The Agent Skills Directory

[SAFE]: The skill follows security best practices by requiring user consent before transmitting any data externally. It explicitly instructs the agent to inform the user that their message will leave the local machine.
[SAFE]: Includes specific recovery procedures to handle sensitive data, requiring the removal of secret-looking text if the backend tool detects credential-like content.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user messages to perform its routing task. While this provides a surface for indirect prompt injection, the skill acts as a guide/gateway and does not execute the user input directly as code or shell commands. The impact is limited to routing suggestions presented to the user for clarification.
Ingestion points: User message processed in SKILL.md (Step 2) and references/full-contract.md (Workflow Contract).
Boundary markers: None explicitly defined in the prompt templates, but the system relies on the omh_jev_ask tool interface.
Capability inventory: Uses omh_jev_ask (external tool call) and omh_route_answer (local record logging).
Sanitization: Implements a safety rule to remove credential_like_content from the message state if the tool refuses processing.

omh-jev-route