commit-sentinel is an open-source AI-powered guardrail for Git commits.
It analyzes diffs, commit messages, and contextual metadata to detect low-quality changes, AI-generated noise and risky code patterns before they enter your repository.
This project focuses on signal over noise, not stylistic nitpicking.
Modern repositories suffer from:
- AI-generated code without understanding
- Developers commit the code they do not understand
- Large, unfocused diffs
commit-sentinel acts as an automated sentinel at commit, enforces to commit the code that is fully clear for the developer.
- Analyzes Git diffs and commit messages
- Applies LLM-backed reasoning to detect:
- Meaningless or low-signal changes
- Vague AI-generated code
- No dashboard
- No authentication
- No billing
- No long-term state
- No training on your private code
- Local CLI (manual or pre-commit)
- Deterministic inputs, explainable outputs
- Human-readable failures
- LLMs as reasoning engines, not oracles
- Open-source core, no lock-in
- v0.1: CLI, single-rule pipeline
- v0.2: Duck for debugging
- v0.3: Local model support
- v0.x: Optional hosted services (separate)
Apache License 2.0
This project is in active early development. Expect breaking changes before v1.0.