Metaclaw Evolving Agent
Deploy MetaClaw as an OpenAI-compatible proxy that injects learned skills, optional RL training, and madmax idle scheduling for a self-improving agent stack.
Overview
metaclaw-evolving-agent is an agent skill most often used in Build (also Operate infra) that deploys MetaClaw—a proxy agent with skills injection, RL training, and madmax scheduling.
Install
npx skills add https://github.com/aradotso/trending-skills --skill metaclaw-evolving-agentWhat is this skill?
- Three runtime modes: skills_only, RL training, and madmax scheduler combining skills + RL + deferred updates
- Optional pip extras: [rl], [evolve], [scheduler] with recommended [rl,evolve,scheduler] install
- Interactive `metaclaw setup` then `metaclaw start` with flags such as `--mode skills_only`
- OpenAI-compatible proxy intercepts conversations for skills injection and meta-learning from live traffic
- 3 documented runtime modes including skills_only and madmax
- 4 optional pip extra groups: rl, evolve, scheduler, plus combined [rl,evolve,scheduler]
Adoption & trust: 1.3k installs on skills.sh; 31 GitHub stars; 2/3 security scanners passed (skills.sh audits).
What problem does it solve?
You want an agent endpoint that improves from live chats but lack a documented path for proxy deploy, skills mode, RL backends, and safe update scheduling.
Who is it for?
Indie builders experimenting with self-evolving agents who can manage Python installs, optional GPU RL, and proxy routing.
Skip if: Teams needing a managed hosted agent with no self-hosted proxy, GPU, or RL operations overhead.
When should I use this skill?
set up metaclaw agent; configure evolving agent; metaclaw skills mode; metaclaw rl training; metaclaw madmax scheduler; agent meta-learning setup; tinker rl backend configuration; metaclaw proxy deployment
What do I get? / Deliverables
You run MetaClaw with a chosen mode—from lightweight skills_only to full madmax RL—and a configured Tinker or proxy stack ready for OpenAI-compatible clients.
- Running MetaClaw proxy with selected mode and configuration from setup wizard
- Optional RL and scheduler backends wired per pip extras
Recommended Skills
Journey fit
Spans multiple journey phases - primary shelf plus alternate fits below.
Initial setup wizard, pip extras, and mode selection are primary Build agent-tooling work. MetaClaw configures the runtime agent layer—proxy, skills injection, and training backends—not a single app feature.
Where it fits
Run `metaclaw setup` and start skills_only before pointing your IDE agent at the local OpenAI-compatible endpoint.
Enable [scheduler] and Google Calendar hooks so madmax defers RL updates outside your working hours.
Operate the proxy under load while monitoring when madmax applies weight updates during idle windows.
Turn on [evolve] summarization so new skills distilled from conversations merge into injection without manual prompt edits.
How it compares
Skill package for deploying a learning proxy—not a static prompt library or a standalone MCP tool server.
Common Questions / FAQ
Who is metaclaw-evolving-agent for?
Solo builders and small teams setting up MetaClaw for skills injection, optional RL on Tinker, and scheduled weight updates.
When should I use metaclaw-evolving-agent?
During Build when wiring the agent proxy and modes; during Operate when tuning madmax scheduler, evolve extras, or RL backends for production traffic.
Is metaclaw-evolving-agent safe to install?
It installs Python packages and may enable network proxies and training stacks—review Security Audits on this page and lock down secrets and egress in your environment.
SKILL.md
READMESKILL.md - Metaclaw Evolving Agent
# MetaClaw Evolving Agent > Skill by [ara.so](https://ara.so) — Daily 2026 Skills collection MetaClaw is an OpenAI-compatible proxy agent that intercepts conversations, injects learned skills, and continuously improves itself through real-world interactions. It supports three modes: lightweight skills injection, immediate RL training, and a smart "madmax" scheduler that defers weight updates to idle/sleep windows. --- ## Installation ```bash # Minimal — skills injection only, no GPU required pip install -e . # Full RL training support (torch, transformers, tinker) pip install -e ".[rl]" # Skill evolution via LLM summarization pip install -e ".[evolve]" # Google Calendar scheduler for madmax mode pip install -e ".[scheduler]" # Recommended: everything pip install -e ".[rl,evolve,scheduler]" ``` --- ## Quick Start ```bash # One-time interactive config wizard metaclaw setup # Start in default madmax mode (skills + RL + smart scheduler) metaclaw start # Skills only — no GPU, no Tinker needed metaclaw start --mode skills_only # RL mode — trains immediately when batch is full metaclaw start --mode rl # RL without scheduler (same as above, explicit) metaclaw start --mode rl ``` After `metaclaw start`, a local OpenAI-compatible proxy is running. Point your client (OpenClaw or any OpenAI SDK consumer) at `http://localhost:<port>` instead of the upstream LLM endpoint. --- ## Configuration `metaclaw setup` writes a config file (default: `~/.metaclaw/config.yaml`). You can also edit it directly: ```yaml # ~/.metaclaw/config.yaml proxy: host: 0.0.0.0 port: 8080 llm: provider: kimi # kimi | qwen | claude | minimax | openai | gemini base_url: https://api.moonshot.cn/v1 model: moonshot-v1-8k # api_key loaded from env: METACLAW_LLM_API_KEY skills: enabled: true max_injected: 5 # max skills injected per turn summarize_after_session: true rl: enabled: true backend: auto # auto | tinker | mint batch_size: 32 algorithm: grpo opd_teacher: false # optional teacher distillation scheduler: # madmax mode only enabled: true sleep_hours: [22, 7] # local 22:00–07:00 idle_timeout_minutes: 15 google_calendar: false # set true + configure OAuth for meeting detection logging: level: info log_dir: ~/.metaclaw/logs ``` ### Environment Variables ```bash export METACLAW_LLM_API_KEY="your-llm-api-key" export METACLAW_TINKER_API_KEY="your-tinker-api-key" # rl mode export METACLAW_MINT_API_KEY="your-mint-api-key" # if backend=mint export GOOGLE_CALENDAR_CREDENTIALS_PATH="path/to/creds.json" # scheduler ``` --- ## Operating Modes | Mode | Command | GPU Required | Description | |------|---------|--------------|-------------| | `skills_only` | `metaclaw start --mode skills_only` | No | Proxy + skills injection + auto-summarization | | `rl` | `metaclaw start --mode rl` | Via API | Skills + GRPO training when batch fills | | `madmax` | `metaclaw start` | Via API | Skills + RL + scheduler (trains only during idle/sleep/meetings) | --- ## Python API ### Programmatic startup ```python import asyncio from metaclaw import MetaClawAgent, AgentConfig, Mode async def main(): config = AgentConfig.from_yaml("~/.metaclaw/config.yaml") agent = MetaClawAgent(config, mode=Mode.MADMAX) await agent.start() asyncio.run(main()) ``` ### Manual skill injection ```python from metaclaw.skills import SkillStore, SkillInjector store = SkillStore(path="~/.metaclaw/skills") # Add a skill manually store.add(