design
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the main agent to include user-provided visual preferences verbatim in the creative brief for the design subagent. Ingestion points: The 'task' field in the startAsyncSubagent call as defined in SKILL.md. Boundary markers: The suggested 'The user wants' prefix does not provide robust isolation or instructions to ignore embedded commands within the interpolated string. Capability inventory: The design subagent has documented access to file operations, package management, and web searching tools. Sanitization: The skill does not recommend any sanitization or validation of user input before its inclusion. Mitigation: Wrap user-provided text in strong delimiters (e.g., XML tags) and explicitly instruct the subagent to ignore any instructions contained within those sections.
Audit Metadata