The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's Python script (scripts/check_plugin_copies.py) executes the git command-line utility using the subprocess module. These calls are used to determine repository status, branch information, and file history to verify if local copies have drifted from the source. The commands are invoked using a safe list-based argument format where parameters are passed as individual elements rather than a single shell string.\n- [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests and processes data from external files (SKILL.md) located on the filesystem.\n
Ingestion points: The scan_for_skill_dirs function in scripts/check_plugin_copies.py performs a recursive walk of the filesystem to locate and read SKILL.md files from both known configuration paths and sibling directories.\n
Boundary markers: The script does not utilize specific boundary markers in its console output for the content of the files, although the tool's logic focuses on reporting status and paths rather than echoing the raw content of the audited instructions.\n
Capability inventory: The skill enables the agent to execute a local Python script which has read access to sensitive agent configuration directories (~/.claude and ~/.agents). The tool instructs the agent on how to interpret and act on the audit report findings.\n
Sanitization: The script uses cryptographic hashing (SHA-256) and line-overlap calculations to classify file status. This algorithmic approach effectively ignores the semantic meaning of instructions within the files, reducing the likelihood of the script itself being subverted by injected content.

copy-audit