The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill communicates with an external service to perform vulnerability lookups.
Evidence: src/lib.ts uses secureFetch to send package information to https://api.osv.dev/v1/query for analysis.
Context: This interaction is with a well-known, trusted security database and is core to the skill's auditing functionality.
[PROMPT_INJECTION]: The skill processes untrusted local files which could contain malicious instructions designed to influence downstream AI tasks.
Ingestion points: src/index.ts reads package.json and various script files (.js, .cjs, .sh) from the user-provided directory at lines 48 and 81.
Boundary markers: There are no delimiters or explicit instructions to ignore embedded commands when processing or outputting the file content.
Capability inventory: The skill has permissions to write to the file system (safeWriteFile) and make network requests (secureFetch).
Sanitization: The skill performs simple regex scanning for suspicious patterns but does not sanitize or escape the content for safe consumption by an LLM.

supply-chain-sentinel