architect
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted project code and user requirements to generate architectural designs and source code. 1. Ingestion points: Codebase subsystems and user-provided design tasks (SKILL.md Phases A and B). 2. Boundary markers: No specific delimiters are defined to isolate grounding data from instructions. 3. Capability inventory: The skill is designed to write new source code and module maps to the filesystem (SKILL.md Phase D). 4. Sanitization: Implementation against generated sketches is automated by default; human review is an optional opt-in (Phase C), potentially allowing malicious design patterns to be implemented without prior inspection.
Audit Metadata