* Add E2E lifecycle test for create/destroy workflow
Bash script that exercises the full droplet lifecycle: create (no
Tailscale), verify SSH config, run commands over SSH, destroy, and
verify cleanup
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: dm <alexis.challande@trailofbits.com>
* Show actual disk size instead of size spec in resize command
After a --no-disk resize, `dropkit resize` showed the size spec's disk
(e.g. 80 GB for s-2vcpu-4gb) instead of the droplet's actual disk
(25 GB). This is confusing for users planning temporary scale-ups —
they need to see their real disk to know if they can scale back down.
Read disk from `droplet["disk"]` (actual allocation) instead of
`droplet["size"]["disk"]` (what the size tier offers). When the two
differ, show an inline hint: "25 GB (size spec: 80 GB, disk was not
resized)" so the situation is immediately clear.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Use actual droplet vcpus and memory, not just disk
Read all three resource fields from the droplet object directly
(droplet["vcpus"], droplet["memory"], droplet["disk"]) instead of
from the size spec (droplet["size"]["vcpus"], etc.). After a
--no-disk resize these can diverge — the size spec reflects the
new tier while the droplet retains its original resources.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Show per-resource and total monthly costs in the list/ls command so
engineers can see at a glance what their infrastructure is costing.
* Droplet cost from DO API size object (price_monthly field)
* Snapshot cost calculated at $0.06/GB/month (DO standard rate)
* Total cost shown in summary line at the bottom
* --cost/--no-cost flag to toggle the column (shown by default)
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Offer to delete leftover hibernation snapshot on droplet destroy
When a user hibernates a droplet then wakes it up but declines to delete
the snapshot, the snapshot is left behind. If they later `dropkit destroy`
the droplet, the orphaned snapshot persists silently, incurring storage
costs. After destroying the droplet, check for a matching hibernation
snapshot and prompt the user to delete it.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Change snapshot deletion default to 'no' in destroy command
Respect the user's earlier choice: if they declined to delete the
snapshot during wakeup, defaulting to 'yes' here would second-guess
that decision. Default to 'no' so the prompt is informational rather
than presumptive.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* Update default Ubuntu image from 25.04 to 25.10
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Extract default region/size/image slugs into constants
Replace duplicated hardcoded strings across config.py, main.py, and
tests with DEFAULT_REGION, DEFAULT_SIZE, and DEFAULT_IMAGE constants
defined in config.py.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* Remove escaped chars in comments
* Add Jinja2 template validation test and pre-commit hook
Catch template syntax errors early with a unit test that parses and
renders the default cloud-init template, and a pre-commit hook that
validates all templates in dropkit/templates/ on change.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* Fix Jinja2 escaping of zsh PROMPT in cloud-init template
The zsh PROMPT string contains `%` sequences that Jinja2 interprets
as template syntax. Wrap in {% raw %}...{% endraw %} to preserve
the literal prompt definition.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Add a warning to the cloud-init jinja template
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Ran /claude-md-management:revise-claude-md to audit CLAUDE.md against
the actual codebase. The file-level tree was stale (missing lock.py,
ui.py, version_check.py and several test files). Replace with a
directory-only layout that won't drift when files are added.
Also document the --cov-fail-under=29 threshold from pyproject.toml.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add a Tailscale IP column to `dropkit ls` output. The IP is read from
the SSH config — if the configured hostname is a Tailscale IP (100.x.x.x),
it's displayed; otherwise shows "—".
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Update both `prepare_for_hibernate` and `destroy` to switch SSH config
to the droplet's public IP before logging out of Tailscale. Previously,
the Tailscale logout happened first, which killed the Tailscale IP route
that SSH was using — causing the logout command itself to hang or fail.
Also adds `--verbose` flag to the `hibernate` command.
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* allow resizing hibernated droplets by modifying the size tag
* Prevent removal of protected tags (owner, firewall) in untag_resource
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* Fix macOS Tailscale CLI discovery for App Store installs
On macOS, Tailscale installed via the App Store places its CLI at
/Applications/Tailscale.app/Contents/MacOS/Tailscale, which is not in
PATH. Add find_tailscale_cli() helper that checks PATH first, then
falls back to the macOS App Store location on darwin.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* Address PR feedback: better error message, remove excessive test
Distinguish "Tailscale not found" from "not running" when the local
check fails, so users know whether to install or start Tailscale.
Remove the overly comprehensive binary_vanishes race condition test.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
- Rename .claude/settings.local.json to .claude/settings.json
- Add .claude/settings.local.json to .gitignore
- Replace uv run tobcloud with uv run dropkit in permissions
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* Generalize email domain handling in documentation and tests
- Add .worktrees/ to .gitignore for git worktree support
- Update _sanitize_email_for_username docstring to be domain-agnostic
- Change CLAUDE.md example from trailofbits.com to example.com
- Add tests for google.com, gmail.com, corporate, and plus addressing
The code already supported any email domain; this updates docs and
tests to reflect that capability.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* Address PR review comments
- Remove redundant trailofbits.com regex, simplify to split("@")[0]
- Rename test to clarify backwards compatibility intent
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
Co-authored-by: Dan Guido <dan@trailofbits.com>
Replace manual file parsing with ssh-keygen -R which properly handles
both hashed (|1|...) and unhashed entries. macOS and many Linux systems
use HashKnownHosts by default, making the previous implementation unable
to remove entries.
Changes:
- Use ssh-keygen -R for each hostname instead of parsing the file
- Detect successful removal via "updated" in stdout
- Backup files are now .old (ssh-keygen default) instead of .bak
- Bracketed entries ([host]:port) now require exact format
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
* Rename tobcloud references to dropkit in tests
Update test data and test method names to use the new project name.
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* Format test_ssh_config.py
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>
dependabot[bot]