omh-jev-ask

Pass

Audited by Gen Agent Trust Hub on Sep 25, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided text as input for the omh_jev_ask tool, which creates a potential surface for instructions embedded in data to influence model behavior. However, the skill implements strong mitigations.
  • Ingestion points: The state parameter in the omh_jev_ask tool call, which contains the user-provided text to be judged (SKILL.md).
  • Boundary markers: The skill instructions require the agent to tell the user in one line exactly what data will be sent before the request is made (full-contract.md).
  • Capability inventory: The skill is limited to the single tool omh_jev_ask and does not permit arbitrary code execution or file writes of user data.
  • Sanitization: The contract includes a recovery procedure to identify and remove credential_like_content from the state parameter before the tool is called.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 25, 2026, 08:33 PM
Security Audit — agent-trust-hub — omh-jev-ask