Skills
Navigate
skills/rlaope/oh-my-hermes/omh-jev-done-check/Gen Agent Trust Hub

omh-jev-done-check

Pass

Audited by Gen Agent Trust Hub on Nov 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted evidence data which creates a potential surface for prompt injection from external sources.
  • Ingestion points: The evidence_excerpt parameter (referenced in SKILL.md) contains observed command or test output that is passed to the external reviewer.
  • Boundary markers: The agent is instructed to explicitly disclose the content being sent externally to the user before the first tool call.
  • Capability inventory: The skill invokes the omh_jev_ask tool for external review and records metadata to a local file at <omh_home>/jev/asks.jsonl.
  • Sanitization: The full-contract.md includes instructions to remove credential_like_content from the payload before transmission.
  • [DATA_EXFILTRATION]: The skill is designed to transmit task goals and evidence to an external reviewer via the omh_jev_ask tool.
  • Evidence: This transmission is a documented core feature of the workflow and includes a disclosure step. The service utilizes OpenRouter, a well-known service for model access.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 9, 2026, 04:57 AM