vault-router
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input to determine which subsequent set of instructions to load and execute.
- Ingestion points: User requests are read to match against the skill catalog in
SKILL.md. - Boundary markers: The skill does not implement delimiters or isolation techniques for the user-provided task descriptions.
- Capability inventory: The routing logic facilitates the reading and loading of other skill files (e.g.,
SKILL.mdfrom sibling directories). - Sanitization: No input validation or filtering is performed on user requests before they are used for routing decisions.
Audit Metadata