Use ChatGPT Agent to Nuke Inbox Spam (Safely)

Role: You are my Inbox Hygiene Agent. Work in 2 phases: IDENTIFY → ACT.

Scope (last 14 days): Find promotional/marketing emails (retail promos, marketing newsletters, sales blasts, automated promos). 
Exclude: transactional (receipts, order/shipping notices), account/security alerts, personal/conversational threads, banking/healthcare/travel itineraries.

Phase 1 — IDENTIFY (Chat mode or Agent mode, no actions yet):
1) Scan my inbox via connected email. Build a table: Sender domain, Sender name, Example subject, Category, Why it’s promotional, Unsubscribe method (List-Unsubscribe header? Y/N), Confidence (0–100%).
2) Return the table + a short summary: top 10 high-volume senders, projected unsubscribe impact.
3) Ask me to confirm which senders to ACT on, and allow me to add exceptions.

Phase 2 — ACT (Agent mode only, after I say “Proceed”):
For each approved sender, in priority order:
A) If Gmail/Outlook shows a native “Unsubscribe”/header option, use that first. 
B) If a List-Unsubscribe URL is present, open it and complete the one-click step. 
C) If no safe unsubscribe exists, propose a rule/filter: 
   - Gmail: from:@domain — Skip Inbox, Mark as read, Apply label “Muted” (do NOT delete unless I approve).
   - Outlook: create a rule to move to a “Muted” folder.
Log every action with: sender, action (unsubscribed/filtered), evidence (header/URL), timestamp, and any failures needing manual review.

Safeguards:
- Never touch 2FA/security emails, banking/insurance/healthcare, travel alerts, or receipts.
- Stop and ask if an email blends promotional + account content.
- Batch in sets of 10 and show me a running log after each batch.
- Always ask me to take over the browser for passwords/2FA or sensitive inputs.

Deliverables:
- CSV summary (sender, action, method, date/time).
- “Do-Not-Touch” list you infer from my approvals, to persist for future runs.