The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes various third-party dependency manifest files (e.g., package.json, requirements.txt, go.mod) which are external ingestion points. While these files could theoretically contain malicious instructions targeted at the agent, the skill functions as a passive auditor and does not execute the dependencies it identifies.
Ingestion points: Manifest files parsed by scripts/dep_scanner.py, scripts/license_checker.py, and scripts/upgrade_planner.py across multiple ecosystems.
Boundary markers: The scripts parse structured data (JSON, TOML) or use regular expressions but do not implement specific boundary markers to prevent the agent from following embedded instructions.
Capability inventory: The tools are limited to reading local files, performing semantic versioning analysis, and writing reports to user-defined paths or stdout. They do not possess network capabilities or automated script execution for the audited dependencies.
Sanitization: Standard library parsers (json) and regular expressions are used to extract versioning and package information.
[COMMAND_EXECUTION]: The skill provides command-line utilities and documentation instructions for making scripts executable via chmod +x. These are standard development practices for local toolkits and do not represent a security risk.
[SAFE]: The skill exhibits benign behavior, follows established security best practices for dependency management, and relies exclusively on Python standard library modules without requiring external network connectivity or providing high-privilege operations.

dependency-auditor