deep-interview
Pass
Audited by Gen Agent Trust Hub on Apr 28, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [SAFE]: The skill implements a structured workflow for requirements gathering with clear boundaries and restricted autonomy.
- [PROMPT_INJECTION]: The skill contains an indirect prompt injection surface. (1) Ingestion points: User answers enter the agent context in Stage 2.2 (SKILL.md). (2) Boundary markers: Prompts use explicit headers to delimit user input. (3) Capability inventory: The skill utilizes Read, Grep, Glob, Write, and Agent tools (SKILL.md). (4) Sanitization: The skill mitigates risks through 'Hard Rules' that explicitly forbid the generation of planning files, implementation code, or git commands.
- [COMMAND_EXECUTION]: Uses Read, Grep, and Glob tools for codebase context gathering in '--deep' mode. This is restricted to architecture and pattern discovery for the purpose of clarifying requirements.
- [DATA_EXFILTRATION]: The skill uses the Write tool to save insights to the local project directory ('deep-interview-outputs/'). No network-based exfiltration patterns or unauthorized data transmission were detected.
Audit Metadata