AMAP-ML/SkillClaw

What is it?

Key features

• 01Local proxy exposing /v1/chat/completions and /v1/messages endpoints for agents.
• 02Records session artifacts and manages a local SKILL.md skill library.
• 03Optional evolve server that deduplicates and improves skills from shared session data.
• 04Shared storage via local directory, Alibaba OSS, or S3, with group-based namespaces.
• 05Optional validation step where opted-in clients review candidate skills before publication.
• 06Local dashboard for inspecting skills, validation progress and session traces.

Use cases

• A single developer who wants a coding agent's skills to improve automatically over time.
• A team sharing one skill library across several engineers' agents.
• One user running multiple agents or devices that should share learned skills.
• Testing skill evolution on a local machine with a local shared directory before adding cloud storage.

Quick start

1. 1

   Clone the repository and run the macOS/Linux install script

   git clone https://github.com/AMAP-ML/SkillClaw.git && cd SkillClaw
2. 2

   Run the bundled install script from the repository

   bash scripts/install_skillclaw.sh
3. 3

   Activate the virtual environment

   source .venv/bin/activate
4. 4

   Generate a local config with the setup wizard

   skillclaw setup

Things to know

• !Needs an upstream provider account exposing an OpenAI-compatible API or AWS Bedrock.
• !Python 3.10 or newer is required.
• !Shared evolution across users requires shared storage (OSS, S3, or local) and an evolve server.

In collections

Topics

FAQ

What is SkillClaw?

A local API proxy for AI agents that records sessions and evolves shared skill libraries across users, agents and devices. SkillClaw sits between an AI agent and its LLM provider as a local OpenAI-compatible proxy. It captures session artifacts from real interactions, manages a local skill library in SKILL.md format, and lets the agent keep improving its skills without extra effort from the user.

Is SkillClaw free to use? What license does it have?

SkillClaw is open source under the MIT license. Check the license terms before using it in commercial products.

Is SkillClaw still maintained?

Actively maintained. The last commit to SkillClaw was on Oct 21, 2026.

What can I use SkillClaw for?

A single developer who wants a coding agent's skills to improve automatically over time. A team sharing one skill library across several engineers' agents. One user running multiple agents or devices that should share learned skills. Testing skill evolution on a local machine with a local shared directory before adding cloud storage.

How do I install SkillClaw?

Clone the repository and run the macOS/Linux install script: git clone https://github.com/AMAP-ML/SkillClaw.git && cd SkillClaw Run the bundled install script from the repository: bash scripts/install_skillclaw.sh Activate the virtual environment: source .venv/bin/activate Generate a local config with the setup wizard: skillclaw setup

Where can I get SkillClaw?

The source code and installation instructions are on GitHub: https://github.com/AMAP-ML/SkillClaw