What problem does it solve? Maintaining a large library of SKILL.md skill files manually leads to structural drift, stale metadata, missing I/O contracts, and unverified quality claims. This Skill runs a self-evolution engine that continuously probes, benchmarks, diagnoses, and patches the skill tree while keeping every change auditable in Git. ## Core Features & Use Cases - Eight-dimension diagnosis: Scores structural integrity, benchmark coverage, constitutional compliance, content quality, reference coverage, absorption potential, gene-layer liveness, and behavioral execution traces via diagnose.py. - Automated improvement loops: auto-loop.py runs unattended multi-cycle evolution with hard convergence guardrails (edit budgets, rejected buffers, burnout protection) and Git-as-memory commits per cycle. - External absorption pipeline: Searches GitHub, Hermes skills, and papers for complementary projects, scores them on five dimensions, and generates approval-gated absorption proposals with methodology extraction. - Use Case: A research agent system with 150+ SKILL.md files runs a nightly evolution cycle that detects 12 skills missing IO_CONTRACT sections, batch-injects them, re-runs benchmarks, commits the changes, and logs the score delta from 0.86 to 0.95. ## Quick Start Run the evolution engine to diagnose all skills, fix structural issues, and record the results in evolution-state.json and evolution-log.md.
