claude-email-triage

KO | EN

gitlite — search

#javascript #python #ai #android #rust #deep-learning #linux #nodejs #java #kubernetes #security #mcp #openai #open-source

claude-email-triage

(★ 10)

Autonomous email triage agent for Claude Code — runs on a schedule, classifies Gmail, drafts/auto-sends replies, kills spam

claude-email-triage 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Claude Email Triage

An autonomous email triage agent for Claude Code that runs on a schedule, classifies unread Gmail, takes action (reply, draft, archive, spam-kill, schedule), and produces a triage report. Fully customizable — edit one skill file to match your workflow.

## What it does

- Fetches unread emails from one or more Gmail accounts
- Classifies each email (P0 urgent, P1 action, P2 review, P3 archive, spam/cold)
- Takes action based on confidence: auto-send on 95%+, draft on 70-94%, flag below 70%
- Auto-kills spam and cold outbound (replies "stop", marks spam, trashes)
- Handles meeting requests against your calendar
- Learns over time via a persistent state file (spam senders, auto-send patterns, follow-up queue)
- Produces a daily/per-run triage report

## How it runs

On a schedule you set (e.g., 8am / 12pm / 3pm / 6pm / 9pm daily), via **Claude Code Desktop's local Scheduled Tasks** feature. Runs fully unattended with the bypass-permissions setup in `SETUP.md`.

Uses the [`gog` CLI](https://github.com/steipete/gogcli) for Gmail/Calendar/Drive access. The orchestration (gog calls, scheduled tasks, state file) runs locally on your Mac, but **email content is sent to Anthropic** for Claude to read, classify, and draft replies — same as any Claude conversation. See [Privacy](#privacy) below.

## Quick start

Full setup takes ~30 minutes from scratch. See **[SETUP.md](SETUP.md)** for step-by-step instructions.

In short:
1. Install Homebrew and [`gog`](https://github.com/steipete/gogcli)
2. Set up a Google Cloud project for OAuth
3. Connect `gog` to your Gmail
4. Copy `config.example.md` → `~/email-triage/config.md` and fill in your accounts, VIPs, tone, meeting prefs
5. Copy `skill/SKILL.md` → `~/.claude/scheduled-tasks/email-triage/SKILL.md` (do not edit it)
6. Enable bypass-permissions mode so it runs without prompts
7. Create the scheduled task(s) in Claude Code Desktop

## Architecture: logic vs config

The repo separates what you shouldn't touch from what you must customize:

- **`skill/SKILL.md`** — pure agent logic: classification rules, actions, safety checks. **Do not edit.** Drop upstream updates in freely.
- **`config.example.md`** → user copies to **`~/email-triage/config.md`** — your accounts, VIPs, tone, meeting preferences, project context. **This is the only file you edit.**

When the skill runs, it reads `~/email-triage/config.md` at Step 0 before doing anything else. Your settings never get overwritten when the skill is updated.

## Customization

Edit one file: `~/email-triage/config.md` (copied from `config.example.md`).

Sections to fill in:
- **Accounts** — one row per Gmail alias you set up with `gog`
- **VIP Safe Senders** — people who should never be filtered
- **Meeting Preferences** — default length, timezones, calendar link, hours
- **Tone and Style** — how drafts should sound in your voice
- **Project Context** — who you are, what you work on, sensitive topics to flag

## Safety

The skill has hard-coded safety rules:
- Never auto-send to investors, board, or legal contacts
- Never auto-send anything about money, contracts, or commitments
- Never auto-send to people you haven't emailed before (draft instead)
- Never delete unless matching explicit delete rules; archive when unsure
- Never modify calendar events created by others
- On phishing suspicion: flag P0, do not click links

Even in bypass-permissions mode, the skill creates **drafts** for anything below 95% confidence — you review before it goes out.

## Requirements

- macOS (tested on Sonoma+)
- Claude Code Desktop app ([download](https://claude.com/claude-code))
- Claude Pro or Max plan
- Gmail account(s) with API access (free to set up)

## Privacy

Be aware of what data goes where:

- **Stays on your Mac**: the SKILL.md, the `~/email-triage/state.json` learning state, the `~/email-triage/log.md` run log, your `gog` OAuth tokens, and the cron schedule.
- **Goes to Google**: your Gmail/Calendar/Drive API requests, via `gog` (same as any Gmail client).
- **Goes to Anthropic**: the **content of every email you triage**, the prompt, the state file, and the drafts Claude generates. This is how the model classifies and writes replies — it has to read the email content. Anthropic's data handling policies for your Claude plan apply (see https://www.anthropic.com/legal).

If you have emails you don't want sent to a third-party LLM at all (NDAs, legal, medical, etc.), don't triage those accounts with this — or filter them out of the unread set before classification.

## License

MIT — see [LICENSE](LICENSE).

## Credits

Originally built for personal use across multiple Gmail accounts; generalized and open-sourced so others can run the same setup.

claude-email-triage

claude-email-triage 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

claude-email-triage

claude-email-triage 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README