brainstorm
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection through its automated discovery mechanism.
- Ingestion points: The
references/discovery.mdfile instructs the agent to "explore the codebase instead of asking" to map current systems and features, which could lead to the ingestion of malicious instructions hidden in code comments or documentation. - Boundary markers: The instructions specify that implementation details should not cross into the final artifact and that the agent should "Extract the current-state answer, not the code," providing a logical constraint on the data processing.
- Capability inventory: The skill performs file reads across the codebase and writes results to a structured markdown file at
docs/product/brainstorm.md. - Sanitization: No technical sanitization or escaping mechanisms are defined for content read from the codebase before it is used to influence the agent's logic.
Audit Metadata