second-opinion

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill contains a vulnerability surface as it processes external code diffs that could contain malicious instructions. It incorporates defensive measures including structural delimiters and explicit instructions in references/review-input.md to treat code as data.
  • Ingestion points: Code diffs are collected in references/review-input.md using standard Git commands.
  • Boundary markers: Instructions specify clear labels for requirements and data sections.
  • Capability inventory: The skill uses Bash to invoke analysis CLIs (codex, agy, gemini).
  • Sanitization: Explicitly instructs the backend model to treat repository material as data to analyze, not instructions to execute.
  • [EXTERNAL_DOWNLOADS]: The skill provides instructions for installing necessary review tools from verified organizations. It references @openai/codex and @google/gemini-cli from the NPM registry, which are trusted resources from well-known services.
  • [COMMAND_EXECUTION]: Shell commands are used to invoke the review tools. The skill implements secure configuration flags such as --sandbox read-only and --ephemeral for Codex, and --disable-slash-commands for Antigravity to ensure the analysis environment remains restricted.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 01:17 PM
Security Audit — agent-trust-hub — second-opinion