The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill downloads a JavaScript file from 'https://fittings.sh/skill/call.mjs' and immediately executes it using 'node'. This 'download and execute' pattern is a high-risk vector as the remote script's content is not part of the audited skill and can be modified by the server at any time without user oversight.
[CREDENTIALS_UNSAFE]: The instructions direct the user to export a blockchain private key to the 'FITTINGS_PRIVATE_KEY' environment variable. While the skill advises using a 'throwaway wallet,' storing sensitive private keys in plaintext environment variables exposes them to the runtime environment and any code executed within it, including the remotely downloaded script.
[EXTERNAL_DOWNLOADS]: The skill fetches the execution script ('call.mjs') and configuration information from 'fittings.sh' via 'curl'. While these are vendor-owned resources, the reliance on fetching and running remote files during skill operation introduces significant supply chain and runtime risks.
[COMMAND_EXECUTION]: The skill utilizes shell commands to download files ('curl -sO') and execute code ('node call.mjs'), integrating external execution into its primary workflow.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by taking user-supplied 'gtin' data and passing it directly into a shell command for script execution. 1. Ingestion points: The 'gtin' parameter in the 'node' execution command. 2. Boundary markers: No delimiters or warnings are provided to prevent the agent from obeying instructions embedded in the 'gtin' input. 3. Capability inventory: The agent can perform network requests, write files, and execute JavaScript. 4. Sanitization: There is no evidence of validation or escaping for the data passed to the execution script.

fittings-valid-gtin