用 Jev 给多 bot 编排器做类型化分流
开发者给 Firstmate 编排器加了一层 Jev 判断：一次 System One 调用返回 route/kind/urgency 等结构化结果，先影子运行再决定是否自动路由。
Next layer on my @grok @bot build: typed triage with @typesafeai Jev sitting in front of the Grok Ship by @kunchenguid vs direct-handoff split. The problem I’m trying to solve is routing the wrong kind of work to the wrong agent. My Firstmate orchestrator has specialist bots for ops, school, inbox, code, etc. Earlier I split the flow into two buckets: Repo work → Grok Ship (scout → ship → review → PR; I merge) Everything else → direct handoff to the right specialist That already stopped a lot of “turn this scheduling task into a software project” behavior. Jev sits in front of that split. One System One call takes the ask + source and returns structured judgments: 'route' Choice - which bot 'kind' Choice - scout / ship / ops_live / fyi / decision 'urgency' Score - can_wait / today / now Nouls - needs me? software repo? live customer? noise? Code still owns the handoff + factory DB. Jev doesn’t merge or auto-route. Every judgment gets logged to append-only JSONL (ask, source, route + confidence, kind, urgency, Nouls, model id). After I dispatch I stamp actual_route and score match rate - “what would Jev have done?” next to “what did I do?” Early tests with jev-1.13.0 are already separating things like: “ship a fix to the messaging repo” → coding / ship / now “move tomorrow’s appointment to noon” → ops “FYI: customer confirmed” → fyi / no action I’m running it in shadow for a couple weeks of real handoffs. If high-confidence route match stays strong and it doesn’t mislabel software vs ops vs noise, I take those lanes out of shadow. Until then it only recommends.
这条是引用着另一条写的：x.com/ssbrouhard/status/2094992750378664267