Morning skill: Auto-label emails on archive using semantic classification

[omriariav]

Problem

During morning triage, label operations are expensive. Every time the user wants to label + archive, a sub-agent is spawned that:

1. Fetches the full label list (925+ labels)
2. Does fuzzy matching
3. Applies the label + archives

This happened 4 times in a single session (Privacy 2x, LiveRamp, Tracking Products), adding significant latency and token usage.

Proposed Solution

Make "Archive" smarter by auto-classifying the best label:

1. Cache the label list at session start (see Morning skill: Reduce triage agent output verbosity #38)
2. When user picks "Archive", spawn a lightweight label-classifier sub-agent that:
   • Reads email subject + sender + snippet + any OKR/task context
   • Uses the cached label list
   • Semantic-matches to the best label
   • Applies label + archives in one shot
3. User sees: Archived with label: 14 - PRIVACY — no manual input needed
4. User can override: "No, use label X instead"

Alternative: Pre-classify Labels in Triage Agent

The triage agent already has email context. It could suggest the best label as part of its output, so the main agent can auto-apply without spawning another sub-agent.

Impact

• Saves ~20-30s per label operation (no sub-agent spawn + label fetch)
• Saves ~15k tokens per label operation
• Better UX: user doesn't need to remember label names

Related

• Morning skill: Reduce triage agent output verbosity #38 (label list caching)
• Morning skill v0.3.0 QA session findings

[omriariav]

Closing — morning skill development has moved to the pulse plugin. Out of scope for this repo.

[omriariav]

Correction: the pulse plugin lives in an internal repo. Morning skill development has moved there and is out of scope for this repo.