ads-router
Pass
Audited by Gen Agent Trust Hub on Aug 22, 2026
Risk Level: SAFE
Full Analysis
- [DYNAMIC_EXECUTION]: The skill identifies user intent and loads corresponding instruction files from the
skills/directory. This is a safe implementation as the loaded filenames are restricted to a predefined list, preventing directory traversal or the execution of unintended files.\n- [INDIRECT_PROMPT_INJECTION]: The skill uses natural language input to select specialized sub-skills, which presents an attack surface for indirect prompt injection. However, the use of a hardcoded matching table limits the impact of potentially malicious user queries.\n - Ingestion points: User requests are parsed for keywords to match routing rules in SKILL.md.\n
- Boundary markers: The skill does not employ specific boundary delimiters for user input.\n
- Capability inventory: Reading and following instructions from the local
skills/folder.\n - Sanitization: No sanitization is applied beyond intent matching against a static list.
Audit Metadata