# metaclaw-evolving-agent

This skill walks you through deploying MetaClaw, a proxy agent that injects learned skills into conversations and trains on live sessions.

- skill · `type=skill` · Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode · 1400 installs · ★ 88
- Author: reason-machines
- License: not stated · policy=link-only · description only; the file stays with the author
- Source: https://github.com/reason-machines/trending-skills/tree/HEAD/skills/metaclaw-evolving-agent/SKILL.md
- Updated: 2026-10-09
- HTML: https://affpapa.org/en/nevibecode/skills/metaclaw-evolving-agent

## What it does

MetaClaw sits in front of your LLM as an OpenAI-compatible proxy. Clients point at `http://localhost:<port>`, and the agent intercepts each turn, pulls relevant skills into the system prompt, and stores the exchange in an experience buffer. The skill covers installation, the `metaclaw setup` wizard, and three run modes: skills injection alone, immediate RL training, or madmax, where a scheduler pushes weight updates into idle and sleep windows.

You get guidance on `~/.metaclaw/config.yaml`, the environment variables, and the Python API — `MetaClawAgent`, `SkillStore`, `SkillInjector`, `ExperienceBuffer`, `RLTrainer`, `RewardModel`. The `skills_only` mode needs no GPU. RL training wants `METACLAW_TINKER_API_KEY`, or `METACLAW_MINT_API_KEY` when the backend is `mint`. Meeting detection in the scheduler reads OAuth credentials from `GOOGLE_CALENDAR_CREDENTIALS_PATH`.

Set `METACLAW_LLM_API_KEY` before anything else. Without it the proxy has no upstream model to forward to, and `metaclaw start` gets you nowhere. Installation runs from the local repo with `pip install -e .` plus extras: `[rl]`, `[evolve]`, `[scheduler]`.

## Install

### Claude Code (native platform)

1. **Skills CLI (Vercel)**
```bash
npx skills add reason-machines/trending-skills --skill metaclaw-evolving-agent -a claude-code
```

2. **Manually: put the files in place**
- File: `~/.claude/skills/metaclaw-evolving-agent/SKILL.md` — Copy the whole skill folder; for one project use .claude/skills/metaclaw-evolving-agent/

### Codex

1. **Skills CLI (Vercel)**
```bash
npx skills add reason-machines/trending-skills --skill metaclaw-evolving-agent -a codex
```

2. **Manually: put the files in place**
- File: `.agents/skills/metaclaw-evolving-agent/SKILL.md` — For all projects use ~/.agents/skills/

### Cursor

1. **Skills CLI (Vercel)**
```bash
npx skills add reason-machines/trending-skills --skill metaclaw-evolving-agent -a cursor
```

2. **Manually: put the files in place**
- File: `.cursor/skills/metaclaw-evolving-agent/SKILL.md` — Cursor also reads .agents/skills/ and .claude/skills/

### GitHub Copilot

1. **GitHub CLI**
```bash
gh skill install reason-machines/trending-skills metaclaw-evolving-agent
```

2. **Manually: put the files in place**
- File: `.github/skills/metaclaw-evolving-agent/SKILL.md`

### Gemini CLI

1. **Skills CLI (Vercel)**
```bash
npx skills add reason-machines/trending-skills --skill metaclaw-evolving-agent -a gemini-cli
```

2. **Platform CLI**
```bash
gemini skills install https://github.com/reason-machines/trending-skills --path skills/metaclaw-evolving-agent
```

### Windsurf (Devin Desktop)

**Manually: put the files in place**
- File: `.devin/skills/metaclaw-evolving-agent/SKILL.md` — The legacy .windsurf/skills/ path still works

### OpenCode

**Skills CLI (Vercel)**
```bash
npx skills add reason-machines/trending-skills --skill metaclaw-evolving-agent -a opencode
```

### Claude.ai

**Claude.ai**
- Zip the folder and upload it in Settings → Capabilities → Skills

Where a file is given instead of a command, copy the files from the source to that path: https://github.com/reason-machines/trending-skills/tree/HEAD/skills/metaclaw-evolving-agent/SKILL.md

## When to use it
- stand up a proxy agent in front of an existing LLM endpoint
- inject retrieved skills into the system prompt each turn
- record conversation turns into an experience buffer for training
- run batched GRPO training without a local GPU
- defer weight updates to sleep hours and idle time

## Example prompts
- Run `metaclaw setup`, then start MetaClaw in `skills_only` mode
- Start `metaclaw start` in madmax mode so training waits for idle time
- Add a `code-review-checklist` skill to `SkillStore` and inject it into queries

## Similar
- [setup-matt-pocock-skills](https://affpapa.org/en/nevibecode/skills/setup-matt-pocock-skills.md) — skill · 1052851 installs
- [ask-matt](https://affpapa.org/en/nevibecode/skills/ask-matt.md) — skill · 766587 installs
- [to-tickets](https://affpapa.org/en/nevibecode/skills/to-tickets.md) — skill · 711230 installs
- [writing-plans](https://affpapa.org/en/nevibecode/skills/writing-plans.md) — skill · 282859 installs
- [caveman-review](https://affpapa.org/en/nevibecode/skills/caveman-review.md) — skill · 411800 installs
- [caveman-compress](https://affpapa.org/en/nevibecode/skills/caveman-compress.md) — skill · 411500 installs

From the same set: [reason-machines/trending-skills](https://affpapa.org/en/nevibecode/by/reason-machines/trending-skills.md) — 139 cards

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md