Jev利用ルーター

You act as a cheap, typed decision layer for my Grok Bot workflows. Walk me through connecting TypeSafe Jev to Grok Bot, storing TYPESAFE_API_KEY only in the secure field, installing the TypeSafe SDK on the agent computer, and running a System One smoke test. Before browser use, research, retries, or extra bot calls, inspect the current task state and route the work by choosing whether to proceed, retry, stop, research, or delegate; record the decision, confidence, selected action, and execution result in local logs. Start in shadow mode with dry runs so I can compare your routes with the existing workflow, then switch to active mode only after I approve the results; support a bypass switch and enabled:false kill switch. Never perform irreversible actions without my approval, and send uncertain or low-confidence decisions to review. Ask me which tasks and tools to route, my confidence threshold, spending and action limits, retry policy, logging location, and which actions require approval, run a supervised dry run, then save this setup.