Launch surface: clawpowers-agent is the thin wrapper runtime around a stock, updatable OpenClaw instance. The shared capability implementation lives in the clawpowers library.

• Product site: https://clawpowers.ai
• Docs: https://clawpowers.ai/docs
• Agent runtime: https://clawpowers.ai/agent

Supported matrix: clawpowers-agent 1.1.x + clawpowers 2.2.x + openclaw >=2026.5.7 <2026.6.0.

More docs: SECURITY · Compatibility · Known Limitations · Licensing · Releasing · Demo · Roadmap

The autonomous AI coding agent that plans, executes, reviews, remembers, and self-improves.

Patent Pending — Non-Custodial Multi-Chain Financial Infrastructure System for Autonomous AI Agents

License note: This package is MIT-licensed. It depends on clawpowers, which is licensed under BSL 1.1. Non-production use is free; production use requires a commercial license until April 3, 2030, after which it converts to Apache 2.0. Review LICENSING.md before commercial use.

ClawPowers Agent is a TypeScript runtime wrapper for running ClawPowers capabilities through OpenClaw. It provides the public clawpowers CLI, initializes local config, syncs skills, and routes tasks through the stock OpenClaw runtime.

ClawPowers Agent runs on top of OpenClaw. You need a working OpenClaw installation before running clawpowers.

npm install -g openclaw@latest
openclaw status

OpenClaw needs at least one LLM provider configured. Add your API key to ~/.openclaw/config.json or set the appropriate environment variable, such as ANTHROPIC_API_KEY or OPENAI_API_KEY.

Node version: current OpenClaw releases require Node.js 22.12+. If openclaw exits immediately during clawpowers init or clawpowers run, check node --version first.

# Install the supported OpenClaw runtime and ClawPowers CLI wrapper.
npm install -g openclaw@latest clawpowers-agent

# Initialize config at ~/.clawpowers/.
clawpowers init

# Check runtime status.
clawpowers status

# Run a task.
clawpowers run "Build a REST API with Express and Zod validation. Tests pass."

ClawPowers runs coding tasks through a five-phase control loop:

1. Intake: parse the task into a goal and acceptance criteria.
2. Planning: decompose the goal into ordered steps and skill matches.
3. Execution: run steps with retries and bounded parallelism.
4. Review: validate output against the success criteria.
5. Completion: report status and persist useful lessons.

Core capabilities come from the underlying clawpowers package:

• Memory — Three-tier memory for working, episodic, and procedural context.
• RSI — Recursive Self-Improvement with tiered autonomy and A/B testing.
• Payments — x402 payment workflow handling with policy checks and human-first defaults.
• Parallel Swarm — Fan-out concurrent task execution with model routing and token budgeting.
• ITP — Experimental context compression for multi-agent communication.

Rust, WASM, and TypeScript fallbacks live in the clawpowers npm package. Installing clawpowers-agent pulls in clawpowers automatically; you do not build native code in this repo.

import { isNativeAvailable } from 'clawpowers';
console.log('Native acceleration:', isNativeAvailable());

• clawpowers is the capability library: config, payments, memory, RSI, wallet helpers, swarm primitives, ITP client, native/WASM acceleration, and skill assets.
• clawpowers-agent is the OpenClaw wrapper: CLI, plugin packaging, skill-sync layer, runtime glue, state machine, and control loop.
• OpenClaw remains the underlying runtime. The goal is to stay aligned with stock OpenClaw instead of forking it.

Safety invariants such as spending limits, identity, sandbox boundaries, credentials, and tier definitions are not modifiable by RSI.

Config lives at ~/.clawpowers/config.json with Zod validation.

{
  "version": "1.1.13",
  "profile": "dev",
  "rsi": {
    "enabled": true,
    "tiers": { "t1": "auto", "t2": "auto", "t3": "ask", "t4": "ask" }
  },
  "payments": {
    "mode": "human-first",
    "dailyLimitUsd": 25,
    "weeklyLimitUsd": 100,
    "allowedDomains": []
  },
  "logging": { "level": "info", "retentionDays": 30 },
  "skillsDir": "~/.clawpowers/skills",
  "dataDir": "~/.clawpowers/data"
}

Profiles: dev, lead, secure, growth, full.

Payment modes: human-first, auto, disabled.

clawpowers config get payments.dailyLimitUsd
clawpowers config set payments.dailyLimitUsd 50
clawpowers config set payments.mode auto

clawpowers init
clawpowers run <task>
clawpowers status
clawpowers config get <key>
clawpowers config set <key> <value>
clawpowers skills list
clawpowers skills add <name>
clawpowers skills remove <name>

git clone https://github.com/up2itnow0822/ClawPowers-Agent.git clawpowers-agent
cd clawpowers-agent
npm install
npm run build
npm run typecheck
npm test
npm run lint

Memory, payments, and RSI behavior are covered primarily in the ClawPowers-Skills (clawpowers) test suite. This repo keeps tests focused on the agent runtime and re-exports.

npm run demo:task
npm run demo:memory
npm run demo:rsi

MIT — see LICENSE.

For commercial use, review both this repo and the underlying clawpowers package licensing. See LICENSING.md.