# SkillClaw

A background service that evolves an AI agent's skill library from every conversation, merging, deduplicating and improving skills across sessions, multiple agents and multiple devices for one user.

SkillClaw runs alongside an AI agent and evolves its skill library from ordinary use, without changing how a person works with the agent day to day. Skills are automatically evolved, deduplicated and quality-improved in the background as sessions happen.

When a user runs several agents, for example one for frontend work and one for backend work, SkillClaw merges and cross-pollinates their skills into one shared library and redistributes it to all of them. The same library also follows a user across devices, so skills learned on one machine are available on another instead of starting over each time.

It installs with a shell installer on macOS and Linux, or a manual Python install path on Windows, then runs as a background daemon. It natively integrates with Hermes, Codex, Claude Code, OpenClaw and several other agent harnesses, plus any OpenAI-compatible API.

Website: https://github.com/AMAP-ML/SkillClaw