SkillClaw

Why we included this project

Any agent that speaks an OpenAI-compatible API, whether Hermes, OpenClaw, Codex, or Claude Code, tends to pile up a skill library full of duplicates and half-finished fragments, since nothing digests what the agent actually learns from real sessions. SkillClaw runs as a local proxy that intercepts the agent's chat traffic, records session artifacts, then deduplicates, improves, and verifies reusable skills in the background without interrupting your workflow. It also merges skill libraries across agents, devices, and shared team groups, so a debugging win by one teammate can update the shared skill once and benefit everyone else right away. The optional evolve server extends this to multi-user setups, and a bundled dashboard shows local and shared skills, validation progress, version history, and session traces.