claude-spec-interviewer
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill maintains an attack surface for indirect prompt injection because it ingests untrusted data from the repository to generate instructions for downstream implementation agents.
- Ingestion points: Processes current user requests, project documentation (README.md, AGENTS.md), issue descriptions, and repository source code.
- Boundary markers: The skill generates specifications using structured templates with explicit 'In scope' and 'Non-goals' sections, and requires a 'Verification checkpoint' for user approval.
- Capability inventory: Produces a 'Claude execution prompt' designed to guide automated code implementation; identifies validation commands for shell execution.
- Sanitization: Requires the agent to summarize assumptions, resolve unknowns through targeted questioning, and wait for explicit user verification before persisting any artifacts.
- [SAFE]: The skill incorporates strong safety guidelines, including explicit instructions to exclude secrets, credentials, or private identifiers from generated artifacts. It implements a 'Plan mode' preflight that restricts repository writes until user verification is obtained.
Audit Metadata