verify
Pass
Audited by Gen Agent Trust Hub on Jul 25, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes externally provided audit findings (issue lists) which could theoretically contain malicious instructions meant to influence the verification process. However, the skill uses structured prompts and clear instructions for the agent to use independent judgment.
- Ingestion point: User-provided issue lists interpolated into the prompt in SKILL.md.
- Boundary markers: Uses clear headers (e.g., 'ISSUES TO VERIFY:') but lacks explicit 'ignore instructions' delimiters.
- Capability inventory: Uses mcp__claude-code tools which have file-read and high-effort reasoning capabilities.
- Sanitization: No explicit sanitization of the input findings.
- [COMMAND_EXECUTION]: The skill is designed to interact with 'mcp__claude-code', an environment-specific tool for code analysis. While powerful, this is the primary intended function for verification and is configured for explicit invocation only.
- [SAFE]: The configuration in agents/openai.yaml explicitly disables implicit invocation (allow_implicit_invocation: false). This is a security best practice that prevents the agent from triggering this high-effort verification task automatically or entering unintended self-review loops.
Audit Metadata