self-learning

SKILL.md

Enables agents to learn from their execution history by capturing traces, extracting actionable insights, and adapting behavior. Transforms raw execution patterns into reusable knowledge structures that improve future performance.

| Success Patterns | Approaches that consistently work | "Loading context.md before task start reduces errors by 40%" | | Failure Patterns | Recurring causes of failures | " Skipping lspdiagnostics leads to type errors in production" | | Efficiency Patterns | Ways to complete tasks faster | "Reading multiple files in parallel reduces context load time" |

| Coordination Patterns | Best practices for multi-agent work | "Orchestrator spawning 2 workers instead of 3 reduces merge conflicts" | | Knowledge Gaps | Information that should have been available | "Project lacks centralized error handling documentation" |

查看原文