ai-skills
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of instructions and metadata from external SKILL.md files through the load_skill tool.
- Ingestion points: SKILL.md files processed by load_skill and read_skill_resource.
- Boundary markers: The library generates <available_skills> XML blocks for Anthropic models and markdown formatting for others to separate skill content from the system prompt.
- Capability inventory: The skill provides discovery (load_skill) and file reading (read_skill_resource). It is designed to work alongside user-provided execution tools.
- Sanitization: Implements path validation in read_skill_resource to prevent path traversal attacks escaping the skill root.
Audit Metadata