brainstorm
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest data from local repository files, including Architecture Decision Records (ADRs) and general project documentation, to provide context for the brainstorming session. This behavior creates a surface for indirect prompt injection, where maliciously crafted content in these files could attempt to influence the agent's behavior or session logic.
- Ingestion points: The agent reads context from existing files in
plans/brainstorm/, ADRs, and general repository documentation (repo docs), in addition to processing free-text ideas from the user. - Boundary markers: The skill implements a basic tagging system, labeling ideas as
[user]or[agent]within the generated session file to distinguish between participants. - Capability inventory: The skill possesses the capability to read repository files, write markdown files to the
plans/brainstorm/directory, and interact with the user via theAskUserQuestiontool. - Sanitization: No explicit sanitization, filtering, or validation of the content ingested from repository files is specified in the instructions.
Audit Metadata