Support ✦ 95outlineNot a production recipe — calibrate thresholds

客服工单分流

工单状态 + Choice/Score;低置信度升级人工。

Who: Support / ops engineers owning queues. Steps: (1) Build state — message, sender, open orders, policy lists. (2) One call: Choice intent + Score complexity (+ optional Nouls). (3) Code confidence-gates → human, or routes to lookup / specialist LLM / human. Expected effect: cheap classify before expensive handlers; uncertain tickets escalate.

1

复制提示词(保留英文原文效果最好)。

2

在代码里实现——调用 System One / Jev,由你的程序组合答案。

3

回答它的设置问题——大多数提示词会先采访你,然后进入固定节奏。

提示词

提示词
Audience: support / ops.
Steps:
1) Deterministic pre-checks in code (closed tickets, missing fields).
2) State: message, sender, open orders, policy lists — omit secrets.
3) Parallel: Choice intent/topic; Score complexity; optional Nouls (needs_human, requests_credentials).
4) Gate: high confidence → route; mid → suggest; low → human.
Expected effect: closed-label routing with confidence-aware escalation.
Do not let Jev mutate tickets or send mail.

需要授权:typesafe-sdkyour-ticketing

它是怎么搭起来的

以下说明为英文原文(来自社区作者),提示词本身建议保留英文。

Map queues to Choice criteria (what / not_for / examples). Call System One once per ticket; compose the route in application code. Cross-link: Hands-on support routing. Sources: intent-routing · how-to-build-with-system-one · awesome-jev-usecases · Routing.

Include an other / exception Choice option so unknown intents are not forced into a wrong queue. Optional speculative Nouls can branch follow-up questions in one round-trip. Production-scale routing claims from community repos are author-reported — calibrate on your ticket mix.

Sources (wave-2 deepen)

为什么它好用

Canonical first System One shape: closed answer space, high repeat rate, judgment instead of brittle rules.

本案例收集自公开的 Jev 社区,版权归原作者所有。提示词属于参考资料:运行前请先审阅内容, 并且不要让 bot 超出你实际授予的权限行事。