reviewing-code

Pass

Audited by Gen Agent Trust Hub on Aug 6, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill instructions and configuration are purely instructional and do not include any executable scripts, binaries, or requests for sensitive permissions.
  • [PROMPT_INJECTION]: The skill has an indirect prompt injection surface due to its core function of analyzing external code and pull request data. Evidence: 1. Ingestion points: requested changes, repository/PR context, and code diffs (SKILL.md). 2. Boundary markers: No explicit delimiters or instructions are provided to ignore commands embedded in the processed code. 3. Capability inventory: No tool execution, subprocess calls, or file-write capabilities are defined within the skill. 4. Sanitization: No validation or escaping of the ingested code content is performed. This vulnerability is inherent to the primary purpose of code review.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 6, 2026, 04:07 PM
Security Audit — agent-trust-hub — reviewing-code