The Agent Skills Directory

[COMMAND_EXECUTION]: The SKILL.md documentation provides copy-pasteable bash commands for users to execute, including fetching a script via curl and running it with node. While these point to the vendor's own infrastructure (fittings.sh), they represent a manual remote code execution pattern common in developer tools.
[CREDENTIALS_UNSAFE]: The skill instructions and code (call.mjs) explicitly require and handle a plaintext blockchain private key (FITTINGS_PRIVATE_KEY) from environment variables. While the documentation correctly advises using a throwaway wallet and notes the risks of plaintext storage, this pattern involves handling highly sensitive credentials.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and validate external product feeds (GTIN barcodes). This creates an ingestion surface where malicious instructions could theoretically be embedded in data processed by an agent, though the specific data format (numeric barcodes) significantly limits this risk. The analysis notes the presence of this surface but identifies no specific malicious payload.

valid-gtin