416rehmanandClaude Opus 5
523ed1a738
A pipeline records what it concluded. Whether a person has verified that conclusion is a separate claim, and nothing in a run can make it, so the report now carries marks the reader adds: confirmed, or outstanding with a note naming what is still unproven. Deliberately generic. What counts as outstanding is the reader's business, and reproducing a crash, proving a precondition or reading a diff are the same shape of unfinished work as far as the report is concerned, so it only provides somewhere to put it. Marks show against each row and can be filtered on, including the results nobody has reviewed yet. The report is a file, so they are kept in the browser and keyed on the pipeline and target rather than the path, which means a rerun over the same corpus keeps them. Exporting writes marks.json; saved next to the report it is read back at generation time, so the marks render for anyone who opens it and reach inventory.csv alongside everything else instead of being stranded in one person's browser. Closes #24 Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Automated vulnerability research pipeline engine
Define pipelines as YAML. DeepZero handles orchestration, parallelism, fault tolerance, and state.
- 🔗 Pipeline-as-YAML - chain ingest, filter, transform, and LLM-assess stages declaratively
- ⚡ Parallel execution - ThreadPoolExecutor with configurable concurrency per stage
- 💾 Resumable runs - atomic per-sample state on disk; Ctrl+C and re-run to pick up where you left off
- 🤖 LLM integration - Jinja2 prompt templates with any LLM provider via LiteLLM
- 🌐 REST API (WIP) - query run state and sample data over HTTP (currently experimental and incomplete)
- 🧩 Extensible - write custom processors as Python classes, reference them by path in YAML
📚 Documentation
DeepZero features extensive, exhaustive documentation covering architecture, pipeline schemas, CLI references, and custom processor development.
👉 Read the Official Documentation here
⚡️ Quickstart
DeepZero requires a target corpus of files to analyze and a pipeline configuration detailing how to process them.
-
Clone & Install (Python 3.11+)
git clone https://github.com/416rehman/DeepZero.git cd DeepZero pip install -e . -
Configure Environment
cp .env.example .env -
Run a Pipeline
deepzero run C:\drivers -p .\pipelines\loldrivers\pipeline.yaml
For detailed setup instructions and example corpora, see the Quickstart Documentation.
📁 Repository Structure
src/deepzero/
├── api/ # REST API (starlette)
├── engine/ # orchestration, state persistence, pipeline execution
└── stages/ # built-in processors (map, reduce, ingest)
processors/ # external processors (shipped as examples)
├── ghidra_decompile/ # ghidra headless decompiler (MapProcessor)
├── loldrivers_filter/ # loldrivers.io hash exclusion filter (MapProcessor)
├── pe_ingest/ # PE header parser and driver metadata extractor (IngestProcessor)
└── semgrep_scanner/ # semgrep batch scanner (BulkMapProcessor)
pipelines/
└── loldrivers/ # BYOVD kernel driver vulnerability research pipeline
├── pipeline.yaml
├── assessment.j2 # LLM prompt template
└── rules/ # semgrep rules
docs/ # Jekyll-based GitHub Pages documentation
tests/ # pytest suite
🤝 Contributing
CI runs on Python 3.11 and 3.12 via GitHub Actions.
Run linting and security checks before submitting:
ruff check . && ruff format --check . && bandit -ll -ii -c pyproject.toml -r .
Please refer to the Contributing Guide and the Code of Conduct before submitting pull requests.
📄 License
DeepZero is released under the MIT License.