AIMM: OpenClaw Multi-Agent Trading Skill
A hedge-fund-style multi-agent crypto trading system packaged for OpenClaw, with a hard risk-control veto and quant-grade backtesting.
Positives: Flow API enforces x-api-key auth, empty secrets auto-generated, defaults are paper/testnet only, SECURITY.md exists. Deductions: a Nexus demo API key is hardcoded and published in SKILL.md and examples, and claw_runner.py silently writes this shared key into the user's .env when .env.example is missing (shared-secret leakage, poor key hygiene); no explicit user-confirmation gate before any trading, no rollback mechanism documented; external trading effects rest only on an unverified Risk Guard description.
Runner, verify script and manifest entrypoints are broadly self-consistent, failure messages are readable (import errors, TA-Lib guidance), and the repo ships CI plus pytest. Deductions: nothing executed in this static review; version inconsistency (SKILL.md 1.0.0 vs manifest 0.2.0 vs pyproject 0.1.0); the runner auto-starts paper trading when no mode is given, and failure feedback when --verify passes with missing optional-but-needed components is incomplete.
Scenarios are clear (backtest/paper), with English, Korean and partial Chinese docs, and an FAQ covering common install issues. Deductions: non-fit boundaries (live trading, non-crypto assets) not declared; core function depends entirely on overseas services (Binance, Nexus, OpenAI/AtlasCloud) with no mainland-China reachability provision; manifest declares 14 agents while SKILL.md claims 7 desks, weakening trigger semantics.
Well-layered docs (SKILL.md, examples, docs/, README), clear AGPL-3.0 license, structured manifest, contributing guide and SECURITY.md. Deductions: three inconsistent version numbers with no changelog; SKILL.md carries a future date (2026-04-17) which undermines credibility; unverified publisher means maintenance responsibility and update path are unclear.
Provides runnable CLI entrypoints, a backtest pipeline and structured traces, with theoretical marginal value over manual setup. Deductions: headline performance numbers (+14.95%, Sharpe 1.79) cannot be reproduced in a static review and the README itself admits a single backtest is not proof of edge; the agentic path hard-requires an LLM key and outputs need human review, so comparative-benefit evidence is limited.
Committed pytest suite, deterministic fixtures, a CI workflow and data MANIFEST. hashes give decent auditability. Deductions: within static cap only — key performance claims lack third-party execution evidence, test coverage of the skill's key paths is unconfirmed, and factual content is mixed with marketing (community hashtags, institutional-grade claims).
- A Nexus demo API key is hardcoded in SKILL.md/examples and can be silently written into your .env by the runner; treat it as public and set your own keys for any real use.
- This is a static review; no code was executed. All performance figures (returns, Sharpe) are unverified and not investment advice.
- Three inconsistent version numbers (1.0.0/0.2.0/0.1.0) plus a future-dated stamp indicate weak version governance.
- Core functionality depends on Binance, Nexus and OpenAI — reachability from mainland-China networks is unverified.
- The agentic path hard-requires an LLM API key; assess cost and model dependence before use.
What does this skill do, and when should you use it?
AI Market Maker (AIMM) is an open-source crypto trading skill that orchestrates seven 'trading desk' agents (Market Scan, Technical TA, Statistical Alpha, Sentiment, Risk Management, Portfolio Management, Risk Guard) via LangGraph. Its defining feature is a hard Risk Guard veto — no trade executes without risk approval. The skill is pre-configured for OpenClaw environments with a dedicated runner, an installation verifier, and paper-trading/backtest commands. A sample backtest (BTC/ETH/SOL, 100 days of data) shows 14.95% total return, 1.79 Sharpe ratio, and 62.5% win rate — but the authors themselves warn that one profitable backtest is not proof of edge.
Runs the openclaw/scripts/claw_runner.py entry point with --backtest, --paper (e.g. --ticker BTC/USDT), and --verify modes; drives a multi-agent LangGraph workflow that generates signals and manages portfolio allocation; forces every execution through the Risk Guard veto layer; paper trades on Binance Testnet; produces trade ledgers and risk-event logs with Sharpe ratio, excess return versus buy-and-hold, and max drawdown; exposes a mostly read-only Flow HTTP API (/runs/latest, /runs/{id}/payload, /pm/portfolio-health, etc.) secured by an x-api-key header.
- A quant trader who wants to backtest multi-asset crypto strategies without building an agent framework from scratch
- A developer validating an agentic trading workflow on Binance Testnet before considering live execution
- An OpenClaw user deploying a risk-governed trading skill with a single claw install command
- A researcher who needs auditable trading runs with full traces, reasoning logs, and an event ledger
- A team piping run data into their own dashboard via the Flow API
What are this skill's strengths and limitations?
- Risk Guard is a real governance layer with veto power, not just logging
- Every backtest automatically benchmarks against buy-and-hold with excess-return math
- Well-tailored OpenClaw packaging: dedicated runner, install verifier, multilingual docs (English, Korean)
- Full traces, reasoning logs, and an event ledger make runs auditable
- Paper-trading mode lets you test strategies without real funds
- The sample backtest (14.95% return) is a single data point; the authors themselves state it proves no edge
- The default Nexus API key is rate-limited; production requires your own key
- TA-Lib is a C-library dependency with a nontrivial install (conda or source compile without sudo)
- Execution is limited to Binance Testnet paper trading; the Hyperliquid adapter is dry-run only, with no live trading evidence
- The full stack (database, API, web dashboard) requires Docker, and that setup sits outside SKILL.md's coverage
How do you install this skill?
Via OpenClaw: claw install https://github.com/olaxbt/ai-market-maker. Locally: git clone https://github.com/olaxbt/ai-market-maker.git && cd ai-market-maker && claw skill install ./openclaw. Then verify with ./openclaw/scripts/verify_installation.sh. Note: full monorepo setup also needs Python 3.11+, uv, and TA-Lib (use conda install -y ta-lib -c conda-forge on sudo-less machines); a production Nexus API key goes in .env (a rate-limited demo key ships by default).
How do you use this skill?
Backtest: python3 openclaw/scripts/claw_runner.py --backtest (customize with --symbols "BTC/USDT,ETH/USDT" --steps 150). Paper trade: python3 openclaw/scripts/claw_runner.py --paper --ticker BTC/USDT. Verify: python3 openclaw/scripts/claw_runner.py --verify. Equivalent claw run ai-market-maker ... commands are pre-configured. Platform-level setup (.env, strategy presets, Docker stack, dashboard) is documented in the root README and docs/ directory, not fully in SKILL.md.