The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a bundled Python script (scripts/scan_skill.py) and perform directory listings (ls) and searches (grep) to audit other skills.
[PROMPT_INJECTION]: The skill functions by ingesting and analyzing untrusted content from third-party agent skills, which presents an indirect prompt injection surface. 1. Ingestion points: Path to the target skill directory provided by the user. 2. Boundary markers: Not explicitly defined in the instructions for the agent. 3. Capability inventory: Includes file system read operations (ls, cat, grep) and execution of bundled scripts via uv run. 4. Sanitization: The process relies on the agent's internal evaluation and a manual human-in-the-loop review workflow to identify malicious intent.

skill-scanner