brainstorming
Pass
Audited by Gen Agent Trust Hub on Sep 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill follows established patterns for creative assistance and research-based brainstorming. The use of subagents and local temporary file storage is within the expected functional scope and appropriate for the tool's purpose.
- [INDIRECT_PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection through its subagent dispatch mechanism. 1. Ingestion points: User-provided brainstorming anchors and ideas enter the agent context and are stored in the case-file ($TMPDIR/brainstorming-.md). 2. Boundary markers: The instructions do not specify delimiters or safety warnings to wrap user-provided data within the subagent prompts. 3. Capability inventory: The skill performs local file writes to $TMPDIR and dispatches multiple parallel subagents for lateral thinking and steelmanning. 4. Sanitization: There is no mention of sanitization or filtering of user input before it is used to populate subagent prompts.
Audit Metadata