claude-code-review

Pass

Audited by Gen Agent Trust Hub on Jun 20, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local CLI tools including git, gh (GitHub CLI), and claude to perform its primary function. It includes explicit instructions to resolve paths to absolute values and quote user-provided inputs to prevent shell injection attacks.
  • [EXTERNAL_DOWNLOADS]: The skill identifies the Claude Code CLI as a prerequisite and directs users to the official Anthropic site (https://claude.com/claude-code) for installation. This is a reference to a well-known service and follows safe practices.
  • [PROMPT_INJECTION]: The skill is designed to process external repository code which could contain malicious instructions (indirect prompt injection). To mitigate this, the skill implements a comprehensive adversarial debate protocol and a 'reversed-role' stress-test to verify findings through multiple rounds of interaction.
  • [DATA_EXFILTRATION]: While the skill reads local repository data to perform reviews, no unauthorized network exfiltration patterns were detected. The data processing is limited to the interaction with the user-authorized Claude CLI.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 20, 2026, 01:11 PM
Security Audit — agent-trust-hub — claude-code-review