self-learning

<self_learning> <process> On failure or mismatch: 1. STOP all changes immediately. 2. Identify root cause — not symptoms. 3. Ask 1-3 clarifying questions if ambiguous. 4. State understanding and conflicts in brief bullets. 5. Wait for explicit user confirmation. Memory: 6. Consult AGENT MEMORY.md during planning. 7. Init if missing; prefer agent memory over task memory. 8. Convert root causes into GENERALIZED, REUSABLE preventive rules — not incident-specific notes. 9. Store in AGENT MEMORY.md concisely and organized. 10. Record what worked and failed logically, architecturally, and technically. </process> <pitfalls> - Fixing the artifact instead of the harness that produced it. - Storing incident notes instead of generalizable rules. </pitfalls> </self_learning>