Hi HN! I notice I have a lot of session data and wonder if everything I put into AGENTS.md is actually being followed in my sessions. And I realized I can also mine agent's repeated mistakes and suggest is as rules to be added, or even actual enforcement. That's why I built this fully local tool to mine your local sessions and audit existing AGENTS.md. The rules are mostly heuristic and patterns I see from actual real world AGENTS.md. Audit is deterministic. But it can also outputs evidence for your agent to synthesize into suggestions. Same thing for mining new rules, it outputs evidence with some context window, then your agent synthesize it into actionable rules.
Let me know if you got any surprising suggestion from it!
Let me know if you got any surprising suggestion from it!