SkillClaw: Let Skills Evolve Collectively with Agentic Evolver

Note

SkillClaw is a framework for skill collective evolution in multi-user OpenClaw-style agent ecosystems. It automatically distills real-world experience from multiple users and agents into reusable Skills, and shares them via the cloud to enable continuous evolution across the entire agent cluster.

中文

Easy to install, effortless to use — just two commands to set up, and supports a wide range of scenarios. Users simply interact with OpenClaw as usual, and skill evolution happens seamlessly in the background — zero extra effort required.

Broad framework support — SkillClaw natively integrates with various Claw frameworks, including CoPaw, IronClaw, PicoClaw, ZeroClaw, NanoClaw, and NemoClaw.

Proven results — In real-world scenario evaluations on WildClawBench, SkillClaw significantly improved Qwen3-Max's performance even under limited group interaction and feedback conditions — not by using a bigger model, but by leveraging smarter experience.

---

Overview

SkillClaw makes LLM agents progressively better by evolving reusable skills from real session data and sharing them across a group of agents.

The system has three components:

1. Client Proxy — A local API proxy (/v1/chat/completions, /v1/messages) that intercepts agent requests, records session artifacts, and syncs skills with shared storage.

2. Workflow Evolve Server (evolve_server) — A fixed 3-stage LLM workflow (Summarize → Aggregate → Execute) that reads session data from shared storage, evolves or creates skills, and writes them back.

3. Agent Evolve Server (agent_evolve_server) — An autonomous agent-driven alternative to the workflow evolve server. Uses an OpenClaw agent to read sessions, analyze patterns, and directly write evolved skill files with full tool access (read/write/exec).

All three share the same storage layer (Alibaba OSS / S3 / local filesystem) and skill format (SKILL.md), so they are fully interchangeable in a deployment.

---

Quick Start

Prerequisites

• Python >= 3.10
• An OpenAI-compatible LLM API endpoint

1. Install

Choose one:

# Client / local development
git clone <repo-url> SkillClaw && cd SkillClaw
bash scripts/install_skillclaw.sh
source .venv/bin/activate

# Server deployment (installs both server entrypoints)
bash scripts/install_skillclaw_server.sh
source .venv-server/bin/activate

# Only needed for the agent evolve server
npm install -g openclaw

2. Configure & Start the Client Proxy

export OPENAI_BASE_URL="https://your-api-gateway/v1"
export OPENAI_API_KEY="sk-..."

skillclaw setup
skillclaw start
skillclaw status
skillclaw config show

3. Start the Evolve Server

SkillClaw provides two evolve servers — pick either one:

Workflow evolve server:

skillclaw-evolve-server --port 8787 --interval 300 \
  --storage-backend oss \
  --oss-endpoint "$EVOLVE_STORAGE_ENDPOINT" \
  --oss-bucket "$EVOLVE_STORAGE_BUCKET" \
  --group-id my-group

Agent evolve server:

skillclaw-agent-evolve-server --port 8787 --interval 300 --no-fresh \
  --storage-backend oss \
  --oss-endpoint "$EVOLVE_STORAGE_ENDPOINT" \
  --oss-bucket "$EVOLVE_STORAGE_BUCKET" \
  --group-id my-group

skillclaw-agent-evolve-server is installed by scripts/install_skillclaw_server.sh, but still requires openclaw in PATH. See agent_evolve_server/README.md for model and runtime details.

4. Skill Management

skillclaw skills pull          # download shared skills
skillclaw skills push          # upload local skills
skillclaw skills sync          # bidirectional
skillclaw skills list-remote   # browse shared skills

---

WildClawBench Experiments

SkillClaw includes built-in experiment workflows for evaluating skill evolution on the WildClawBench benchmark. The main public experiment entrypoint is scripts/run_wildclawbench_iterative_evolve_agent.py, and the rest are available via skillclaw benchmark CLI commands. See skillclaw benchmark --help for all available subcommands.

---

Project Structure

SkillClaw/
├── skillclaw/                  # Client proxy, CLI, config, skill sync, experiments
│   ├── cli.py
│   ├── api_server.py
│   ├── launcher.py
│   ├── skill_manager.py / skill_hub.py
│   └── experiments/
├── evolve_server/              # Workflow evolve server
│   ├── __main__.py
│   ├── server.py
│   ├── summarizer.py / aggregation.py / execution.py
│   └── config.py / llm_client.py / skill_registry.py
├── agent_evolve_server/        # OpenClaw-based evolve server
│   ├── __main__.py
│   ├── server.py
│   ├── workspace.py / openclaw_runner.py
│   └── EVOLVE_AGENTS.md
├── scripts/                    # Installers + main public experiment runner
│   ├── install_skillclaw.sh
│   ├── install_skillclaw_server.sh
│   └── run_wildclawbench_iterative_evolve_agent.py
├── assets/                     # Logo and docs assets
└── pyproject.toml              # Package metadata & extras

Configuration

Configuration lives in ~/.skillclaw/config.yaml plus environment variables.

• Client / shared credentials: example_env.sh
• Evolve server env template: evolve_server/.env.example
• Inspect config: skillclaw config show
• Update config: skillclaw config <key> <value>

Acknowledgement

The repo is built upon these open-source repos.

MetaClaw - Just talk to your agent — it learns and evolves

WildClawBench - Can an AI agent do real work, end-to-end, without hand-holding

OpenClaw-RL - Train a personalized agent simply by talking to it

Contributing

SkillClaw is a community-driven project. We welcome contributions of all kinds — bug reports, feature requests, new skills, documentation improvements, and more. Feel free to open an issue or submit a pull request!

License

See LICENSE for details.