Lessons Learned Loop

Overview

This skill enforces persistent learning across agent sessions, preventing agents from repeating known mistakes or forgetting hard-won knowledge. It mandates that an agent reads a LESSONS.md file upon starting work and appends any non-trivial bugs encountered during the session before concluding.

It acts as a critical contract for complex workflows, ensuring that every run starts smarter than the last by leveraging accumulated institutional memory within the project's directory structure.

Capabilities

• Knowledge Retrieval: Reads LESSONS.md at the start of execution to inform current decisions.
• Failure Logging: Appends structured entries detailing bugs, attempted fixes, and outcomes upon completion.
• State Enforcement: Acts as a checkpoint that must be satisfied before declaring work complete.
• History Management: Handles appending new lessons without modifying or deleting older, critical records.

Example Use Cases

• Debugging Cycles: When an agent repeatedly fails on the same type of error (e.g., API rate limits), this loop ensures the fix is documented and prioritized for the next attempt.
• Complex System Integration: In multi-stage development, it guarantees that lessons from one module's failure inform the design constraints of a subsequent module.
• Process Refinement: After a major project milestone, running this skill forces a review of all encountered pitfalls to build a robust playbook for future iterations.