Email Triage Skill

Classify emails and draft replies with guardrails to prevent wrong sends.

emailtriageproductivitycommunication

Email Triage Skill

TL;DR

The Email Triage skill reads your inbox, classifies each message by type and urgency, and drafts replies for your review — without sending anything until you explicitly approve. It’s designed for people who spend more than 90 minutes a day on email and want to reclaim that time without losing control of what goes out.

This is a High risk skill because it touches your email account directly. A misconfigured skill could draft replies to the wrong thread, expose sensitive information in logs, or — in the worst case — send emails without your review. Read the guardrails section carefully before enabling automated sending.

What it does

  • Classifies incoming messages into categories you define: action required, FYI only, newsletter, support request, invoice, spam candidate, and so on.
  • Assigns urgency scores based on sender VIP status, keywords (“urgent,” “deadline,” “contract”), and thread age.
  • Drafts context-aware replies that reference the original message, match your writing tone, and include relevant details from your calendar or previous threads.
  • Flags sensitive content — messages containing SSNs, credit card numbers, legal language, or HR topics — for human-only handling rather than automated drafting.
  • Surfaces follow-up reminders for threads where you sent a message but received no reply within a configurable window (e.g., 3 business days).
  • Generates a daily triage summary showing how many messages were classified, how many drafts are waiting for approval, and which threads need attention today.

Best for

Customer support inboxes: Classify incoming tickets by product area, draft templated responses for common issues, and escalate anything that mentions legal action or billing disputes to a human queue.

Executive assistants: Triage a principal’s inbox, draft replies in their voice, and surface only the messages that genuinely require their attention — not every newsletter and calendar notification.

Sales follow-up: Flag threads where a prospect hasn’t replied in 5 days, draft a gentle follow-up, and queue it for review before sending. Combine with calendar scheduler to propose meeting times in the same draft.

Personal inbox zero: Run a weekly triage pass on your personal inbox. Archive newsletters, draft replies to friends and family, and surface the three emails that actually need action today.

This skill is not appropriate for inboxes containing attorney-client privileged communications, medical records, or financial account credentials — the classification model may not handle these categories reliably, and the risk of mishandling is too high.

How to use (example)

Scenario: Support inbox triage for a SaaS product

Your support inbox receives 80–120 emails per day. You want to classify them, draft replies for common issues, and escalate billing and legal emails to a separate queue.

Classification rules:

categories:
  - name: "billing_dispute"
    keywords: ["refund", "charge", "invoice", "overcharged"]
    action: "escalate_to_human"
    do_not_draft: true

  - name: "bug_report"
    keywords: ["error", "broken", "not working", "crash", "500"]
    action: "draft_reply"
    template: "bug_acknowledgment"

  - name: "feature_request"
    keywords: ["would love", "wish", "can you add", "feature request"]
    action: "draft_reply"
    template: "feature_request_thanks"

  - name: "newsletter"
    sender_patterns: ["@mailchimp.com", "noreply@", "no-reply@"]
    action: "archive"
    do_not_draft: true

Draft output for a bug report:

Subject: Re: Error when exporting CSV

Hi Sarah,

Thanks for reaching out — I'm sorry you're running into trouble with the CSV export.

I've flagged this for our engineering team to investigate. In the meantime, could you share:
1. The browser and OS version you're using
2. A screenshot of the error message, if possible

We'll follow up within 1 business day.

[DRAFT — awaiting approval before send]

Common variations:

  • Enable send_after_approval: true to send drafts with a single click rather than copy-pasting.
  • Add a vip_senders list so emails from your CEO or key clients always skip automated drafting and go straight to your attention queue.
  • Combine with support macros to pull pre-approved response templates rather than generating new text each time.

Permissions & Risks

Required permissions: Email (read + draft; optionally send)
Risk level: High

Email triage is the highest-risk skill in the productivity category. Here’s why:

Wrong-recipient risk: If the skill misidentifies a thread (e.g., two contacts with similar names), a draft reply could go to the wrong person. Always review drafts before sending, especially for threads involving personal or financial topics.

Sensitive data leakage: Email content passes through the AI model for classification and drafting. If your emails contain passwords, SSNs, or confidential business information, that data is processed by the model. Check your provider’s data retention policy before enabling this skill on sensitive inboxes.

Classification accuracy on edge cases: The skill will misclassify some emails — a legal notice that doesn’t use obvious keywords, a VIP sender writing from a personal address, or a phishing email that looks like a legitimate support request. Build in a “review queue” for low-confidence classifications rather than acting on them automatically.

Recommended guardrails:

  • Never enable auto-send until you’ve reviewed at least 50 drafts manually and confirmed quality.
  • Maintain a do-not-draft list for senders and topics that should always go to human review.
  • Log every classification decision with the reasoning, so you can audit and improve the rules over time.
  • Set a daily draft cap (e.g., max 20 drafts per day) to prevent runaway automation if the skill misbehaves.

Troubleshooting

Skill is classifying newsletters as action-required emails
Your classification rules don’t have a strong enough newsletter pattern. Add sender domain patterns (noreply@, @mailchimp.com, @substack.com) and check whether the emails have List-Unsubscribe headers — a reliable newsletter signal.

Drafts don’t match your writing tone
The skill is using a generic tone. Provide 5–10 examples of your actual sent emails as tone references in the skill configuration. Most implementations support a “writing style examples” field.

Sensitive emails are being drafted instead of escalated
Your sensitive content detection keywords are too narrow. Add variations: “attorney,” “counsel,” “legal hold,” “HIPAA,” “PII,” “confidential” — and test with real examples from your inbox.

Draft replies are missing context from earlier in the thread
The skill is only reading the most recent message, not the full thread. Enable include_thread_context: true in your config, or increase the thread_depth parameter to include the last 5 messages.

Skill is creating duplicate drafts
This happens when the skill runs twice on the same message (e.g., a webhook fires twice). Add idempotency checking using the message’s unique ID to prevent duplicate draft creation.

Alternatives

SaneBox — A managed service that filters and prioritizes your inbox using behavioral signals (not AI drafting). Lower risk than full triage automation; doesn’t draft replies. Good starting point if you want inbox organization without the complexity of AI-generated responses.

Gmail filters + canned responses — Zero AI, maximum control. Create filters for known senders and categories, then use Gmail’s canned responses (Templates) for common replies. Requires manual setup but is completely transparent and free.

Missive AI — A team email client with built-in AI drafting and shared inbox features. Better for teams than individuals; requires migrating to Missive as your email client rather than layering a skill on top of your existing setup.

Links & sources

  • Gmail API documentation: developers.google.com/gmail/api
  • Microsoft Graph API (Outlook): learn.microsoft.com/en-us/graph/api/resources/mail-api-overview
  • Related guide: Safe Skill Workflows
  • Related guide: Skills Security & Privacy

Related Skills

Explore more AI skills in our directory.

Back to Skills