Handle your inbox and support messages with AI
Skills used in this guide (4)
给收件箱做分诊：排优先级、起草安全的回复草稿。
npx skills add NousResearch/hermes-agent -g把 Drive、Docs、Sheets、Slides 当成一个工作流来用，处理计划、跟踪表、演示稿与共享文档，不必下沉到裸工具调用。
npx skills add affaan-m/ECC -g通过 Stripe 等已接入的计费工具处理订阅、退款、流失分级、计费门户恢复与套餐分析。适合协助客户、查看订阅状态或处理影响收入的计费操作。
npx skills add affaan-m/ECC -g用 Himalaya CLI 在终端里通过 IMAP/SMTP 收发邮件。
npx skills add NousResearch/hermes-agent -g
The red line for email automation is "draft yes, auto-send careful". "Triage + draft + you hit send" is safest; letting AI blast-send means you own the mistake, not it.
Who this is for / not for
For:
- Inbox perpetually overflowing, want to triage then act
- Support / presales repetitive Q&A, need consistent draft口径
- You want "draft + evidence", not full auto-send
Not for:
- Legal / contract / money-deciding replies (human must see)
- Blast marketing email (that's a different compliance flow)
- Unauthorized handling of messages with others' PII
Before you start
- Email access: Gmail auth (google-workspace-ops) or IMAP/SMTP (himalaya).
- 口径 library: standard reply templates (refund policy, lead time).
- Send rules: what can auto-send, what needs human confirm — decide first.
Steps
1. Install the skill
email-inbox-triage is an evidence-first inbox skill. Copy its folder into your client's skills directory (same as any Agent Skill):
| Client | Directory |
|---|---|
| Claude Code | ~/.claude/skills/ |
| Codex | ~/.codex/skills/ |
| Cursor | ~/.cursor/skills/ |
| Trae | ~/.trae/skills/ |
| CodeBuddy | ~/.codebuddy/skills/ |
| Tongyi Lingma | ~/.lingma/skills/ |
2. Triage first, then draft
Read my inbox, triage into [urgent / this week / later].
For each: sender intent, suggested category, a ready-to-send reply draft.
For refunds or billing, quote our refund policy first, don't promise amounts.3. Support tiers
For support use customer-billing-ops: refunds, churn tiers, billing-portal recovery follow policy; AI only drafts, money needs human confirm.
4. You hit send
After AI drafts, confirm each. For traceability, have it write back "sent / pending" status.
Pitfalls
| Pit | Symptom | Fix |
|---|---|---|
| Auto-send fails | Wrong person / over-promise | Default draft only, send needs confirm |
| Inconsistent口径 | Two replies differ | Feed standard templates, no free-form policy |
| Privacy leak | Draft carries others' info | Mask PII first |
| Money promised wrong | Refund amount off | Money forced human-confirm, quote policy |
| No priority | Important buried | Triage by urgency before drafting |
FAQ
Can AI send for me?
Better not. Default draft only, send by you; full auto needs explicit rules and low-risk only.
Consistent support tone?
Give standard reply templates (refund, lead time, quote) as context; require "strict template, no improv".
Multiple mailboxes?
Yes, google-workspace-ops treats Gmail/Drive/calendar as a workflow; IMAP multi-account via himalaya.
Can AI decide refund amount?
No. AI quotes policy, drafts comms; final amount must be human-confirmed.
Prove "it was sent"?
Have AI write back send status + log, or wire ticketing for trail.
Related guides
Continue across these scenes:
- Write cold emails and outreach with AI
- Write weekly reports and standups with AI
- Turn meeting recordings into a task list with AI
相关源码
- NousResearch/hermes-agent · email-inbox-triage source
- affaan-m/ECC · google-workspace-ops source
- Agent Skill open format spec
一句话结论
结论：邮件自动化的红线是「草拟可以、代发要慎」——分诊加草稿加你点发送，最稳。
References
Related Scenes
Write PRDs and product specs with AI
The pain of PRD isn't "wrong format" — it's "goal and acceptance unclear, dev guesses". The point: give background, target user, problem; let AI output "feature list + acceptance + edges", not decide "why build" for you.
Write weekly reports and standups with AI
The pain of weekly reports isn't "can't write" — it's "turning a log into outcomes". The point: give AI the raw material (what you did, what's stuck), have it close into "progress / risk / next", not invent content for you.
Translate and localize docs with AI
Machine translation's scare is "fluent but distorted" — wrong terms, wrong tone, stiff local phrasing. The point: give a glossary and target-reader tone, have it "translate + keep proper nouns", then you verify key sentences — not trust the whole blindly.
Turn meeting recordings into a task list with AI
The biggest waste in meetings isn't the meeting — it's nobody remembering "who, by when, what" afterward. The point of this flow is to compress scattered discussion into a task table with owners and due dates, not to write another minutes doc nobody reads.