architect
Pass
Audited by Gen Agent Trust Hub on Oct 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the existing codebase during its grounding phase. This information is passed to sub-agents during the sketch phase to inform design decisions.
- Ingestion points: Source code, module structures, and architectural rationales from existing project files are ingested in Phase A (Grounding), as described in SKILL.md.
- Boundary markers: The skill utilizes a structured prompt in references/runner-prompt.md to guide sub-agents, but it lacks specific sanitization or isolation instructions for the content read from the project.
- Capability inventory: The skill facilitates code implementation and rationale generation by orchestrating sub-agents through the arena and pstack-harness skills.
- Sanitization: No explicit logic for sanitizing or escaping ingested project data is documented before it is provided to the agent runners.
Audit Metadata