grill-me
Pass
Audited by Gen Agent Trust Hub on Oct 11, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied plans, decisions, or ideas to generate a structured design tree and follow-up questions. This data is untrusted and could contain instructions designed to influence agent behavior.
- Ingestion points: Untrusted data enters the context via the initial plan/idea in the
/grill-mecommand and subsequent answers provided during the interview rounds described inSKILL.md. - Boundary markers: The instructions do not define specific boundary markers or delimiters for the agent to use when interpolating user-supplied text into its internal reasoning or tool calls.
- Capability inventory: The skill utilizes the
AskUserQuestiontool and possesses the ability to dispatch sub-agents to query the environment (filesystem and tools) for facts. - Sanitization: There is no instruction for the agent to sanitize, escape, or filter user input before it is used to drive the design tree logic or environmental fact-finding.
Audit Metadata