The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill allows the agent to process untrusted code from a project path using powerful system tools, creating a vulnerability surface.
Ingestion points: The agent is instructed to use Read, Glob, and Grep tools to ingest content from arbitrary files within a user-provided project directory as seen in the tools definition of SKILL.md.
Boundary markers: The skill metadata and instructions do not define delimiters or specific safety guidelines to help the agent distinguish between its scanning logic and instructions found inside the scanned data.
Capability inventory: The Bash tool is enabled in the YAML frontmatter, and the skill references an external Python script (scripts/security_scan.py), providing a high-privilege execution environment for potentially malicious instructions.
Sanitization: There is no evidence of sanitization or validation of the ingested file content before it is processed by the agent or its tools.

vulnerability-scanner