deep-interview
Pass
Audited by Gen Agent Trust Hub on Mar 12, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill uses internal role-play modes (Contrarian, Simplifier, Ontologist) to guide the AI's questioning style. These are legitimate instructional constraints for the intended Socratic interview process and do not attempt to bypass system safety protocols.
- [DATA_EXFILTRATION]: The skill accesses the local codebase using an 'explore' sub-agent to provide context for brownfield projects. This information is used locally to refine the interview and is not transmitted to unauthorized external endpoints.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input during the interview loop. While this input influences the generated specification and subsequent execution phases, the iterative nature of the process and the requirement for explicit user confirmation at each stage provide mitigation against accidental or malicious instruction following.
- [COMMAND_EXECUTION]: The skill invokes other specialized skills such as 'omc-plan' or 'autopilot' for final execution. This delegation follows a principle of least privilege where the interview agent itself does not perform direct system modifications, instead relying on dedicated execution agents.
Audit Metadata