architecture-reviewer
Pass
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: LOWPROMPT_INJECTION
Full Analysis
- Indirect Prompt Injection (LOW): The skill is designed to process untrusted user input (architecture diagrams and code) which constitutes an indirect prompt injection surface. Because the skill lacks external capabilities, the impact is limited to influencing agent reasoning. Evidence: 1. Ingestion points: 'architecture-reviewer' processes user-provided diagrams or code in SKILL.md. 2. Boundary markers: Absent; user input is not isolated with delimiters. 3. Capability inventory: No subprocess, file-write, or network capabilities identified in the skill. 4. Sanitization: None identified.
Audit Metadata