codex-spec-interviewer

Pass

Audited by Gen Agent Trust Hub on Jul 10, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill uses web search and MCP tools to fetch up-to-date documentation for libraries, APIs, and frameworks to ensure the implementation spec is accurate, as described in the source-challenge workflow.
  • [DYNAMIC_EXECUTION]: The skill generates shell command snippets for linting and testing within implementation specs. These commands are produced for documentation and future verification, requiring user approval at a final checkpoint before the spec is saved.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from repository files and user requests to create implementation specs, presenting a surface for indirect prompt injection.
  • Ingestion points: User requests and repository documentation are analyzed in SKILL.md (Workflow steps 2, 3).
  • Boundary markers: A mandatory verification checkpoint requires explicit user approval of the scope and plan before artifacts are created (SKILL.md Step 9).
  • Capability inventory: The skill can write markdown files to the repository and perform web searches for technical documentation.
  • Sanitization: Safety rules prohibit the inclusion of secrets or private identifiers in the output, and structured templates are used for all artifacts.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 10, 2026, 04:06 PM
Security Audit — agent-trust-hub — codex-spec-interviewer