full-code-review

Pass

Audited by Gen Agent Trust Hub on Jul 20, 2026

Risk Level: SAFE
Full Analysis
  • [PROMPT_INJECTION]: The skill implements strong defenses against indirect prompt injection. It includes explicit instructions to its sub-agents to ignore any commands or instructions embedded within the reviewed code, diffs, or PR metadata.
  • Ingestion points: Diff content and PR metadata fetched via git and gh (SKILL.md).
  • Boundary markers: Present; the script explicitly labels diffs as "untrusted input" and warns agents to ignore embedded instructions (scripts/full-code-review.js).
  • Capability inventory: Limited to git and gh CLI tools (SKILL.md).
  • Sanitization: Implements a redactSensitiveText function to mask secrets like API keys, tokens, and passwords before they are processed by LLM agents (scripts/full-code-review.js).
  • [CREDENTIALS_UNSAFE]: No hardcoded credentials found. The skill actively identifies and redacts secret-like patterns in the data it processes, which is a significant security best practice for code analysis tools.
  • [COMMAND_EXECUTION]: Tool usage is strictly limited to git and gh for repository context gathering. This is appropriate for a code review skill and no suspicious shell patterns or argument injections were detected.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 20, 2026, 01:29 PM
Security Audit — agent-trust-hub — full-code-review