TypeSafe Jev 模板（System One）

工单路由

将进线工单路由到 tech / sales / billing / human。例：API returns 500 on checkout → tech。

// model: typesafe-ai/jev
questions: {
  route: {
    type: 'choice',
    instructions: 'Route this support ticket to the right queue.',
    criteria: {
      tech: 'Bugs, outages, errors, API failures, integrations, or product not working as designed',
      sales: 'Pricing, plans, upgrades, demos, or new-purchase intent before becoming a customer',
      billing: 'Charges, invoices, receipts, subscriptions, payment method, or invoice questions',
      human: 'Ambiguous, multi-issue, sensitive, legal, account takeover, or otherwise needs a person',
    },
  },
}

内容审核

将 UGC 分流到 allow / block / human。例：buy followers cheap DM me → block。

questions: {
  verdict: {
    type: 'choice',
    instructions: 'How should this content be handled under community policy?',
    criteria: {
      allow: 'Normal participation that clearly complies with policy',
      block: 'Spam, scams, prohibited promo, harassment, or clear policy violations',
      human: 'Borderline, context-dependent, or policy-unclear — needs a moderator',
    },
  },
}

直播评论标注

粘贴评论 → 建议 buy_intent / price_question / after_sales / spam / other。例：in stock? ship to US? → buy_intent。AI 仅建议 — 绝不自动退款或收款。

Agent 命令风险

对拟执行的 shell/工具调用做仅预览的 Score/Boolean。帮助人决定是否运行 — 不是 CMD Guard 产品、DOM 扩展或自动拦截器。

questions: {
  destructive: {
    type: 'boolean',
    instructions: 'Could this command destroy or exfiltrate data if executed as written?',
  },
  risk: {
    type: 'score',
    instructions: 'Overall operational risk if executed as written.',
    criteria: ['low', 'medium', 'high', 'critical'],
  },
}