The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill requires a FITTINGS_PRIVATE_KEY environment variable to process payments. The documentation correctly warns users to use throwaway wallets and notes that keys are stored in plaintext in the execution environment. This is a risk of local credential exposure rather than exfiltration to an attacker, as the key is used for signed authorizations to the vendor's own gateway.- [COMMAND_EXECUTION]: The documentation provides curl and node command lines for users to execute. These target the vendor's own infrastructure (fittings.sh) for fetching configuration and running the payment client. While it instructs users to download and run call.mjs, the script is provided as part of the skill and its behavior is transparent.- [EXTERNAL_DOWNLOADS]: The skill instructions suggest downloading call.mjs via curl if it is missing. It also fetches a SHA256 hash from https://fittings.sh/skill/call.mjs.sha256 at runtime to verify the integrity of the local script before signing payments. These downloads target the vendor's official domain.- [PROMPT_INJECTION]: The skill processes external earthquake data from the USGS (via fittings.sh). This presents a standard indirect prompt injection surface where malicious content could theoretically be embedded in the earthquake 'notes' or metadata fields, though the risk is mitigated by the structured nature of the data (magnitude, coordinates). No explicit boundary markers were detected in the instructions to protect the agent from data-driven instructions.

quake-recent