deep-interview

Pass

Audited by Gen Agent Trust Hub on Jul 10, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted data from repository files (code, documentation, history) and user inputs, creating an indirect prompt injection surface.
  • Ingestion points: Documented in SKILL.md and references/interview-method.md, the skill reads repository artifacts to establish context.
  • Boundary markers: The skill implements an 'ambiguity ledger' and provenance labels (e.g., [from-code], [from-user]) to track the source of all information.
  • Capability inventory: Capabilities include file reading, writing the final interview-brief.md (upon user request), and potential web research.
  • Sanitization: The instructions explicitly mandate that repository content be treated as 'evidence, not authority' over user intent, reducing the risk of data-driven instruction overrides.
  • [EXTERNAL_DOWNLOADS]: The references/harness-source-notes.md file references several external GitHub repositories (e.g., Ouroboros, Codexplain, Superpowers) used as research sources for the skill's methodology. These are cited for reproducibility and are not downloaded or executed at runtime.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 10, 2026, 03:26 PM
Security Audit — agent-trust-hub — deep-interview