The Agent Skills Directory

[COMMAND_EXECUTION]: The instructions direct the agent to run standard ecosystem auditing tools such as npm audit, pip-audit, bundle audit, and govulncheck. These are legitimate tools used for the skill's stated purpose of security auditing.
[DATA_EXPOSURE]: The skill requires access to project manifest files (e.g., package.json, requirements.txt, go.mod) to analyze dependencies. This access is appropriate for a dependency audit skill and does not involve exfiltration of sensitive credentials or private user data.
[PROMPT_INJECTION]: The skill does not contain instructions to override safety filters, bypass constraints, or ignore previous instructions. It maintains a professional and instructional tone throughout.
[INDIRECT_PROMPT_INJECTION]: While the skill processes external data (manifest files and audit tool output), it includes guidelines for verification and reachability analysis, which serve as manual checkpoints.
Ingestion points: Reads package.json, lockfiles, and requirements.txt (SKILL.md, Step 1).
Boundary markers: None explicitly defined for the prompt itself, but it encourages manual verification of findings.
Capability inventory: Uses standard CLI tools like grep and ecosystem-native audit tools.
Sanitization: Relies on the output of established security tools like govulncheck.

Dependency Risk Audit