Morning skill: Reduce triage agent output verbosity

[omriariav]

Problem

Triage agents return ~35k tokens of JSON per batch with verbose fields:

• Full reasoning strings
• Priority signal lists
• Prep materials
• OKR alignment details
• Calendar conflict notes

Most of this isn't displayed to the user. The main agent only needs: id, category, priority, one-line summary, sender, and action_taken.

Proposed Solution

Define a lean output schema for triage agents:

{
  "auto_handled": [
    {"id": "x", "thread_id": "y", "action": "archived", "reason": "noise"}
  ],
  "needs_input": [
    {
      "message_id": "x",
      "thread_id": "y",
      "category": "ACT_NOW",
      "priority": 5,
      "sender": "name <email>",
      "subject": "...",
      "summary": "One line: why this matters",
      "message_count": 4,
      "matches": ["TOP 5: task name", "OKR: objective"]
    }
  ],
  "batch_stats": {"processed": 7, "noise": 3, "scheduling": 1, "act_now": 1, "review": 2}
}

Impact

• ~80% token reduction in triage agent output
• Faster response times (less to generate and parse)
• Main context stays leaner

Implementation

Update skills/morning/prompts/triage-agent.md output format section.

Related

• Morning skill v0.3.0 QA session findings

[omriariav]

Closing — lean triage output was implemented in pulse (Phase 1, PR #38 in pulse repo). Out of scope for this repo.

[omriariav]

Correction: lean triage output was implemented in the pulse plugin (internal repo). Out of scope for this repo.