se
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied free-text questions to select and read content from external knowledge packs located in sibling directories or specific application paths.
- Ingestion points: Untrusted data enters the agent context through the free-text question argument in the
/secommand located inSKILL.md. - Boundary markers: The skill includes explicit instructions for the agent to provide chapter-specific citations, limit the number of claims per pack, and refrain from using model memory when sources are silent.
- Capability inventory: The skill directs the agent to perform file system read operations on relative paths (
../<slug>/) and specific application-specific directories (e.g.,~/.claude/skills/,~/.codex/prompts/) to retrieve its knowledge base. - Sanitization: The skill utilizes a hardcoded routing map (Topics and Agency contexts tables) which functions as a whitelist for allowed pack identifiers, preventing the use of arbitrary strings in file paths during the routing process.
Audit Metadata