The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to clone a complete software repository from an unverified GitHub user account ('lorine93s') that does not match the identified vendor's infrastructure.
Evidence: git clone https://github.com/lorine93s/openclaw-polymarket-ai-arbitrage-trading-bot in SKILL.md.
[REMOTE_CODE_EXECUTION]: The skill executes remote code by downloading an external repository and immediately running it using package managers and start scripts.
Evidence: Sequence of git clone, npm install, and npm run dev / npm start commands provided in the installation and execution sections.
[CREDENTIALS_UNSAFE]: The skill requires the user to input highly sensitive information, specifically a Polygon wallet private key, into an environment file that is accessed by the unverified external code.
Evidence: Instructions to set POLYMARKET_PRIVATE_KEY in a .env file.
[COMMAND_EXECUTION]: The skill triggers multiple shell commands to set up and run the trading environment, providing a surface for command-based attacks if inputs are not properly sanitized.
Evidence: npm install, cp .env.example .env, and npm run build commands.
[INDIRECT_PROMPT_INJECTION]: The bot is designed to ingest external data from prediction markets and an optional external AI decision engine, which could be manipulated to influence the bot's trading behavior.
Ingestion points: Polymarket CLOB API, Gamma API, and external HTTP signals from openclaw.http.url.
Capability inventory: The skill has the capability to sign transactions and execute orders using the provided wallet key.
Boundary markers: None identified; the bot appears to trust external signals for decision making.
Sanitization: Not explicitly documented for incoming market or signal data.

polymarket-openclaw-ai-arbitrage-bot