The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted blockchain data from the Blockscout PRO API, which could be exploited to deliver malicious instructions via transaction data or logs.\n- Ingestion points: External data enters the agent context through REST and JSON-RPC responses from https://api.blockscout.com.\n- Boundary markers: The skill does not define explicit delimiters or 'ignore' instructions for the API response data placed in the prompt.\n- Capability inventory: The skill uses shell-based tools including curl, oastools, and jq, and instructions include writing API responses to temporary files on disk.\n- Sanitization: The skill effectively mitigates risk by instructing the agent to use jq to project only the necessary fields, which prevents unvalidated raw payloads from occupying the full context window.\n- [EXTERNAL_DOWNLOADS]: The skill recommends the local installation of oastools (from github.com/erraggy/oastools) and jq as utility dependencies. This is documented as a manual setup requirement for the user rather than an automated or runtime remote code download.

web3-dev