design-review
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted design context which could potentially contain instructions aimed at biasing the review.
- Ingestion points: Step 2 in SKILL.md ('Gather context: the screen(s)/flow').
- Boundary markers: No delimiters or warnings are used to isolate user-provided design data from the agent's instructions.
- Capability inventory: The skill calls a local Python script (scripts/contrast.py).
- Sanitization: No input validation or filtering of the ingested context is described.
- [COMMAND_EXECUTION]: The skill instructs the agent to run a specific script for technical design checks.
- Evidence: Use of 'scripts/contrast.py' in Step 4 to resolve color-pair doubts.
Audit Metadata