodys-imac\n"
+        "file_path: x/y.md\n"
+        "permalink: hermes-jodys-imac/folder/slug-name\n"
+        "checksum: unknown\n"
+    )
+    text = json.dumps({"text": md})
+    assert bm._extract_permalink(text, "fb") == "hermes-jodys-imac/folder/slug-name"
+
+
+def test_extract_permalink_from_raw_markdown(bm):
+    md = "# Created note\npermalink: proj/folder/slug\n"
+    # Raw, not wrapped — strategy 4 path
+    assert bm._extract_permalink(md, "fb") == "proj/folder/slug"
+
+
+def test_extract_permalink_no_match(bm):
+    assert bm._extract_permalink('{"foo":"bar"}', "fallback-title") == "fallback-title"
+
+
+def test_extract_permalink_empty(bm):
+    assert bm._extract_permalink("", "fb") == "fb"
+
+
+def test_extract_permalink_invalid(bm):
+    assert bm._extract_permalink("not json or markdown", "fb") == "fb"
+
+
+def test_extract_permalink_strips_trailing_punct(bm):
+    md = "# Created note\npermalink: proj/folder/slug,"
+    assert bm._extract_permalink(md, "fb") == "proj/folder/slug"
+
+
+# ---- _translate_args ----
+
+
+def test_translate_search(bm):
+    tool, args = bm._translate_args("bm_search", {"query": "hi", "limit": 7}, "proj")
+    assert tool == "search_notes"
+    assert args == {"project": "proj", "query": "hi", "page_size": 7}
+
+
+def test_translate_search_no_limit(bm):
+    tool, args = bm._translate_args("bm_search", {"query": "hi"}, "proj")
+    assert tool == "search_notes"
+    assert args == {"project": "proj", "query": "hi"}
+
+
+def test_translate_read(bm):
+    tool, args = bm._translate_args("bm_read", {"identifier": "x/y"}, "proj")
+    assert tool == "read_note"
+    assert args == {"project": "proj", "identifier": "x/y"}
+
+
+def test_translate_read_workspace_qualified_identifier_self_routes(bm):
+    tool, args = bm._translate_args(
+        "bm_read",
+        {"identifier": "personal/main/scratch/note"},
+        "hermes-memory",
+    )
+    assert tool == "read_note"
+    assert args == {"identifier": "personal/main/scratch/note"}
+
+
+def test_translate_read_org_workspace_qualified_identifier_self_routes(bm):
+    tool, args = bm._translate_args(
+        "bm_read",
+        {"identifier": "basic-memory-7020de4e925843c68c9056c60d101d9e/main/scratch/note"},
+        "hermes-memory",
+    )
+    assert tool == "read_note"
+    assert args == {"identifier": "basic-memory-7020de4e925843c68c9056c60d101d9e/main/scratch/note"}
+
+
+def test_translate_write(bm):
+    tool, args = bm._translate_args(
+        "bm_write",
+        {"title": "T", "content": "C", "folder": "F", "tags": ["a", "b"]},
+        "proj",
+    )
+    assert tool == "write_note"
+    assert args == {
+        "project": "proj",
+        "title": "T",
+        "content": "C",
+        "directory": "F",
+        "tags": ["a", "b"],
+    }
+
+
+def test_translate_write_no_tags(bm):
+    tool, args = bm._translate_args(
+        "bm_write",
+        {"title": "T", "content": "C", "folder": "F"},
+        "proj",
+    )
+    assert tool == "write_note"
+    assert "tags" not in args
+    assert args["directory"] == "F"
+
+
+def test_translate_edit_minimal(bm):
+    tool, args = bm._translate_args(
+        "bm_edit",
+        {"identifier": "x", "operation": "append", "content": "more"},
+        "proj",
+    )
+    assert tool == "edit_note"
+    assert args == {
+        "project": "proj",
+        "identifier": "x",
+        "operation": "append",
+        "content": "more",
+    }
+
+
+def test_translate_edit_find_replace(bm):
+    tool, args = bm._translate_args(
+        "bm_edit",
+        {
+            "identifier": "x",
+            "operation": "find_replace",
+            "content": "new",
+            "find_text": "old",
+        },
+        "proj",
+    )
+    assert args["find_text"] == "old"
+
+
+def test_translate_edit_replace_section(bm):
+    tool, args = bm._translate_args(
+        "bm_edit",
+        {
+            "identifier": "x",
+            "operation": "replace_section",
+            "content": "new",
+            "section": "## Notes",
+        },
+        "proj",
+    )
+    assert args["section"] == "## Notes"
+
+
+def test_translate_context(bm):
+    tool, args = bm._translate_args("bm_context", {"url": "memory://x", "depth": 2}, "proj")
+    assert tool == "build_context"
+    assert args == {"project": "proj", "url": "memory://x", "depth": 2}
+
+
+def test_translate_context_workspace_qualified_url_self_routes(bm):
+    tool, args = bm._translate_args(
+        "bm_context",
+        {"url": "memory://personal/main/scratch/note", "depth": 1},
+        "hermes-memory",
+    )
+    assert tool == "build_context"
+    assert args == {"url": "memory://personal/main/scratch/note", "depth": 1}
+
+
+def test_translate_context_org_workspace_qualified_url_self_routes(bm):
+    tool, args = bm._translate_args(
+        "bm_context",
+        {
+            "url": "memory://basic-memory-7020de4e925843c68c9056c60d101d9e/main/scratch/note",
+            "depth": 1,
+        },
+        "hermes-memory",
+    )
+    assert tool == "build_context"
+    assert args == {
+        "url": "memory://basic-memory-7020de4e925843c68c9056c60d101d9e/main/scratch/note",
+        "depth": 1,
+    }
+
+
+def test_translate_delete(bm):
+    tool, args = bm._translate_args("bm_delete", {"identifier": "x"}, "proj")
+    assert tool == "delete_note"
+    assert args == {"project": "proj", "identifier": "x"}
+
+
+def test_translate_move(bm):
+    tool, args = bm._translate_args(
+        "bm_move", {"identifier": "x", "new_folder": "archive/2026"}, "proj"
+    )
+    assert tool == "move_note"
+    assert args == {
+        "project": "proj",
+        "identifier": "x",
+        "destination_folder": "archive/2026",
+    }
+
+
+def test_translate_recent_defaults(bm):
+    tool, args = bm._translate_args("bm_recent", {}, "proj")
+    assert tool == "recent_activity"
+    assert args == {"project": "proj"}
+
+
+def test_translate_recent_full(bm):
+    tool, args = bm._translate_args(
+        "bm_recent",
+        {"timeframe": "2 weeks", "limit": 25, "type": "entity"},
+        "proj",
+    )
+    assert tool == "recent_activity"
+    assert args == {
+        "project": "proj",
+        "timeframe": "2 weeks",
+        "page_size": 25,
+        "type": "entity",
+    }
+
+
+# ---- Per-call project routing ----
+
+
+def test_translate_uses_default_project_when_no_override(bm):
+    """Existing behavior preserved: with no project override, the configured
+    default flows through."""
+    _, args = bm._translate_args("bm_search", {"query": "hi"}, "default-proj")
+    assert args["project"] == "default-proj"
+    assert "project_id" not in args
+
+
+def test_translate_uses_project_name_override(bm):
+    """Agent passes project="main" → that name reaches BM, not the default."""
+    _, args = bm._translate_args("bm_search", {"query": "hi", "project": "main"}, "default-proj")
+    assert args["project"] == "main"
+    assert "project_id" not in args
+
+
+def test_translate_uses_project_id_override(bm):
+    """Agent passes project_id=<uuid> → reaches BM as project_id, with no
+    project name in the call (would be redundant and risk server-side
+    precedence surprises)."""
+    uuid = "bf2a4c1e-d77f-4b7a-9c3e-5d8a1f0e2b6d"
+    _, args = bm._translate_args("bm_search", {"query": "hi", "project_id": uuid}, "default-proj")
+    assert args["project_id"] == uuid
+    assert "project" not in args
+
+
+def test_translate_project_id_wins_when_both_supplied(bm):
+    """If the agent passes both, project_id is the more specific identifier
+    (UUID across workspaces) and takes precedence. Only project_id reaches BM."""
+    uuid = "bf2a4c1e-d77f-4b7a-9c3e-5d8a1f0e2b6d"
+    _, args = bm._translate_args(
+        "bm_search",
+        {"query": "hi", "project": "main", "project_id": uuid},
+        "default-proj",
+    )
+    assert args["project_id"] == uuid
+    assert "project" not in args
+
+
+def test_translate_routing_coerces_to_string(bm):
+    """Defensive: if a model passes a non-string identifier (e.g. an int),
+    coerce rather than crash. BM accepts strings."""
+    _, args = bm._translate_args("bm_search", {"query": "hi", "project_id": 12345}, "default-proj")
+    assert args["project_id"] == "12345"
+
+
+@pytest.mark.parametrize(
+    "tool,base_args",
+    [
+        ("bm_search", {"query": "x"}),
+        ("bm_read", {"identifier": "x"}),
+        ("bm_write", {"title": "t", "content": "c", "folder": "f"}),
+        ("bm_edit", {"identifier": "x", "operation": "append", "content": "c"}),
+        ("bm_context", {"url": "memory://x"}),
+        ("bm_delete", {"identifier": "x"}),
+        ("bm_move", {"identifier": "x", "new_folder": "f"}),
+        ("bm_recent", {}),
+    ],
+)
+def test_translate_routing_works_for_every_tool(bm, tool, base_args):
+    """Routing applies uniformly across every per-project tool. Global
+    discovery tools (bm_projects, bm_workspaces) are tested separately."""
+    args_with = dict(base_args, project="main")
+    _, out = bm._translate_args(tool, args_with, "default-proj")
+    assert out["project"] == "main"
+
+    args_with_id = dict(base_args, project_id="e1d3a5b8-0492-4c1f-8e7d-2a4b6c8d0e2f")
+    _, out = bm._translate_args(tool, args_with_id, "default-proj")
+    assert out["project_id"] == "e1d3a5b8-0492-4c1f-8e7d-2a4b6c8d0e2f"
+    assert "project" not in out
+
+    _, out = bm._translate_args(tool, base_args, "default-proj")
+    assert out["project"] == "default-proj"
+
+
+# ---- Global discovery tools (bm_projects, bm_workspaces) ----
+
+
+def test_translate_bm_projects_no_routing(bm):
+    """bm_projects is a global discovery tool — it lists across all projects
+    and workspaces. _translate_args must NOT inject a default project
+    (would make BM scope the listing) and MUST request JSON so the agent
+    can parse identifiers out of the response."""
+    tool, out = bm._translate_args("bm_projects", {}, "default-proj")
+    assert tool == "list_memory_projects"
+    assert "project" not in out
+    assert "project_id" not in out
+    assert out == {"output_format": "json"}
+
+
+def test_translate_bm_workspaces_no_routing(bm):
+    tool, out = bm._translate_args("bm_workspaces", {}, "default-proj")
+    assert tool == "list_workspaces"
+    assert "project" not in out
+    assert "project_id" not in out
+    assert out == {"output_format": "json"}
+
+
+def test_translate_global_tools_ignore_project_kwargs(bm):
+    """Even if a confused caller passes project/project_id to a global tool,
+    those args are dropped — BM doesn't accept them and silently scoping
+    the listing would be worse than ignoring the args."""
+    _, out = bm._translate_args(
+        "bm_projects",
+        {"project": "main", "project_id": "e1d3a5b8-0492-4c1f-8e7d-2a4b6c8d0e2f"},
+        "default-proj",
+    )
+    assert "project" not in out
+    assert "project_id" not in out
+
+
+# ---- TOOL_SCHEMAS routing properties ----
+
+
+def test_every_tool_schema_advertises_project_routing(bm):
+    """Every per-project bm_* tool must expose `project` and `project_id` so
+    the agent sees them in the tool surface. Regression: forgetting to add
+    routing props to a new tool would silently lock the agent into the
+    active project — exactly the friction Drew's note flagged.
+
+    Global discovery tools (bm_projects, bm_workspaces) are excluded — they
+    list across projects/workspaces and don't take routing args."""
+    for schema in bm.TOOL_SCHEMAS:
+        props = schema["parameters"]["properties"]
+        if schema["name"] in bm._GLOBAL_TOOLS:
+            assert "project" not in props, (
+                f"{schema['name']} is a global tool; should not have project prop"
+            )
+            assert "project_id" not in props, (
+                f"{schema['name']} is a global tool; should not have project_id prop"
+            )
+            continue
+        assert "project" in props, f"{schema['name']} missing project prop"
+        assert "project_id" in props, f"{schema['name']} missing project_id prop"
+        # Routing is always optional — never in `required`.
+        required = schema["parameters"].get("required", [])
+        assert "project" not in required
+        assert "project_id" not in required
+
+
+# ---- _default_project / _hostname ----
+
+
+def test_default_project_format(bm):
+    p = bm._default_project()
+    assert p.startswith("hermes-")
+    # Hostnames are lowercased and stripped
+    assert " " not in p
+
+
+def test_hostname_lowercased(bm, monkeypatch):
+    monkeypatch.setattr(bm.socket, "gethostname", lambda: "Some.Long.Host")
+    assert bm._hostname() == "some"
+
+
+# ---- TOOL_SCHEMAS ----
+
+
+def test_tool_schemas_complete(bm):
+    names = {s["name"] for s in bm.TOOL_SCHEMAS}
+    expected = {
+        "bm_search",
+        "bm_read",
+        "bm_write",
+        "bm_edit",
+        "bm_context",
+        "bm_delete",
+        "bm_move",
+        "bm_recent",
+        "bm_projects",
+        "bm_workspaces",
+    }
+    assert names == expected
+
+
+def test_tool_schemas_have_descriptions(bm):
+    for s in bm.TOOL_SCHEMAS:
+        assert s["description"], f"{s['name']} missing description"
+        assert "parameters" in s
+        assert s["parameters"]["type"] == "object"
+
+
+def test_hermes_to_bm_complete(bm):
+    assert set(bm._HERMES_TO_BM.keys()) == {s["name"] for s in bm.TOOL_SCHEMAS}

integrations/hermes/tests/test_integration.py

@@ -0,0 +1,423 @@
+"""
+Gated integration tests against a real basic-memory MCP server.
+
+These tests spin up the real `bm mcp` subprocess via the production actor and
+exercise every tool through `handle_tool_call`, mirroring the production code
+path. They are skipped unless BOTH:
+
+    BM_INTEGRATION=1
+    AND `bm` is installed AND `mcp` Python package is importable
+
+A throwaway BM project is created for the test session and removed afterward,
+so these tests never touch your real BM projects.
+
+Run them with:
+
+    BM_INTEGRATION=1 uv run --with pytest --with mcp pytest tests/test_integration.py
+"""
+
+from __future__ import annotations
+
+import json
+import os
+import shutil
+import subprocess
+import tempfile
+import time
+import uuid
+
+import pytest
+
+
+# ---- Gating ----
+
+_INTEGRATION_ENABLED = os.environ.get("BM_INTEGRATION") == "1"
+_BM_BIN = shutil.which("bm") or (
+    os.path.expanduser("~/.local/bin/bm")
+    if os.path.isfile(os.path.expanduser("~/.local/bin/bm"))
+    else None
+)
+
+try:
+    import mcp  # noqa: F401
+
+    _MCP_OK = True
+except Exception:
+    _MCP_OK = False
+
+pytestmark = [
+    pytest.mark.skipif(
+        not _INTEGRATION_ENABLED,
+        reason="set BM_INTEGRATION=1 to run integration tests",
+    ),
+    pytest.mark.skipif(_BM_BIN is None, reason="bm CLI not on PATH"),
+    pytest.mark.skipif(not _MCP_OK, reason="mcp Python package not installed"),
+]
+
+
+# ---- Session-scoped BM project ----
+
+
+@pytest.fixture(scope="session")
+def temp_bm_project():
+    """Create a throwaway BM project for the session; remove when done."""
+    project_name = f"hermes-bm-test-{uuid.uuid4().hex[:8]}"
+    project_dir = tempfile.mkdtemp(prefix=f"{project_name}-")
+    # Register
+    subprocess.run(
+        [_BM_BIN, "project", "add", project_name, project_dir],
+        check=False,
+        capture_output=True,
+        timeout=20,
+    )
+    yield project_name, project_dir
+    # Tear down
+    subprocess.run(
+        [_BM_BIN, "project", "remove", project_name],
+        check=False,
+        capture_output=True,
+        timeout=20,
+    )
+    shutil.rmtree(project_dir, ignore_errors=True)
+
+
+@pytest.fixture
+def provider(bm, temp_bm_project, tmp_path):
+    """Initialized provider pointing at the temp project."""
+    project_name, project_dir = temp_bm_project
+    # Pre-write a config file in this test's hermes_home so initialize picks it up
+    cfg = {
+        "mode": "local",
+        "project": project_name,
+        "project_path": project_dir,
+        "capture_per_turn": True,
+        "capture_session_end": True,
+        "capture_folder": "test-sessions",
+    }
+    (tmp_path / "basic-memory.json").write_text(json.dumps(cfg))
+
+    p = bm.BasicMemoryProvider()
+    p.initialize(
+        session_id=f"integration-{uuid.uuid4().hex[:6]}",
+        hermes_home=str(tmp_path),
+        platform="cli",
+    )
+    if not p._initialized:
+        pytest.fail("Provider failed to initialize against the real bm MCP server")
+    yield p
+    p.shutdown()
+
+
+def _parse_tool_result(raw):
+    try:
+        d = json.loads(raw)
+    except Exception:
+        return None
+    return d
+
+
+# ---- Actor smoke ----
+
+
+def test_actor_starts_and_lists_expected_tools(provider, bm):
+    tools = {t["name"] for t in provider._actor.list_tools()}
+    expected = set(bm._HERMES_TO_BM.values())
+    missing = expected - tools
+    assert not missing, f"BM MCP server missing tools we depend on: {missing}"
+
+
+# ---- Tool surface ----
+
+
+def test_bm_write_returns_full_permalink(provider, bm):
+    title = f"Integration Write Test {uuid.uuid4().hex[:6]}"
+    raw = provider.handle_tool_call(
+        "bm_write",
+        {
+            "title": title,
+            "content": f"# {title}\n\nbody.\n",
+            "folder": "tests",
+            "tags": ["integration"],
+        },
+    )
+    permalink = bm._extract_permalink(raw, "")
+    assert permalink, f"no permalink extracted from: {raw[:300]}"
+    # BM permalinks include the project prefix
+    assert permalink.split("/")[0] == provider._project, (
+        f"permalink should start with project name: {permalink}"
+    )
+
+
+def test_bm_read_round_trips_a_written_note(provider, bm):
+    title = f"Read RT {uuid.uuid4().hex[:6]}"
+    body = f"# {title}\n\nMARKER-{uuid.uuid4().hex}\n"
+    raw = provider.handle_tool_call(
+        "bm_write",
+        {
+            "title": title,
+            "content": body,
+            "folder": "tests",
+        },
+    )
+    permalink = bm._extract_permalink(raw, "")
+    raw = provider.handle_tool_call("bm_read", {"identifier": permalink})
+    d = _parse_tool_result(raw)
+    text = (d or {}).get("text") or json.dumps(d or {})
+    assert title in text
+
+
+def test_bm_edit_append_lands_in_note(provider, bm):
+    title = f"Append Test {uuid.uuid4().hex[:6]}"
+    raw = provider.handle_tool_call(
+        "bm_write",
+        {
+            "title": title,
+            "content": f"# {title}\nseed\n",
+            "folder": "tests",
+        },
+    )
+    permalink = bm._extract_permalink(raw, "")
+    marker = f"APPEND-MARKER-{uuid.uuid4().hex}"
+    provider.handle_tool_call(
+        "bm_edit",
+        {
+            "identifier": permalink,
+            "operation": "append",
+            "content": f"\n{marker}\n",
+        },
+    )
+    raw = provider.handle_tool_call("bm_read", {"identifier": permalink})
+    d = _parse_tool_result(raw)
+    text = (d or {}).get("text") or json.dumps(d or {})
+    assert marker in text
+
+
+def test_bm_edit_replace_section_swaps_content(provider, bm):
+    title = f"ReplaceSection {uuid.uuid4().hex[:6]}"
+    body = f"# {title}\n\n## Notes\noriginal-body\n"
+    raw = provider.handle_tool_call(
+        "bm_write",
+        {
+            "title": title,
+            "content": body,
+            "folder": "tests",
+        },
+    )
+    permalink = bm._extract_permalink(raw, "")
+    new_marker = f"REPLACED-{uuid.uuid4().hex}"
+    provider.handle_tool_call(
+        "bm_edit",
+        {
+            "identifier": permalink,
+            "operation": "replace_section",
+            "section": "## Notes",
+            "content": new_marker,
+        },
+    )
+    raw = provider.handle_tool_call("bm_read", {"identifier": permalink})
+    d = _parse_tool_result(raw)
+    text = (d or {}).get("text") or json.dumps(d or {})
+    assert new_marker in text
+    assert "original-body" not in text
+
+
+def test_bm_search_finds_a_freshly_written_note(provider, bm):
+    unique = f"SEARCH-MARKER-{uuid.uuid4().hex}"
+    title = f"Search Test {unique}"
+    provider.handle_tool_call(
+        "bm_write",
+        {
+            "title": title,
+            "content": f"# {title}\nbody.\n",
+            "folder": "tests",
+        },
+    )
+    raw = provider.handle_tool_call("bm_search", {"query": unique, "limit": 5})
+    d = _parse_tool_result(raw)
+    text = (d or {}).get("text") or json.dumps(d or {})
+    assert unique in text or title in text
+
+
+def test_bm_context_returns_results(provider, bm):
+    title = f"Context Test {uuid.uuid4().hex[:6]}"
+    raw = provider.handle_tool_call(
+        "bm_write",
+        {
+            "title": title,
+            "content": f"# {title}\n",
+            "folder": "tests",
+        },
+    )
+    permalink = bm._extract_permalink(raw, "")
+    raw = provider.handle_tool_call(
+        "bm_context",
+        {
+            "url": f"memory://{permalink}",
+            "depth": 1,
+        },
+    )
+    d = _parse_tool_result(raw)
+    assert d is not None
+    # build_context returns a JSON dict with `results` (and other fields)
+    text_blob = json.dumps(d)
+    assert "results" in text_blob
+
+
+def test_bm_move_relocates_note(provider, bm):
+    """
+    BM permalinks are stable IDs that don't change on move — only the
+    file_path moves. So we verify by:
+    1. The move response itself reports the new destination
+    2. Reading by the original permalink still succeeds (note wasn't lost)
+    """
+    title = f"Move Test {uuid.uuid4().hex[:6]}"
+    raw = provider.handle_tool_call(
+        "bm_write",
+        {
+            "title": title,
+            "content": f"# {title}\n",
+            "folder": "tests",
+        },
+    )
+    permalink = bm._extract_permalink(raw, "")
+    assert permalink, "expected a permalink from bm_write"
+
+    raw = provider.handle_tool_call(
+        "bm_move",
+        {
+            "identifier": permalink,
+            "new_folder": "tests/archive",
+        },
+    )
+    d = _parse_tool_result(raw)
+    move_text = (d or {}).get("text") or json.dumps(d or {})
+    # Move response text reports both the old and new locations
+    assert "moved successfully" in move_text.lower() or "moved" in move_text.lower(), (
+        f"move response missing success indicator: {move_text[:200]}"
+    )
+    assert "tests/archive" in move_text, f"move response missing new folder: {move_text[:200]}"
+
+    # Permalink is stable — reading by it should still work
+    raw = provider.handle_tool_call("bm_read", {"identifier": permalink})
+    d = _parse_tool_result(raw)
+    read_text = (d or {}).get("text") or json.dumps(d or {})
+    assert title in read_text, "note should still be readable after move"
+
+
+def test_bm_delete_removes_note(provider, bm):
+    title = f"Delete Test {uuid.uuid4().hex[:6]}"
+    raw = provider.handle_tool_call(
+        "bm_write",
+        {
+            "title": title,
+            "content": f"# {title}\n",
+            "folder": "tests",
+        },
+    )
+    permalink = bm._extract_permalink(raw, "")
+    provider.handle_tool_call("bm_delete", {"identifier": permalink})
+    # Read should now indicate "not found"
+    raw = provider.handle_tool_call("bm_read", {"identifier": permalink})
+    d = _parse_tool_result(raw)
+    text = (d or {}).get("text") or json.dumps(d or {})
+    assert "not found" in text.lower() or "no notes found" in text.lower(), (
+        f"expected a 'not found' indication, got: {text[:200]}"
+    )
+
+
+# ---- Capture pipeline ----
+
+
+def test_sync_turn_writes_then_appends_to_same_session_note(provider, bm):
+    # First turn — creates the session note
+    provider.sync_turn("integration turn-1 user", "integration turn-1 assistant")
+    if provider._sync_thread:
+        provider._sync_thread.join(timeout=20.0)
+    sid_1 = provider._session_note_id
+    assert sid_1, "first sync_turn should set _session_note_id"
+    assert sid_1.startswith(provider._project + "/"), (
+        f"session_note_id should include project prefix, got: {sid_1}"
+    )
+
+    # Second turn — should append to the same note
+    marker = f"TURN-2-MARKER-{uuid.uuid4().hex}"
+    provider.sync_turn("integration turn-2 user", marker)
+    if provider._sync_thread:
+        provider._sync_thread.join(timeout=20.0)
+    sid_2 = provider._session_note_id
+    assert sid_2 == sid_1, "session_note_id should NOT change between turns"
+
+    # Verify both turn markers are present in the persisted note
+    raw = provider.handle_tool_call("bm_read", {"identifier": sid_1})
+    d = _parse_tool_result(raw)
+    text = (d or {}).get("text") or json.dumps(d or {})
+    assert "integration turn-1 user" in text
+    assert marker in text
+
+
+def test_on_session_end_writes_summary_with_relations(provider, bm):
+    # Seed a session note via sync_turn
+    provider.sync_turn("first message", "first reply")
+    if provider._sync_thread:
+        provider._sync_thread.join(timeout=20.0)
+    sid = provider._session_note_id
+    assert sid
+
+    provider.on_session_end(
+        [
+            {"role": "user", "content": "first message"},
+            {"role": "assistant", "content": "first reply"},
+        ]
+    )
+
+    # Search for the summary
+    raw = provider.handle_tool_call(
+        "bm_search",
+        {
+            "query": "Hermes Session Summary",
+            "limit": 5,
+        },
+    )
+    d = _parse_tool_result(raw)
+    text = (d or {}).get("text") or json.dumps(d or {})
+    assert "Hermes Session Summary" in text
+
+
+def test_prefetch_against_real_bm(provider, bm):
+    # Seed a recognizable note
+    unique = f"PREFETCH-MARKER-{uuid.uuid4().hex}"
+    provider.handle_tool_call(
+        "bm_write",
+        {
+            "title": f"Prefetch Test {unique}",
+            "content": f"# Prefetch Test\n{unique}\n",
+            "folder": "tests",
+        },
+    )
+
+    # BM's FTS index is updated synchronously inside the write_note API
+    # path (knowledge_router.py:272), so this loop is really only smoothing
+    # over the round-trip cost of a few RPCs on a slow runner. prefetch
+    # explicitly requests search_type="text" so we don't get pulled onto
+    # BM's hybrid path, where vector indexing is async and would race the
+    # search.
+    budget_secs = 10.0
+    deadline = time.monotonic() + budget_secs
+    out = ""
+    attempts = 0
+    while time.monotonic() < deadline:
+        attempts += 1
+        out = provider.prefetch(unique)
+        if out:
+            break
+        time.sleep(0.25)
+
+    assert out, (
+        f"prefetch returned nothing after {attempts} attempt(s) over "
+        f"{budget_secs}s; provider._failure_count={provider._failure_count}, "
+        f"circuit_open={provider._is_circuit_open()}. "
+        f"Either BM didn't index the note in time or prefetch's actor.call "
+        f"is timing out internally."
+    )
+    assert "Basic Memory Recall" in out
+    assert unique in out or "Prefetch Test" in out

integrations/hermes/tests/test_prefetch.py

@@ -0,0 +1,296 @@
+"""Tests for prefetch / queue_prefetch / _format_prefetch."""
+
+from __future__ import annotations
+
+import json
+import time
+from unittest.mock import MagicMock
+
+import pytest
+
+
+def _initialized_provider(bm):
+    p = bm.BasicMemoryProvider()
+    p._initialized = True
+    p._project = "test-proj"
+    p._actor = MagicMock()
+    return p
+
+
+# ---- prefetch ----
+
+
+def test_prefetch_returns_cached_value_drained(bm):
+    p = _initialized_provider(bm)
+    p._pending_prefetch = "## cached recall"
+    out = p.prefetch("any query")
+    assert out == "## cached recall"
+    # Cache must be drained so the next prefetch doesn't return stale results
+    assert p._pending_prefetch == ""
+    p._actor.call.assert_not_called()
+
+
+def test_prefetch_calls_search_when_cache_empty(bm):
+    p = _initialized_provider(bm)
+    p._actor.call.return_value = json.dumps(
+        {"results": [{"title": "T", "permalink": "p/t", "content": "c"}]}
+    )
+    out = p.prefetch("hello world")
+    assert "## Basic Memory Recall" in out
+    assert "**T**" in out
+    p._actor.call.assert_called_once()
+    bm_tool, bm_args = p._actor.call.call_args[0][:2]
+    assert bm_tool == "search_notes"
+    assert bm_args["query"] == "hello world"
+    assert bm_args["page_size"] == 5
+    assert bm_args["output_format"] == "json"
+    # Pin search_type=text so BM doesn't fall into the hybrid+async-vector
+    # path on the prefetch hot path. See prefetch() comment for rationale.
+    assert bm_args["search_type"] == "text"
+
+
+def test_prefetch_returns_empty_when_uninitialized(bm):
+    p = bm.BasicMemoryProvider()
+    assert p.prefetch("x") == ""
+
+
+def test_prefetch_returns_empty_when_circuit_open(bm):
+    p = _initialized_provider(bm)
+    p._failure_pause_until = time.monotonic() + 60.0
+    assert p.prefetch("x") == ""
+    p._actor.call.assert_not_called()
+
+
+def test_prefetch_records_failure_on_actor_error(bm):
+    p = _initialized_provider(bm)
+    p._actor.call.side_effect = RuntimeError("boom")
+    assert p.prefetch("x") == ""
+    assert p._failure_count == 1
+
+
+def test_prefetch_returns_empty_for_empty_results(bm):
+    p = _initialized_provider(bm)
+    p._actor.call.return_value = json.dumps({"results": []})
+    assert p.prefetch("x") == ""
+
+
+# ---- queue_prefetch ----
+
+
+def test_queue_prefetch_fills_cache_in_background(bm):
+    p = _initialized_provider(bm)
+    p._actor.call.return_value = json.dumps(
+        {"results": [{"title": "Bg", "permalink": "p/bg", "content": "c"}]}
+    )
+    p.queue_prefetch("user typed something")
+    # Wait for the daemon thread to finish
+    if p._prefetch_thread:
+        p._prefetch_thread.join(timeout=5.0)
+    # Cache should now have the formatted result
+    assert "**Bg**" in p._pending_prefetch
+    # And subsequent prefetch returns it without making another call
+    p._actor.call.reset_mock()
+    out = p.prefetch("anything")
+    assert "**Bg**" in out
+    p._actor.call.assert_not_called()
+
+
+def test_queue_prefetch_uses_longer_timeout_than_sync_prefetch(bm):
+    """queue_prefetch runs in background, so it can afford a longer timeout."""
+    p = _initialized_provider(bm)
+    p._actor.call.return_value = json.dumps({"results": []})
+    p.queue_prefetch("q")
+    if p._prefetch_thread:
+        p._prefetch_thread.join(timeout=5.0)
+    timeout = p._actor.call.call_args.kwargs.get("timeout") or p._actor.call.call_args[1].get(
+        "timeout"
+    )
+    # Background prefetch is more patient than the foreground 3.0s
+    assert timeout is not None and timeout > 3.0
+
+
+def test_queue_prefetch_skipped_when_thread_in_flight(bm):
+    p = _initialized_provider(bm)
+
+    # Simulate an already-running prefetch thread
+    class _StillAlive:
+        def is_alive(self):
+            return True
+
+    p._prefetch_thread = _StillAlive()  # type: ignore[assignment]
+    p.queue_prefetch("q")
+    p._actor.call.assert_not_called()
+
+
+def test_queue_prefetch_skipped_when_circuit_open(bm):
+    p = _initialized_provider(bm)
+    p._failure_pause_until = time.monotonic() + 60.0
+    p.queue_prefetch("q")
+    p._actor.call.assert_not_called()
+
+
+def test_queue_prefetch_skipped_when_uninitialized(bm):
+    p = bm.BasicMemoryProvider()
+    p._actor = MagicMock()
+    p.queue_prefetch("q")
+    p._actor.call.assert_not_called()
+
+
+def test_queue_prefetch_records_failure_on_bg_error(bm):
+    p = _initialized_provider(bm)
+    p._actor.call.side_effect = RuntimeError("backend down")
+    p.queue_prefetch("q")
+    if p._prefetch_thread:
+        p._prefetch_thread.join(timeout=5.0)
+    assert p._failure_count >= 1
+
+
+# ---- _format_prefetch ----
+
+
+def _format(bm, payload):
+    """Helper: call _format_prefetch on a fresh provider."""
+    return bm.BasicMemoryProvider()._format_prefetch(payload)
+
+
+def test_format_prefetch_with_results(bm):
+    payload = json.dumps(
+        {
+            "results": [
+                {"title": "A", "permalink": "p/a", "content": "first line"},
+                {"title": "B", "permalink": "p/b", "content": "second line"},
+            ]
+        }
+    )
+    out = _format(bm, payload)
+    assert "## Basic Memory Recall" in out
+    assert "**A**" in out and "**B**" in out
+    assert "p/a" in out and "p/b" in out
+
+
+def test_format_prefetch_caps_at_5_entries(bm):
+    results = [{"title": f"T{i}", "permalink": f"p/{i}", "content": "x"} for i in range(20)]
+    payload = json.dumps({"results": results})
+    out = _format(bm, payload)
+    # Five lines + one heading = 6 lines max
+    assert out.count("\n- **") == 5
+
+
+def test_format_prefetch_caps_preview_length(bm):
+    payload = json.dumps(
+        {
+            "results": [
+                {"title": "T", "permalink": "p/t", "content": "x" * 5000},
+            ]
+        }
+    )
+    out = _format(bm, payload)
+    # Each result line includes the preview, capped to 200 chars
+    line = [l for l in out.split("\n") if l.startswith("- ")][0]
+    # Some boilerplate around the preview, but the long preview is capped
+    assert len(line) < 400
+
+
+def test_format_prefetch_collapses_whitespace(bm):
+    payload = json.dumps(
+        {
+            "results": [
+                {"title": "T", "permalink": "p/t", "content": "first\n\n  second\tthird"},
+            ]
+        }
+    )
+    out = _format(bm, payload)
+    assert "first second third" in out
+
+
+def test_format_prefetch_falls_back_to_preview_field(bm):
+    """BM may use 'preview' instead of 'content' in some response shapes."""
+    payload = json.dumps(
+        {
+            "results": [
+                {"title": "T", "permalink": "p/t", "preview": "preview-only"},
+            ]
+        }
+    )
+    out = _format(bm, payload)
+    assert "preview-only" in out
+
+
+def test_format_prefetch_handles_non_string_content(bm):
+    """Defensive: BM could conceivably return non-string content fields."""
+    payload = json.dumps(
+        {
+            "results": [
+                {"title": "T", "permalink": "p/t", "content": 12345},
+            ]
+        }
+    )
+    # Must not raise
+    out = _format(bm, payload)
+    assert "12345" in out
+
+
+def test_format_prefetch_handles_missing_title_permalink(bm):
+    payload = json.dumps(
+        {
+            "results": [
+                {"content": "orphan note"},
+            ]
+        }
+    )
+    out = _format(bm, payload)
+    assert "(untitled)" in out
+    assert "orphan note" in out
+
+
+def test_format_prefetch_skips_non_dict_entries(bm):
+    payload = json.dumps(
+        {
+            "results": [
+                "not-a-dict",
+                {"title": "Real", "permalink": "p/r", "content": "x"},
+            ]
+        }
+    )
+    out = _format(bm, payload)
+    assert "Real" in out
+    assert "not-a-dict" not in out
+
+
+def test_format_prefetch_with_text_wrapped_results(bm):
+    """BM text-format responses arrive wrapped as {"text": "..."} by _extract_mcp_text."""
+    inner = json.dumps({"results": [{"title": "X", "permalink": "p/x", "content": "c"}]})
+    payload = json.dumps({"text": inner})
+    out = _format(bm, payload)
+    assert "**X**" in out
+
+
+def test_format_prefetch_empty(bm):
+    assert _format(bm, json.dumps({"results": []})) == ""
+
+
+def test_format_prefetch_no_results_key(bm):
+    assert _format(bm, json.dumps({"foo": "bar"})) == ""
+
+
+def test_format_prefetch_malformed(bm):
+    assert _format(bm, "not-json") == ""
+
+
+def test_format_prefetch_handles_extra_unknown_fields(bm):
+    """Forward-compatibility: unknown fields shouldn't break formatting."""
+    payload = json.dumps(
+        {
+            "results": [
+                {
+                    "title": "T",
+                    "permalink": "p/t",
+                    "content": "c",
+                    "future_field_42": {"nested": "value"},
+                    "score": 0.9,
+                },
+            ]
+        }
+    )
+    out = _format(bm, payload)
+    assert "**T**" in out

integrations/hermes/tests/test_provider.py

@@ -0,0 +1,637 @@
+"""Tests for BasicMemoryProvider with the MCP actor mocked out."""
+
+from __future__ import annotations
+
+import json
+from pathlib import Path
+from unittest.mock import MagicMock
+
+
+def test_is_available_no_mcp(bm, monkeypatch):
+    monkeypatch.setattr(bm, "_MCP_AVAILABLE", False)
+    p = bm.BasicMemoryProvider()
+    assert p.is_available() is False
+
+
+def test_is_available_no_bm_no_uv(bm, monkeypatch):
+    """No bm AND no uv → can't install, can't operate → unavailable."""
+    monkeypatch.setattr(bm, "_MCP_AVAILABLE", True)
+    monkeypatch.setattr(bm, "_bm_binary_path", lambda: None)
+    monkeypatch.setattr(bm, "_uv_binary_path", lambda: None)
+    p = bm.BasicMemoryProvider()
+    assert p.is_available() is False
+
+
+def test_is_available_bm_present(bm, monkeypatch):
+    """bm already installed → available regardless of uv."""
+    monkeypatch.setattr(bm, "_MCP_AVAILABLE", True)
+    monkeypatch.setattr(bm, "_bm_binary_path", lambda: "/fake/bm")
+    monkeypatch.setattr(bm, "_uv_binary_path", lambda: None)
+    p = bm.BasicMemoryProvider()
+    assert p.is_available() is True
+
+
+def test_is_available_bm_missing_but_uv_present(bm, monkeypatch):
+    """bm missing but uv available → we can install bm at init time → available."""
+    monkeypatch.setattr(bm, "_MCP_AVAILABLE", True)
+    monkeypatch.setattr(bm, "_bm_binary_path", lambda: None)
+    monkeypatch.setattr(bm, "_uv_binary_path", lambda: "/fake/uv")
+    p = bm.BasicMemoryProvider()
+    assert p.is_available() is True
+
+
+def test_name(bm):
+    assert bm.BasicMemoryProvider().name == "basic-memory"
+    assert bm.BasicMemoryProvider().name == bm.PROVIDER_NAME
+
+
+def test_get_tool_schemas_unconditional(bm):
+    """
+    Regression: Hermes captures the schema list at *register* time, before
+    `initialize()` runs. If get_tool_schemas() returns [] when uninitialized,
+    Hermes builds _tool_to_provider with no entries for us and every
+    subsequent bm_* invocation returns "Unknown tool: bm_*" forever.
+    Schemas are static — return them unconditionally.
+    """
+    # Fresh provider, never initialized, should still expose all 10 schemas
+    p = bm.BasicMemoryProvider()
+    assert p._initialized is False
+    schemas = p.get_tool_schemas()
+    assert len(schemas) == 10
+    names = {s["name"] for s in schemas}
+    assert names == {
+        "bm_search",
+        "bm_read",
+        "bm_write",
+        "bm_edit",
+        "bm_context",
+        "bm_delete",
+        "bm_move",
+        "bm_recent",
+        "bm_projects",
+        "bm_workspaces",
+    }
+
+    # Initialized provider also returns 10 (idempotent)
+    p._initialized = True
+    assert len(p.get_tool_schemas()) == 10
+
+
+def test_get_tool_schemas_returns_independent_copies(bm):
+    """Mutating the returned list shouldn't affect the next call."""
+    p = bm.BasicMemoryProvider()
+    schemas = p.get_tool_schemas()
+    schemas.clear()
+    assert len(p.get_tool_schemas()) == 10
+
+
+def test_handle_tool_call_uninitialized(bm):
+    p = bm.BasicMemoryProvider()
+    out = json.loads(p.handle_tool_call("bm_search", {"query": "x"}))
+    assert "error" in out
+
+
+def test_handle_tool_call_unknown_tool(bm):
+    p = bm.BasicMemoryProvider()
+    p._initialized = True
+    p._actor = MagicMock()
+    out = json.loads(p.handle_tool_call("bm_bogus", {}))
+    assert "error" in out
+
+
+def test_handle_tool_call_dispatches(bm):
+    p = bm.BasicMemoryProvider()
+    p._initialized = True
+    p._project = "proj"
+    actor = MagicMock()
+    actor.call.return_value = json.dumps({"results": []})
+    p._actor = actor
+
+    out = p.handle_tool_call("bm_search", {"query": "hi", "limit": 3})
+    actor.call.assert_called_once()
+    bm_tool, bm_args = actor.call.call_args[0][:2]
+    assert bm_tool == "search_notes"
+    assert bm_args == {"project": "proj", "query": "hi", "page_size": 3}
+
+
+def test_handle_tool_call_missing_arg(bm):
+    p = bm.BasicMemoryProvider()
+    p._initialized = True
+    p._actor = MagicMock()
+    out = json.loads(p.handle_tool_call("bm_write", {"title": "x"}))
+    assert "error" in out
+
+
+def test_handle_tool_call_actor_failure(bm):
+    p = bm.BasicMemoryProvider()
+    p._initialized = True
+    p._project = "proj"
+    actor = MagicMock()
+    actor.call.side_effect = RuntimeError("boom")
+    p._actor = actor
+    out = json.loads(p.handle_tool_call("bm_search", {"query": "x"}))
+    assert "error" in out
+    assert p._failure_count == 1
+
+
+def test_circuit_breaker_opens_after_5_failures(bm, monkeypatch):
+    p = bm.BasicMemoryProvider()
+    p._initialized = True
+    p._project = "proj"
+    actor = MagicMock()
+    actor.call.side_effect = RuntimeError("boom")
+    p._actor = actor
+
+    for _ in range(5):
+        p.handle_tool_call("bm_search", {"query": "x"})
+
+    assert p._failure_pause_until > 0
+    assert p._is_circuit_open() is True
+
+
+def test_circuit_breaker_resets_after_pause(bm, monkeypatch):
+    p = bm.BasicMemoryProvider()
+    p._failure_count = 5
+    p._failure_pause_until = 1.0  # already in the past
+    monkeypatch.setattr(bm.time, "monotonic", lambda: 9999.0)
+    assert p._is_circuit_open() is False
+    assert p._failure_count == 0
+
+
+def test_session_note_title_with_session_id(bm):
+    from datetime import datetime, timezone
+
+    p = bm.BasicMemoryProvider()
+    p._session_started_at = datetime(2026, 5, 10, 13, 5, tzinfo=timezone.utc)
+    p._session_id = "20260510_080249_571920"
+    title = p._session_note_title()
+    # Date appears once; trailing random component is the disambiguator
+    assert title == "Hermes Session 2026-05-10 1305 571920"
+
+
+def test_session_note_title_no_session_id(bm):
+    from datetime import datetime, timezone
+
+    p = bm.BasicMemoryProvider()
+    p._session_started_at = datetime(2026, 5, 10, 13, 5, 42, tzinfo=timezone.utc)
+    p._session_id = ""
+    title = p._session_note_title()
+    # Falls back to seconds for disambiguation
+    assert title == "Hermes Session 2026-05-10 1305 42"
+
+
+def test_session_note_title_short_session_id(bm):
+    from datetime import datetime, timezone
+
+    p = bm.BasicMemoryProvider()
+    p._session_started_at = datetime(2026, 5, 10, 13, 5, tzinfo=timezone.utc)
+    p._session_id = "abcdef"
+    title = p._session_note_title()
+    # No `_` in id → use last 6 chars
+    assert title == "Hermes Session 2026-05-10 1305 abcdef"
+
+
+def test_system_prompt_block_uninitialized(bm):
+    p = bm.BasicMemoryProvider()
+    assert p.system_prompt_block() == ""
+
+
+def test_system_prompt_block_mentions_tools(bm):
+    p = bm.BasicMemoryProvider()
+    p._initialized = True
+    p._project = "test-proj"
+    p._mode = "local"
+    out = p.system_prompt_block()
+    assert "bm_search" in out
+    assert "test-proj" in out
+    assert "local" in out
+
+
+def test_system_prompt_block_steers_away_from_cli(bm):
+    """
+    Felix's training data is biased toward `bm tool ...` CLI patterns. The
+    system_prompt_block must explicitly direct the model to the bm_* tools
+    AND give a reason (latency) so it has a justification for following the
+    directive. Without this nudge, agents reach for bash/terminal tools and
+    pay 1-2s per call instead of ~0.1s.
+    """
+    p = bm.BasicMemoryProvider()
+    p._initialized = True
+    p._project = "test-proj"
+    p._mode = "local"
+    out = p.system_prompt_block().lower()
+    # Directive: don't use the bm CLI
+    assert "do not shell out" in out or "do not use the" in out or "do not run" in out
+    assert "bm" in out and "cli" in out
+    # Reason given (latency / capture bypass)
+    assert "mcp" in out  # the persistent connection is the mechanism we cite
+    assert "spawn" in out or "process" in out  # cold-start cost is mentioned
+
+
+def test_save_config_writes_json(bm, tmp_path):
+    p = bm.BasicMemoryProvider()
+    p.save_config({"mode": "local", "project": "x", "capture_per_turn": "true"}, str(tmp_path))
+    written = json.loads((tmp_path / "basic-memory.json").read_text())
+    assert written["mode"] == "local"
+    assert written["project"] == "x"
+    assert written["capture_per_turn"] is True  # coerced from "true"
+
+
+def test_save_config_merges_existing(bm, tmp_path):
+    cfg = tmp_path / "basic-memory.json"
+    cfg.write_text(json.dumps({"mode": "cloud", "project": "old"}))
+    p = bm.BasicMemoryProvider()
+    p.save_config({"project": "new"}, str(tmp_path))
+    after = json.loads(cfg.read_text())
+    assert after["mode"] == "cloud"  # preserved
+    assert after["project"] == "new"  # updated
+
+
+def test_load_config_missing_returns_empty(bm, tmp_path):
+    assert bm._load_config(str(tmp_path)) == {}
+
+
+def test_load_config_corrupt_returns_empty(bm, tmp_path):
+    (tmp_path / "basic-memory.json").write_text("{not json")
+    assert bm._load_config(str(tmp_path)) == {}
+
+
+def test_get_config_schema_shape(bm):
+    schema = bm.BasicMemoryProvider().get_config_schema()
+    keys = {entry["key"] for entry in schema}
+    assert {
+        "mode",
+        "project",
+        "project_path",
+        "capture_per_turn",
+        "capture_session_end",
+        "capture_folder",
+    }.issubset(keys)
+
+
+def test_register_appends_to_active_providers(bm):
+    fake_ctx = MagicMock()
+    bm._active_providers.clear()
+    bm.register(fake_ctx)
+    fake_ctx.register_memory_provider.assert_called_once()
+    assert len(bm._active_providers) == 1
+    bm._active_providers.clear()
+
+
+def test_register_also_registers_bundled_skill(bm):
+    """Plugin's bundled SKILL.md should auto-register when `hermes plugins install`
+    drops the repo into ~/.hermes/plugins/. Avoids the manual symlink step."""
+    fake_ctx = MagicMock()
+    bm._active_providers.clear()
+    bm.register(fake_ctx)
+    fake_ctx.register_skill.assert_called_once()
+    args, kwargs = fake_ctx.register_skill.call_args
+    # First positional arg is the bare skill name
+    assert args[0] == "basic-memory"
+    # Second positional arg is the SKILL.md path; it should resolve to a real file
+    assert args[1].name == "SKILL.md"
+    assert args[1].is_file()
+    bm._active_providers.clear()
+
+
+def test_register_tolerates_old_hermes_without_register_skill(bm):
+    """Older Hermes versions don't have ctx.register_skill; we shouldn't crash."""
+
+    class _OldCtx:
+        def __init__(self):
+            self.calls = []
+
+        def register_memory_provider(self, provider):
+            self.calls.append(("memory", provider))
+
+    ctx = _OldCtx()
+    bm._active_providers.clear()
+    bm.register(ctx)  # must not raise
+    assert any(c[0] == "memory" for c in ctx.calls)
+    bm._active_providers.clear()
+
+
+def test_register_swallows_register_skill_errors(bm, caplog):
+    """If register_skill raises (path validation, ABC mismatch, etc.) we log
+    and continue — don't break the memory-provider registration."""
+    fake_ctx = MagicMock()
+    fake_ctx.register_skill.side_effect = ValueError("invalid skill name")
+    bm._active_providers.clear()
+    with caplog.at_level("WARNING"):
+        bm.register(fake_ctx)
+    fake_ctx.register_memory_provider.assert_called_once()
+    assert "register_skill failed" in caplog.text
+    bm._active_providers.clear()
+
+
+# ---- Edge cases ----
+
+
+def test_handle_tool_call_with_none_args(bm):
+    """args=None must not crash; should be coerced to {} and surface a missing-arg error."""
+    p = bm.BasicMemoryProvider()
+    p._initialized = True
+    p._actor = MagicMock()
+    out = json.loads(p.handle_tool_call("bm_search", None))  # type: ignore[arg-type]
+    assert "error" in out
+
+
+def test_handle_tool_call_unknown_does_not_invoke_actor(bm):
+    p = bm.BasicMemoryProvider()
+    p._initialized = True
+    p._actor = MagicMock()
+    p.handle_tool_call("bm_does_not_exist", {"x": 1})
+    p._actor.call.assert_not_called()
+
+
+def test_translate_args_unknown_tool_raises(bm):
+    """Unknown tool names should raise KeyError so callers handle it explicitly."""
+    import pytest
+
+    with pytest.raises(KeyError):
+        bm._translate_args("not_a_tool", {}, "proj")
+
+
+# ---- Version metadata ----
+
+
+def test_module_version_present(bm):
+    import re
+
+    assert hasattr(bm, "__version__")
+    assert isinstance(bm.__version__, str)
+    # Matches the Python release versions written by scripts/update_versions.py.
+    assert re.fullmatch(r"\d+\.\d+\.\d+(?:(?:b|rc)\d+)?", bm.__version__), bm.__version__
+
+
+def test_module_version_matches_plugin_yaml(bm):
+    """plugin.yaml ships to Hermes; __version__ is what tooling reads. Keep them in sync."""
+    import os
+    import re
+
+    plugin_yaml = os.path.join(
+        os.path.dirname(os.path.abspath(bm.__file__)),
+        "plugin.yaml",
+    )
+    text = open(plugin_yaml).read()
+    m = re.search(r"^\s*version\s*:\s*(\S+)\s*$", text, re.MULTILINE)
+    assert m is not None, "plugin.yaml is missing a version field"
+    assert m.group(1) == bm.__version__, (
+        f"plugin.yaml version ({m.group(1)}) doesn't match __version__ ({bm.__version__})"
+    )
+
+
+# ---- uv bootstrap ----
+
+
+def test_install_bm_via_uv_no_uv(bm, monkeypatch):
+    """If uv isn't available, install returns None without trying to spawn."""
+    monkeypatch.setattr(bm, "_uv_binary_path", lambda: None)
+    assert bm._install_bm_via_uv() is None
+
+
+def test_install_bm_via_uv_runs_uv_tool_install(bm, monkeypatch):
+    """Install shells out to `uv tool install basic-memory`."""
+    calls: list = []
+
+    class _Result:
+        returncode = 0
+        stdout = b""
+        stderr = b""
+
+    def _fake_run(argv, **kwargs):
+        calls.append((argv, kwargs))
+        return _Result()
+
+    monkeypatch.setattr(bm, "_uv_binary_path", lambda: "/fake/uv")
+    monkeypatch.setattr(bm.subprocess, "run", _fake_run)
+    monkeypatch.setattr(bm, "_bm_binary_path", lambda: "/fake/bm-after-install")
+
+    result = bm._install_bm_via_uv()
+    assert result == "/fake/bm-after-install"
+    assert len(calls) == 1
+    argv = calls[0][0]
+    assert argv[0] == "/fake/uv"
+    assert argv[1:] == ["tool", "install", "basic-memory", "--quiet"]
+
+
+def test_install_bm_via_uv_failed_returncode(bm, monkeypatch):
+    """Non-zero exit logs and returns None — doesn't pretend success."""
+
+    class _Result:
+        returncode = 2
+        stdout = b""
+        stderr = b"network unreachable"
+
+    monkeypatch.setattr(bm, "_uv_binary_path", lambda: "/fake/uv")
+    monkeypatch.setattr(bm.subprocess, "run", lambda *a, **kw: _Result())
+
+    assert bm._install_bm_via_uv() is None
+
+
+def test_install_bm_via_uv_subprocess_exception(bm, monkeypatch):
+    """If subprocess raises (timeout, OSError, etc.) we degrade to None, not crash."""
+    monkeypatch.setattr(bm, "_uv_binary_path", lambda: "/fake/uv")
+
+    def _raise(*a, **kw):
+        raise OSError("boom")
+
+    monkeypatch.setattr(bm.subprocess, "run", _raise)
+    assert bm._install_bm_via_uv() is None
+
+
+def test_initialize_invokes_uv_install_when_bm_missing(bm, monkeypatch, tmp_path):
+    """Cold-start path: bm absent, uv present → initialize triggers install."""
+    install_calls: list = []
+
+    def _fake_install():
+        install_calls.append(True)
+        return None  # install reports failure; initialize logs and returns
+
+    monkeypatch.setattr(bm, "_MCP_AVAILABLE", True)
+    monkeypatch.setattr(bm, "_bm_binary_path", lambda: None)
+    monkeypatch.setattr(bm, "_uv_binary_path", lambda: "/fake/uv")
+    monkeypatch.setattr(bm, "_install_bm_via_uv", _fake_install)
+
+    p = bm.BasicMemoryProvider()
+    p.initialize(session_id="test", hermes_home=str(tmp_path))
+
+    assert install_calls == [True], "expected initialize() to attempt the install"
+    assert p._initialized is False  # install reported failure → don't start actor
+
+
+def test_initialize_skips_uv_install_when_bm_present(bm, monkeypatch, tmp_path):
+    """Steady-state path: bm already installed → no install attempt."""
+    install_calls: list = []
+
+    monkeypatch.setattr(bm, "_bm_binary_path", lambda: "/fake/bm")
+
+    def _fake_install():
+        install_calls.append(True)
+        return "/should-not-be-called"
+
+    monkeypatch.setattr(bm, "_install_bm_via_uv", _fake_install)
+    monkeypatch.setattr(bm, "_MCP_AVAILABLE", False)  # short-circuit actor start
+
+    p = bm.BasicMemoryProvider()
+    p.initialize(session_id="test", hermes_home=str(tmp_path))
+
+    assert install_calls == [], "should not invoke install when bm is already present"
+
+
+def test_initialize_bails_when_no_bm_no_uv(bm, monkeypatch, tmp_path, caplog):
+    """No bm, no uv → log clear error, don't try to install, don't initialize."""
+    monkeypatch.setattr(bm, "_MCP_AVAILABLE", True)
+    monkeypatch.setattr(bm, "_bm_binary_path", lambda: None)
+    monkeypatch.setattr(bm, "_uv_binary_path", lambda: None)
+
+    install_calls: list = []
+    monkeypatch.setattr(bm, "_install_bm_via_uv", lambda: install_calls.append(True) or None)
+
+    p = bm.BasicMemoryProvider()
+    with caplog.at_level("ERROR"):
+        p.initialize(session_id="test", hermes_home=str(tmp_path))
+
+    assert install_calls == []  # never attempted — no uv to call
+    assert p._initialized is False
+    assert "uv is not installed" in caplog.text or "uv" in caplog.text.lower()
+
+
+# ---- Defaults: stay out of bm's app dir ----
+
+
+def test_default_project_name_is_hermes_memory(bm):
+    """The default project name no longer carries a hostname suffix.
+    Each machine has its own isolated local store with this same name."""
+    assert bm._default_project() == "hermes-memory"
+
+
+def test_default_project_path_is_in_user_space(bm):
+    """~/.basic-memory/ is reserved for bm's app state. Projects live in user space."""
+    p = bm._default_project_path()
+    assert ".basic-memory" not in p, (
+        f"default project path must not live inside ~/.basic-memory/, got: {p}"
+    )
+    assert p.rstrip("/").endswith("hermes-memory")
+
+
+# ---- bm config introspection ----
+
+
+def test_bm_known_projects_missing_file(bm, monkeypatch, tmp_path):
+    """When bm has never been run, return None (callers should treat as 'unknown')."""
+    monkeypatch.setattr(bm, "_bm_config_path", lambda: tmp_path / "config.json")
+    assert bm._bm_known_projects() is None
+
+
+def test_bm_known_projects_corrupt_file(bm, monkeypatch, tmp_path):
+    cfg = tmp_path / "config.json"
+    cfg.write_text("{not json")
+    monkeypatch.setattr(bm, "_bm_config_path", lambda: cfg)
+    assert bm._bm_known_projects() is None
+
+
+def test_bm_known_projects_returns_dict(bm, monkeypatch, tmp_path):
+    cfg = tmp_path / "config.json"
+    cfg.write_text(json.dumps({"projects": {"main": {}, "hermes-memory": {}}}))
+    monkeypatch.setattr(bm, "_bm_config_path", lambda: cfg)
+    result = bm._bm_known_projects()
+    assert isinstance(result, dict)
+    assert set(result.keys()) == {"main", "hermes-memory"}
+
+
+def test_bm_known_projects_handles_non_dict_root(bm, monkeypatch, tmp_path):
+    cfg = tmp_path / "config.json"
+    cfg.write_text(json.dumps(["not a dict"]))
+    monkeypatch.setattr(bm, "_bm_config_path", lambda: cfg)
+    assert bm._bm_known_projects() is None
+
+
+# ---- Project verification ----
+
+
+def test_verify_project_registered_no_bm_config(bm, monkeypatch):
+    """No bm config yet → assume registration is fine; let downstream surface real failures."""
+    monkeypatch.setattr(bm, "_bm_known_projects", lambda: None)
+    p = bm.BasicMemoryProvider()
+    p._project = "anything-goes"
+    assert p._verify_project_registered() is True
+
+
+def test_verify_project_registered_present(bm, monkeypatch):
+    monkeypatch.setattr(bm, "_bm_known_projects", lambda: {"hermes-memory": {}, "main": {}})
+    p = bm.BasicMemoryProvider()
+    p._project = "hermes-memory"
+    assert p._verify_project_registered() is True
+
+
+def test_verify_project_registered_missing(bm, monkeypatch):
+    monkeypatch.setattr(bm, "_bm_known_projects", lambda: {"main": {}, "other": {}})
+    p = bm.BasicMemoryProvider()
+    p._project = "hermes-memory-cloud"
+    assert p._verify_project_registered() is False
+
+
+def test_log_missing_project_local_hint_includes_path(bm, caplog):
+    p = bm.BasicMemoryProvider()
+    p._mode = "local"
+    p._project = "hermes-memory"
+    p._project_path = "/tmp/somewhere"
+    with caplog.at_level("ERROR"):
+        p._log_missing_project()
+    msg = caplog.text
+    assert "hermes-memory" in msg
+    assert "/tmp/somewhere" in msg
+    assert "--cloud" not in msg
+
+
+def test_log_missing_project_cloud_hint_uses_cloud_flag(bm, caplog):
+    p = bm.BasicMemoryProvider()
+    p._mode = "cloud"
+    p._project = "hermes-memory-cloud"
+    with caplog.at_level("ERROR"):
+        p._log_missing_project()
+    msg = caplog.text
+    assert "hermes-memory-cloud" in msg
+    assert "--cloud" in msg
+
+
+# ---- initialize() bail-out on missing project ----
+
+
+def test_initialize_bails_when_project_missing(bm, monkeypatch, tmp_path):
+    """If bm config says the project doesn't exist, refuse to initialize."""
+    # bm config exists, but our project isn't in it
+    bm_cfg = tmp_path / ".basic-memory" / "config.json"
+    bm_cfg.parent.mkdir(parents=True)
+    bm_cfg.write_text(json.dumps({"projects": {"main": {}}}))
+    monkeypatch.setattr(bm, "_bm_config_path", lambda: bm_cfg)
+
+    # Cloud mode so _ensure_local_project doesn't auto-create
+    plugin_cfg = tmp_path / "basic-memory.json"
+    plugin_cfg.write_text(json.dumps({"mode": "cloud", "project": "not-registered"}))
+
+    p = bm.BasicMemoryProvider()
+    p.initialize(session_id="test", hermes_home=str(tmp_path))
+
+    assert p._initialized is False
+    assert p._actor is None
+
+
+def test_initialize_proceeds_when_bm_config_absent(bm, monkeypatch, tmp_path):
+    """If bm config doesn't exist (fresh install), don't false-reject — let actor try."""
+    monkeypatch.setattr(bm, "_bm_config_path", lambda: tmp_path / "no-such" / "config.json")
+    monkeypatch.setattr(bm, "_MCP_AVAILABLE", False)  # shortcut: actor won't actually start
+
+    plugin_cfg = tmp_path / "basic-memory.json"
+    plugin_cfg.write_text(json.dumps({"mode": "cloud", "project": "anything"}))
+
+    p = bm.BasicMemoryProvider()
+    # We don't fully assert _initialized here because the actor won't start without
+    # MCP — but we DO assert _verify_project_registered didn't gate us out before
+    # actor-start was attempted.
+    p.initialize(session_id="test", hermes_home=str(tmp_path))
+    # Initialization fails at actor-start (MCP unavailable), not at verify.
+    assert p._initialized is False  # expected — actor couldn't start
+    # _project should have been set despite the failure (proves we got past verify)
+    assert p._project == "anything"

integrations/openclaw/.githooks/pre-commit

@@ -0,0 +1,11 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+if ! command -v bun >/dev/null 2>&1; then
+  echo "bun is required to run pre-commit checks." >&2
+  exit 1
+fi
+
+echo "Running pre-commit checks..."
+bun run lint
+bun run check-types

integrations/openclaw/.github/workflows/ci.yml

@@ -0,0 +1,37 @@
+name: CI
+
+on:
+  pull_request:
+  push:
+    branches:
+      - main
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Setup Bun
+        uses: oven-sh/setup-bun@v2
+        with:
+          bun-version: "1.3.8"
+
+      - name: Install dependencies
+        run: bun install --frozen-lockfile
+
+      - name: Fetch skills
+        run: bun scripts/fetch-skills.ts
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Lint
+        run: bun run lint
+
+      - name: Typecheck
+        run: bun run check-types
+
+      - name: Unit tests
+        run: bun test

integrations/openclaw/.github/workflows/release.yml

@@ -0,0 +1,99 @@
+name: Release
+
+on:
+  workflow_dispatch:
+    inputs:
+      version:
+        description: "Version bump (`patch`, `minor`, `major`) or explicit semver (`0.2.0`)"
+        required: true
+        default: "patch"
+
+permissions:
+  contents: write
+  id-token: write
+
+concurrency:
+  group: release-${{ github.ref }}
+  cancel-in-progress: false
+
+jobs:
+  release:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+
+      - name: Setup Bun
+        uses: oven-sh/setup-bun@v2
+        with:
+          bun-version: "1.3.8"
+
+      - name: Setup Node
+        uses: actions/setup-node@v4
+        with:
+          node-version: "24"
+          registry-url: "https://registry.npmjs.org"
+
+      - name: Install dependencies
+        run: bun install --frozen-lockfile
+
+      - name: Fetch skills
+        run: bun scripts/fetch-skills.ts
+        env:
+          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
+
+      - name: Release checks
+        run: bun run release:check
+
+      - name: Configure Git identity
+        run: |
+          git config user.name "github-actions[bot]"
+          git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
+
+      - name: Bump version and create tag
+        id: bump
+        run: |
+          set -euo pipefail
+
+          VERSION_INPUT="${{ github.event.inputs.version }}"
+
+          if [[ "$VERSION_INPUT" =~ ^(patch|minor|major)$ ]]; then
+            npm version "$VERSION_INPUT" -m "chore(release): %s"
+          elif [[ "$VERSION_INPUT" =~ ^[0-9]+\.[0-9]+\.[0-9]+(-.+)?$ ]]; then
+            npm version "$VERSION_INPUT" -m "chore(release): %s"
+          else
+            echo "Unsupported version input: $VERSION_INPUT" >&2
+            echo "Use patch|minor|major or explicit semver like 0.2.0 or 0.2.0-alpha.1" >&2
+            exit 1
+          fi
+
+          TAG="$(git describe --tags --abbrev=0)"
+          VERSION="$(node -p "require('./package.json').version")"
+          echo "tag=$TAG" >> "$GITHUB_OUTPUT"
+          echo "version=$VERSION" >> "$GITHUB_OUTPUT"
+
+      - name: Push commit and tag
+        run: |
+          set -euo pipefail
+          git push origin HEAD:${GITHUB_REF_NAME}
+          git push origin "${{ steps.bump.outputs.tag }}"
+
+      - name: Publish to npm
+        run: |
+          VERSION="${{ steps.bump.outputs.version }}"
+          if [[ "$VERSION" == *-* ]]; then
+            DIST_TAG="${VERSION##*-}"   # e.g. alpha.6 -> alpha
+            DIST_TAG="${DIST_TAG%%.*}"  # strip .N suffix
+            npm publish --provenance --access public --tag "$DIST_TAG"
+          else
+            npm publish --provenance --access public
+          fi
+
+      - name: Create GitHub release
+        uses: softprops/action-gh-release@v2
+        with:
+          tag_name: ${{ steps.bump.outputs.tag }}
+          name: ${{ steps.bump.outputs.tag }}
+          generate_release_notes: true

integrations/openclaw/.gitignore

@@ -0,0 +1,146 @@
+# Logs
+logs
+*.log
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+lerna-debug.log*
+
+# Diagnostic reports (https://nodejs.org/api/report.html)
+report.[0-9]*.[0-9]*.[0-9]*.[0-9]*.json
+
+# Runtime data
+pids
+*.pid
+*.seed
+*.pid.lock
+
+# Directory for instrumented libs generated by jscoverage/JSCover
+lib-cov
+
+# Coverage directory used by tools like istanbul
+coverage
+*.lcov
+
+# nyc test coverage
+.nyc_output
+
+# Grunt intermediate storage (https://gruntjs.com/creating-plugins#storing-task-files)
+.grunt
+
+# Bower dependency directory (https://bower.io/)
+bower_components
+
+# node-waf configuration
+.lock-wscript
+
+# Compiled binary addons (https://nodejs.org/api/addons.html)
+build/Release
+
+# Dependency directories
+node_modules/
+jspm_packages/
+
+# Snowpack dependency directory (https://snowpack.dev/)
+web_modules/
+
+# TypeScript cache
+*.tsbuildinfo
+
+# Optional npm cache directory
+.npm
+
+# Optional eslint cache
+.eslintcache
+
+# Optional stylelint cache
+.stylelintcache
+
+# Optional REPL history
+.node_repl_history
+
+# Output of 'npm pack'
+*.tgz
+
+# Yarn Integrity file
+.yarn-integrity
+
+# dotenv environment variable files
+.env
+.env.*
+!.env.example
+
+# parcel-bundler cache (https://parceljs.org/)
+.cache
+.parcel-cache
+
+# Next.js build output
+.next
+out
+
+# Nuxt.js build / generate output
+.nuxt
+dist
+
+# Gatsby files
+.cache/
+# Comment in the public line in if your project uses Gatsby and not Next.js
+# https://nextjs.org/blog/next-9-1#public-directory-support
+# public
+
+# vuepress build output
+.vuepress/dist
+
+# vuepress v2.x temp and cache directory
+.temp
+.cache
+
+# Sveltekit cache directory
+.svelte-kit/
+
+# vitepress build output
+**/.vitepress/dist
+
+# vitepress cache directory
+**/.vitepress/cache
+
+# Docusaurus cache and generated files
+.docusaurus
+
+# Serverless directories
+.serverless/
+
+# FuseBox cache
+.fusebox/
+
+# DynamoDB Local files
+.dynamodb/
+
+# Firebase cache directory
+.firebase/
+
+# TernJS port file
+.tern-port
+
+# Stores VSCode versions used for testing VSCode extensions
+.vscode-test
+
+# yarn v3
+.pnp.*
+.yarn/*
+!.yarn/patches
+!.yarn/plugins
+!.yarn/releases
+!.yarn/sdks
+!.yarn/versions
+
+# Vite logs files
+vite.config.js.timestamp-*
+vite.config.ts.timestamp-*
+
+.idea/
+benchmark/datasets/
+benchmark/corpus-locomo/
+
+# Generated from ../../../skills by scripts/fetch-skills.ts
+skills/

integrations/openclaw/AGENTS.md

@@ -0,0 +1 @@
+CLAUDE.md

integrations/openclaw/BASIC_MEMORY.md

@@ -0,0 +1,208 @@
+# Basic Memory CLI & Cloud Setup
+
+## How the Plugin Connects
+
+Basic Memory is **local-first** — everything works out of the box with no cloud account, no internet connection, and no external services. Your notes live as markdown files on disk, indexed locally with SQLite.
+
+The plugin spawns a Basic Memory MCP session via stdio:
+
+```
+bm mcp --transport stdio --project <name>
+```
+
+All tool calls route through this MCP session. No cloud configuration is required for normal use.
+
+## Cloud Configuration (Optional)
+
+Cloud sync is entirely optional. If you want to sync your local knowledge base to Basic Memory Cloud for backup or cross-device access, you can configure per-project cloud routing.
+
+### Why Cloud?
+
+- **Your agent's memory travels with you** — laptop, desktop, hosted environment. Same knowledge graph everywhere, synced bidirectionally.
+- **Team knowledge sharing** — org workspaces let multiple agents and team members build on a shared knowledge base.
+- **Durable memory for production agents** — CI runners, containers, and hosted environments are ephemeral. Cloud gives agents persistent memory that survives teardowns.
+- **Multi-agent coordination** — multiple agents (or the same agent across services) can read and write to a shared graph.
+
+Cloud extends local-first — it doesn't replace it. Your notes are still plain markdown, still editable locally, still yours. Start with a [7-day free trial](https://basicmemory.com) — no credit card required. Use code `BMCLAW` for 20% off for 3 months.
+
+### Setup
+
+```bash
+# Authenticate with Basic Memory Cloud
+bm cloud login
+
+# Save API key for per-project cloud routing
+bm cloud set-key bmc_...
+
+# Route a project through the cloud
+bm project set-cloud <name>
+
+# Revert a project to local routing
+bm project set-local <name>
+
+# Check cloud connection state
+bm cloud status
+```
+
+When a project is set to cloud mode, the MCP server routes tool calls for that project through the cloud API using the saved API key as a Bearer token. Local projects (the default) continue to use the local SQLite index. You can mix local and cloud projects freely.
+
+## Project Management
+
+```bash
+# List all projects
+bm project list
+
+# Add a new project
+bm project add "name" ~/path
+
+# Show current project details
+bm project info
+
+# Set the default project
+bm project default "name"
+
+# One-way sync (local -> cloud)
+bm project sync
+
+# Bidirectional sync
+bm project bisync
+```
+
+## Cross-Project Operations
+
+All plugin tools accept an optional `project` parameter to operate on a different project:
+
+```
+search_notes(query="authentication", project="other-project")
+read_note(identifier="notes/api-design", project="docs")
+write_note(title="New Note", content="...", folder="notes", project="research")
+```
+
+## Workspace Support
+
+Workspaces group projects by owner (personal or organization):
+
+```
+list_workspaces()
+
+list_memory_projects(workspace="my-org")
+```
+
+## Auto-Recall
+
+When `autoRecall` is enabled (the default), the plugin injects relevant context at the start of each agent session by listening for the `agent_start` event. On each trigger it:
+
+1. **Queries active tasks** — searches the knowledge graph for notes with `type: Task` and `status: active` (up to 5 results)
+2. **Fetches recent activity** — gets notes modified in the last 24 hours
+3. **Formats and injects context** — returns the results as structured context for the agent
+
+The injected context looks like:
+
+```
+## Active Tasks
+- **Fix login bug** — Description of Fix login bug
+- **Update API docs** — Description of Update API docs
+
+## Recent Activity
+- Daily standup notes (memory/daily-standup-notes.md)
+- API design decisions (memory/api-design-decisions.md)
+
+---
+Check for active tasks and recent activity. Summarize anything relevant to the current session.
+```
+
+The trailing instruction (after `---`) is the `recallPrompt`, which you can customize to change what the agent focuses on. For example:
+
+```json
+{
+  "recallPrompt": "Focus on blocked tasks and any decisions made in the last 24 hours."
+}
+```
+
+To disable auto-recall entirely:
+
+```json
+{
+  "autoRecall": false
+}
+```
+
+## Auto-Capture
+
+When `autoCapture` is enabled (the default), the plugin automatically records agent conversations after each turn:
+
+1. Extracts the last user + assistant messages
+2. Appends them as timestamped entries to a daily conversation note (`conversations-YYYY-MM-DD`)
+3. Skips very short exchanges (< `captureMinChars` chars each, default 10)
+
+This builds a searchable history of agent interactions in the knowledge graph without any manual effort.
+
+## Slash Commands
+
+### Memory commands
+
+- **`/remember <text>`** — Save a quick note to the knowledge graph
+- **`/recall <query>`** — Search the knowledge graph (top 5 results)
+
+### Skill workflows
+
+These commands inject step-by-step workflow instructions from the bundled skill files:
+
+| Command | What it does |
+|---------|-------------|
+| `/tasks` | Task management — create, track, resume structured tasks that survive context compaction |
+| `/reflect` | Memory reflection — review recent activity and consolidate insights into long-term memory |
+| `/defrag` | Memory defrag — reorganize, split, prune, and clean up memory files |
+| `/schema` | Schema management — infer, create, validate, and evolve Picoschema definitions |
+
+Each command accepts optional arguments for context:
+
+```
+/tasks create a task for the API migration
+/reflect focus on decisions from this week
+/defrag clean up completed tasks older than 2 weeks
+/schema infer a schema for Meeting notes
+```
+
+When invoked without arguments, the agent receives the full workflow instructions and follows them interactively.
+
+## Plugin Configuration
+
+The plugin accepts these config fields in `openclaw.config.json`:
+
+```json
+{
+  "plugins": {
+    "entries": {
+      "openclaw-basic-memory": {
+        "enabled": true,
+        "config": {
+          "project": "my-project",
+          "memoryDir": "memory/",
+          "memoryFile": "MEMORY.md",
+          "autoCapture": true,
+          "autoRecall": true,
+          "recallPrompt": "Check for active tasks and recent activity. Summarize anything relevant.",
+          "debug": false
+        }
+      }
+    },
+    "slots": {
+      "memory": "openclaw-basic-memory"
+    }
+  }
+}
+```
+
+| Field | Default | Description |
+|-------|---------|-------------|
+| `project` | `openclaw-<hostname>` | BM project name |
+| `bmPath` | `bm` | Path to BM CLI binary |
+| `memoryDir` | `memory/` | Relative path for memory files |
+| `memoryFile` | `MEMORY.md` | Working memory file name |
+| `projectPath` | same as `memoryDir` | Absolute path to project root |
+| `autoCapture` | `true` | Auto-index conversations after each turn |
+| `captureMinChars` | `10` | Min chars to trigger capture |
+| `autoRecall` | `true` | Inject context (active tasks, recent activity) at session start |
+| `recallPrompt` | *(see above)* | Instruction appended to recalled context |
+| `debug` | `false` | Verbose logging |