brainstorm-coach
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input and writes it verbatim to a session document, creating a potential surface for indirect prompt injection if the resulting logs are subsequently processed by an AI agent.\n
- Ingestion points: User ideas provided during divergent thinking rounds and setup phase (SKILL.md).\n
- Boundary markers: The skill uses labeling (e.g., '(user)' and '(AI)') to distinguish content origins, but these markers are not robust security boundaries.\n
- Capability inventory: The agent is instructed to create and write to a local markdown file (e.g., 'brainstorm-*.md') to maintain a session record.\n
- Sanitization: The instructions mandate verbatim capture of user responses, with no provision for escaping markdown control characters or filtering embedded instructions.
Audit Metadata