The Agent Skills Directory

[DATA_EXPOSURE]: The skill resolves a CertainData API key from environment variables or a local .env file (defaulting to ~/.env). This is a standard practice for secret management in development tools. The instructions include strict safeguards: the agent is forbidden from logging or outputting the key, and it is transmitted only to the issuer's official endpoint (api.certaindata.ai) via an Authorization header. On supported platforms, it even recommends passing the key by reference to ensure the raw secret never enters the model's context.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources, including third-party seller responses and the public x402 Bazaar. To mitigate risk, it defines a 'Trust Boundary' that treats all external content as non-executable data. It uses a strict allowlist for parsing payment fields and requires the agent to display external text (like error messages) in clearly delimited blocks, preventing the agent from following any embedded natural-language commands.
[EXTERNAL_DOWNLOADS]: The skill fetches its service catalog and transaction signatures from api.certaindata.ai (the vendor's domain) and performs discovery searches via api.cdp.coinbase.com (a well-known service). These network operations are necessary for the skill's primary function and target trusted infrastructure.
[REMOTE_CODE_EXECUTION]: The skill is composed entirely of instructions and does not include any bundled scripts, shell commands, or dynamic code execution capabilities, significantly reducing the attack surface.

certaindata