architect
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill operates by 'grounding' on existing subsystems (Phase A) and synthesizing designs based on these artifacts (Phase B). This process involves ingesting content from the user's codebase, which serves as a vector for indirect prompt injection if that code contains adversarial instructions designed to influence the AI's design decisions.
- Ingestion points: Grounding artifacts produced from existing subsystems in Phase A, and task descriptions provided by the user in Phase B.
- Boundary markers: The workflow uses distinct phases (Ground, Sketch, Agree, Implement) to separate concerns, though it lacks explicit instructions to ignore embedded instructions in the source files being analyzed.
- Capability inventory: The skill culminates in Phase D ('Implement against the sketch'), where the agent writes implementation code into the project files.
- Sanitization: There is no evidence of input sanitization or filtering of external content before it is processed by the architectural runners.
Audit Metadata