The Agent Skills Directory

[CREDENTIALS_UNSAFE] (HIGH): The skill requires a PRIVATE_KEY for a blockchain wallet to be exported as an environment variable. This is a highly sensitive credential that, if exposed via logs or environment dumping, leads to a total loss of assets.
[COMMAND_EXECUTION] (HIGH): The skill uses npx ts-node to execute local scripts (e.g., scripts/trade.ts). These scripts have full access to the user's environment, including the provided PRIVATE_KEY, and perform financial transactions.
[EXTERNAL_DOWNLOADS] (MEDIUM): The programmatic usage relies on the pnp-evm package. This package is not from a trusted source (as defined in the security framework) and its integrity cannot be verified in this context.
[BEST_PRACTICE_VIOLATION] (MEDIUM): The documentation explicitly encourages the use of 'Infinite Approvals' (type(uint256).max). This is a dangerous DeFi pattern that allows the smart contract to move an unlimited amount of the user's tokens, creating a massive risk if the protocol's contracts are ever compromised.
[INDIRECT_PROMPT_INJECTION] (LOW):
Ingestion points: Market questions and metadata are ingested via the --question flag or the createMarket function.
Boundary markers: None present; the question is a raw string.
Capability inventory: Executes blockchain transactions via wallet signature.
Sanitization: None mentioned. While the impact is largely limited to the transaction payload, an agent automatically creating markets based on untrusted summaries could be manipulated into creating fraudulent markets.

pnp-markets