The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill requires the user to provide a raw EVM private key via the POLYCLAW_PRIVATE_KEY environment variable. This key is used unencrypted to sign and broadcast on-chain transactions on the Polygon network.
[COMMAND_EXECUTION]: The main entry point scripts/polyclaw.py uses subprocess.run to execute other scripts within the skill directory based on user-provided arguments.
[PROMPT_INJECTION]: The hedge discovery mechanism in scripts/hedge.py is vulnerable to indirect prompt injection.
Ingestion points: Market questions and descriptions are fetched from the Polymarket Gamma API and directly interpolated into LLM prompts.
Boundary markers: The prompt uses standard double quotes for market questions but lacks explicit instructions to ignore embedded commands or structural validation of the untrusted content.
Capability inventory: The skill possesses the capability to execute financial transactions (splitting positions and selling on the CLOB) which could be targeted by an injection.
Sanitization: There is no visible sanitization or filtering of the market questions before they are processed by the LLM.
[EXTERNAL_DOWNLOADS]: The skill interacts with the Polymarket Gamma API and CLOB API for market data and order management, and uses Chainstack's Polygon RPC for blockchain operations. It also makes requests to OpenRouter for LLM-based analysis.

polyclaw