email-triage

name	email-triage
description	Read email threads → categorize by urgency → draft responses for review → flag priorities → identify newsletters for unsubscribe. Use when processing an email backlog.

Email Triage

Source: $ARGUMENTS (defaults: ./emails/ → ./triage/)

Deterministic pipeline: read threads → categorize → draft → output review package.

Urgency Categories

Priority	Label	Criteria
P1	urgent	Deadline within 24h, financial/legal matter, executive sender
P2	action-needed	Requires response or decision, no immediate deadline
P3	fyi	Informational only, no action required
P4	newsletter	Marketing, subscription, automated digest
P5	spam	Unsolicited, phishing indicators, irrelevant

Workflow

1. Discover emails — read $SOURCE_DIR for .eml, .mbox, or .json export files
2. Parse threads — group by thread/subject; for each thread extract:
   • subject, sender, date (most recent message), body (latest + context)
3. Categorize — assign priority label using Urgency Categories table
4. Draft responses — for P1 and P2 threads only:
   • Write a concise draft response (≤150 words)
   • Mark draft as [DRAFT — REVIEW BEFORE SENDING]
   • Do NOT include personal sign-off — leave placeholder [Your name]
5. Flag newsletters — for P4 threads, extract unsubscribe link if present
6. Write outputs:
   • $OUTPUT_DIR/triage-summary.md — all threads sorted by priority with drafts inline
   • $OUTPUT_DIR/urgent.md — P1 threads only (quick action view)
   • $OUTPUT_DIR/unsubscribe-list.md — P4 newsletter unsubscribe links
7. Print summary — count per priority label

Input Contract

$SOURCE_DIR/
  *.eml | *.mbox | *.json   — exported email files or mailbox dumps

Output Contract

$OUTPUT_DIR/
  triage-summary.md     — full categorized list with response drafts
  urgent.md             — P1-only quick action list
  unsubscribe-list.md   — newsletter unsubscribe links
stdout                  — counts: P1=N, P2=N, P3=N, P4=N, P5=N

Constraints

• Do NOT send any emails — output drafts only for human review
• Do NOT store email content outside $OUTPUT_DIR
• Draft responses for P1/P2 only; P3–P5 get categorization only
• If thread body is empty or unreadable, categorize as P3 with note [body-unavailable]
• Phishing indicators (mismatched sender domains, suspicious links) → P5 + note [phishing-suspected]