skill-inventory-auditor
Warn
Audited by Gen Agent Trust Hub on Mar 13, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill implements a workflow to delete directories and files using
rm -rfandrm. These are high-impact operations that can lead to permanent data loss if the target paths are incorrect or if the agent is manipulated into selecting wrong targets. - [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface.
- Ingestion points: Metadata (name and description) is parsed from
SKILL.mdfiles across multiple directories inscripts/scan_inventory.py. - Boundary markers: The skill lacks boundary markers to isolate untrusted skill descriptions when they are presented to the agent in Phase 2.
- Capability inventory: The skill can execute shell commands (
rm,rm -rf) as defined in theSKILL.mdworkflow. - Sanitization: No sanitization or validation of the ingested metadata is performed before it is presented to the agent.
Audit Metadata