Mine historical bug-fix commits from Jira + GitHub, cluster recurring patterns, synthesize Scalafix (Scala) and Opengrep (TypeScript, JavaScript, Python, Go) rules, validate them against a held-out corpus, and ship them as PRs to a rules repository that gates future PRs via CI.
- Language-aware synthesis — Scala clusters become Scalafix rules; TS/JS/Python/Go clusters become Opengrep YAML rules
- Configurable mining — any Jira project + GitHub org; ticket pattern and repo list in
config.toml - Validation loop — precision/recall thresholds before auto-shipping rules
- PR review skill — Claude Code skill at
.claude/skills/repo-bug-review/for scanning local code or GitHub PRs - CI templates — copy workflows from
service-integration/into any target repo - Continuous learning — batch bootstrap plus optional per-bug continuous loop
- Python 3.11+
- Docker (Postgres + pgvector via
docker compose up -d) - uv or pip
- Jira + GitHub API credentials
- Anthropic API key (or compatible endpoint)
make install
cp config/config.example.toml config/config.toml
# Edit config.toml — Jira URL, GitHub org, repos, Claude base_url
export JIRA_USER=... JIRA_TOKEN=... GITHUB_TOKEN=... ANTHROPIC_API_KEY=...
docker compose up -d
python -m src.orchestrator schema
python -m src.orchestrator batch --repo backend --since 2024-05-01 --output out/
make testOpen the repo in Claude Code — CLAUDE.md imports AGENTS.md. Project skill:
.claude/skills/repo-bug-review/.
make install
make update # refresh deps after pulling mainRules live in .cursor/rules/000-project.mdc (points to AGENTS.md). Portable skills:
.agents/skills/.
make install
make updateCodex reads root AGENTS.md for repo policy. Portable skills: .agents/skills/.
make install
make update| Target | Purpose |
|---|---|
make install |
First-time setup (uv sync --extra dev) |
make update |
Refresh dependencies after pulling changes |
make uninstall |
Remove local venv and generated caches |
See docs/README.md for the full doc map.
See docs/CONTRIBUTING.md.
MIT — see LICENSE.
