# Tacit

- **Event:** [Built with Opus 4.6: a Claude Code hackathon](https://cerebralvalley.ai/e/claude-code-hackathon)
- **When:** Feb 10 at 12:00 PM – Feb 17 at 10:00 AM (EST)
- **Where:** Location TBA
- **Team:** [Bayram Annakov](https://cerebralvalley.ai/u/Bayka)
- **GitHub:** https://github.com/BayramAnnakov/tacit
- **Demo video:** https://youtu.be/JEyUfCVTNN4
- **Gallery:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery/40

Tacit extracts unwritten rules your team enforces but never documented. Point it at a GitHub repo — 16 Agent SDK agents mine PR reviews, CI failures, and CHANGES_REQUESTED discussions, then output CLAUDE.md and `.claude/rules/` files Claude Code loads automatically.

The key: anti-pattern mining from rejected PRs extracts "NEVER do X" rules linking each to the exact PR comment where the team learned the lesson. 57 agent runs, 20 MCP tools, ~$13/extraction. 4-phase pipeline with Opus for deep analysis, Sonnet for scanning.

Results: 343 rules from 50 PRs, 100% with provenance. Eval suite: 83%. Rediscovered 87% of documented guidelines from PR data alone — without reading the existing CLAUDE.md.

---

Markdown version of https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery/40. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt