reviewing-with-peer-agents

Pass

Audited by Gen Agent Trust Hub on Aug 25, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to handle and process data from external AI models (peer agents), which is an inherent attack surface. However, the instructions incorporate strong defensive patterns: it mandates that every response be treated as untrusted review input, requires verification of material claims against the source project, and establishes a strict 'resolve-authorized' mode that prevents unauthorized file modifications, commits, or pushes without explicit user intervention.
  • [DATA_EXPOSURE_&_EXFILTRATION]: The skill explicitly prohibits the transmission of private chain-of-thought or unrelated sensitive context to peer agents, limiting data sharing to the minimum sufficient artifacts required for the review.
  • [COMMAND_EXECUTION]: The skill does not define or execute any shell commands, tools, or APIs directly. It functions as a procedural guideline for existing agent capabilities.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 25, 2026, 05:47 AM
Security Audit — agent-trust-hub — reviewing-with-peer-agents