relay-review

Pass

Audited by Gen Agent Trust Hub on Oct 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [SAFE]: The skill implements significant security engineering practices, such as using O_NOFOLLOW during file operations to prevent symlink attacks and performing SHA-256 verification of all staged inputs (diffs, prompts, and criteria) both before and after execution to ensure immutability.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted code diffs for review, creating a potential surface for indirect prompt injection. This is mitigated by the reviewer-prompt.md which uses clear XML-style boundary markers (<task-content>) and provides explicit instructions to the AI to treat the diff content as data rather than instructions, specifically warning against following directives embedded in the diff.
  • [COMMAND_EXECUTION]: The review-runner.js script executes local git commands for repository identification and diff generation. The implementation is secure, as it uses execFileSync and validates commit SHAs against a strict regular expression (SHA1_RE) before including them in command arguments, effectively preventing command injection.
  • [EXTERNAL_DOWNLOADS]: The skill requires the GitHub CLI (gh) and external AI adapters (e.g., Claude, Codex) for review delivery. These dependencies are standard for the skill's purpose and utilize the operator's ambient local authentication state rather than hardcoding or exfiltrating credentials.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 9, 2026, 12:25 AM