What problem does it solve? Agent Skills often fail silently in production: workarounds go unrecorded, routing mistakes repeat, and quality regressions are never traced back to the skill contract. This Skill turns real-use evidence into durable improvements instead of relying on synthetic benchmarks. ## Core Features & Use Cases - Mandatory postmortems: After every material skill use, assess routing, outcome, quality delta, friction, and workarounds with a compact structured record. - Gated feedback issues: Open or update deduplicated GitHub issues only when actionable learning exists, with privacy sanitization rules. - Evidence-driven evolution: Diagnose feedback, apply the smallest justified change, add regression evals only when they protect a learned boundary, and decide between improving, branching, handing off, or creating a sibling skill. - Use Case: After three real tasks required the same workaround with a skill, use this Skill to consolidate the evidence into one feedback issue, diagnose the cause, and apply the minimal instruction fix. ## Quick Start Use the loop-engineering skill to turn the postmortem from my last skill use into a deduplicated feedback issue and the smallest justified change.
