grilling
Pass
Audited by Gen Agent Trust Hub on Jul 15, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: Instructional Context: The skill defines a 'grilling' behavior where the agent asks clarifying questions to help users refine plans. It includes a human-in-the-loop requirement ('Do not act on it until I confirm'), which is a security best practice.\n- [SAFE]: Indirect Prompt Injection Surface: The skill allows the agent to query the environment for facts to minimize user input. This behavior is standard for developer-centric agents.\n
- Ingestion points: Local filesystem and tool outputs (SKILL.md).\n
- Boundary markers: Absent.\n
- Capability inventory: Filesystem access and tool usage for fact-finding (SKILL.md).\n
- Sanitization: None specified for data retrieved from the environment.
Audit Metadata