// the find
tradecatlabs/tradecat-public
交易猫数据系统
A paper-trading runtime for AI agents, specifically built around reading signal feeds from online spreadsheets and executing simulated trades on Binance futures without touching any real keys or orders. Primarily Chinese-language docs and tooling. Aimed at developers building autonomous trading agents who want a structured simulation harness with audit trails.
The security boundary is unusually well-enforced — the code explicitly refuses to read API keys, sign requests, or call private endpoints, and fails closed when sizing/stop-loss parameters are missing rather than guessing. The JSON Schema contracts in `contracts/` are thorough; there are 40+ schemas covering every I/O boundary, which makes the agent protocol machine-verifiable. CI pipeline is genuinely solid: secret scanning via gitleaks, supply chain audit, dependency policy validation, and skill manifest validation all run on every push. The fail-closed default on `agent_sizing_required` is the right call — it prevents autonomous paper trades from silently using bad defaults.
The signal source is an online spreadsheet that you don't control, which means the entire system's usefulness collapses if that sheet goes away or changes format — there's no fallback and no versioned snapshot of the signal schema. The Wyckoff/TA strategy logic lives in a private repo; this public repo is essentially just the plumbing with no actual trading logic you can inspect or backtest against. Documentation is almost entirely Chinese, which is a real adoption barrier for anyone outside that community. The paper cost model uses a hardcoded 4bps taker fee as a 'public fallback' — anyone doing serious backtesting on this will get skewed results since actual fees vary significantly by VIP tier and symbol.