omh-jev-ask
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill follows security best practices by requiring explicit user consent before processing data and by logging only metadata (hashes, cost) while explicitly forbidding the capture of keys or question text. It also includes an automated check to remove credential-like content from inputs.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied text for evaluation. Risk is low as the tool output is restricted to numeric probabilities which cannot contain malicious instructions for the agent.
- Ingestion points: User-supplied 'text to judge' and 'questions' (SKILL.md).
- Boundary markers: Transparent notification to the user before data is sent to the model (SKILL.md).
- Capability inventory: Limited to calling the specialized omh_jev_ask tool (SKILL.md).
- Sanitization: Validates input for secret-looking text and removes it before transmission (references/full-contract.md).
Audit Metadata