Operations Email Report

operations-email-report

Three jobs, in this order:

Job 1 — Categorize: Search Gmail for unread + important emails from the last 24 hours. Categorize each one using references/content-patterns.md (matches by content fingerprint, not sender domain — survives tool swaps). Buckets: INBOUND / CLIENT / REVENUE / ADMIN / FYI / UNKNOWN.

Job 2 — Score: Apply references/email-rubric.md to assign a Priority Score (0-8) and badge (🔥 / 🔴 / 🟡 / ⚪). Score decides surfacing order; bucket decides which section.

Job 3 — Surface dropped balls: Separate query — find threads where you last replied >3 days ago and the other side hasn't responded. Surface in 🟠 Dropped Balls section.

Automation level: AUTONOMOUS — read-only. Nothing is sent, labeled, or modified in Gmail.

Routine position: Morning routine #1 — feeds operations-compile-daily-plan (skill #6).

Setup Required

Replace these placeholders in the context steps below with your actual Notion database IDs:

Also update references/sender-aliases.md with your own contacts and references/noise-blocklist.md with your own noise domains.

Context to read before running

1. Read context/current-priorities.md — current priorities. An email tied to a current priority gets +1 to score (see rubric).
2. Read context/business-profile.md — known client/prospect company domains.
3. Read context/icp.md — for the form submission ICP mini-score.
4. Read all four reference files in this skill:
   • references/email-rubric.md — scoring math
   • references/content-patterns.md — bucket-detection rules
   • references/sender-aliases.md — alias mapping
   • references/noise-blocklist.md — auto-FYI domains and self-suppress
5. Query the Notion Client Deliverables data source (collection://YOUR_CLIENT_DELIVERABLES_NOTION_DB_ID) — list of active clients (page IDs and names). Powers "Active client = +3" sender match.
6. Query the Notion People data source (collection://YOUR_PEOPLE_CRM_NOTION_DB_ID) — prospects in CRM (Status ≠ Lost / Cold). Powers "CRM prospect = +2" sender match.

Steps

Part 1 — Search Gmail

1. Three Gmail queries in parallel using mcp__claude_ai_Gmail__search_threads:

   • Unread (24h): query: "is:unread newer_than:1d", pageSize: 30
   • Important unread (24h): query: "is:important is:unread newer_than:1d", pageSize: 20
   • Dropped balls (you last replied 3-14d ago): query: "from:me older_than:3d newer_than:14d", pageSize: 30

2. Merge results from queries 1 + 2. Deduplicate by thread ID. Keep 30 most recent if total exceeds 30.

3. For each thread in the merged list, call mcp__claude_ai_Gmail__get_thread with messageFormat: FULL_CONTENT. Extract:

   • sender name + sender email + sender domain
   • subject
   • first 200 chars of body
   • thread length
   • who sent the most recent message
   • first sender in thread (to detect "you sent first → reply from prospect" override)

Part 2 — Pre-filter (suppress + blocklist)

4. For each email, check references/noise-blocklist.md:
   • Self-suppress matches → drop entirely. Don't include in counts or report.
   • Domain blocklist matches → force bucket = FYI, score = 0. Skip to Part 5.

Part 3 — Categorize by content pattern

5. For each remaining email, walk through references/content-patterns.md in this order:

   • INBOUND patterns (1-5)
   • CLIENT patterns (1-5)
   • REVENUE patterns (1-5)
   • ADMIN patterns (1-5)
   • FYI patterns (1-6)
   • If none match → bucket = UNKNOWN

   First match wins. Record which pattern matched (used for action-item template).

6. Resolve sender aliases via references/sender-aliases.md before checking active-client / CRM lookup. (Same person across 2 emails should count once.)

Part 4 — Score with rubric

7. Apply references/email-rubric.md:

   • Special-case overrides FIRST. If a content pattern matches a special override (inbound form = score 6, inbound booking = score 5, reply-to-outbound = min 3, time-bombed renewal = 3, SaaS-drip-pretending-to-be-human = 0, etc.) — apply it and skip the additive rubric.
   • Otherwise, sum the 5 signals (sender match + subject + time-sensitivity + thread state + priority bonus). Cap at 8.

8. VIP check — for each non-suppressed, non-blocklisted email:

   • Run a Gmail query: from:<sender> newer_than:30d (cache results across the run — don't re-query for the same sender).
   • Count distinct threads. If ≥ 3 AND domain is not bulk-mail (not mail.*, email.*, notifications.*, noreply.*) → apply +3 sender match (overrides "unknown sender" 0).

9. Assign badge from final score: 🔥 ≥ 6 / 🔴 4–5 / 🟡 2–3 / ⚪ 0–1.

Part 5 — Detect dropped balls

10. From the dropped-balls query (Part 1, query 3): for each thread:
    • Read full thread.
    • Check: did anyone reply AFTER your most recent message?
    • If no → it's a dropped ball.
    • Skip if your last message contained closure language: "ttyl", "no rush", "fyi", "thanks!", "no need to reply".
    • For each true dropped ball, record: who, subject, days dormant, last action.

Part 6 — Action items

11. For each email scoring ≥ 2 in INBOUND, CLIENT, or REVENUE buckets, write one action item using the template from the matched content pattern.

    • Format: "[Verb]: [topic] — from [Name]"
    • Keep under 15 words.
    • INBOUND form submissions get a parsed ICP mini-score appended: "ICP fit: HIGH/MEDIUM/LOW — [reason]"
    • Call recap CLIENT emails get one action item per commitment extracted from the body.

12. ⚪ emails (score 0–1) get no action item — informational only.

13. Add one action item per dropped ball: "Followup: [subject] — [Name] (X days dormant)"

Part 7 — Build and save output

14. Sort each bucket internally by score descending.

15. Save to inbox/outputs/md/YYYY-MM-DD-operations-email-report.md. If a file for today already exists, append with a ### Check — HH:MM header.

16. Print short summary inline (not the full report):

    • "Email report done: X INBOUND, X CLIENT (Y action), X REVENUE (Y action), X ADMIN, X FYI, X UNKNOWN, X dropped balls. Top score: [score] — [subject]."

17. If UNKNOWN section has entries, also print: "X emails couldn't be categorized — check the UNKNOWN section in the report and tell me what they are so I can add a rule to references/content-patterns.md."

Output Format

# Email Report — YYYY-MM-DD HH:MM

> X total scanned. INBOUND X, CLIENT X, REVENUE X, ADMIN X, FYI X, UNKNOWN X. Dropped balls: X. Action items: Y.

---

## Priority rubric (recap)
- 🔥 ≥6 / 🔴 4–5 / 🟡 2–3 / ⚪ 0–1 / 🟠 Dropped ball (separate)
- Active client or VIP (3+ threads/30d): +3 / CRM prospect: +2 / Known company: +1
- Subject "proposal/invoice/blocked/urgent": +2 / "pricing/call/follow-up": +1
- Body "today/EOD/ASAP": +2 / "this week": +1
- You holding the ball: +2 / You last replied: -1
- Priority bonus: +1
- Special overrides (form = 6, booking = 5, reply-to-outbound = min 3, blocklist = 0)

---

## 🆕 INBOUND (X)
| Score | From | Subject | Pattern matched | Action |
|-------|------|---------|-----------------|--------|
| 🔥 6 | [Contact Name] | Free Audit Call | Tally form submission | Reply within 24h. ICP fit: HIGH — [reason] |
| 🔴 5 | [Contact Name] | New booking | Cal.com | Meeting booked: [Name] on YYYY-MM-DD — run meeting prep 24h before. |

---

## 👥 CLIENT (X)
| Score | From | Subject | Preview | Action |
|-------|------|---------|---------|--------|
| 🔥 7 | [Client Name] | Re: project update | "When can I expect..." | Reply: ETA update — from [Client] |

---

## 💰 REVENUE (X)
| Score | From | Subject | Preview | Action |
|-------|------|---------|---------|--------|
| 🔴 4 | [Prospect Name] | Re: proposal | "...may still need your help..." | Warm reply — review thread, draft response |

---

## 🟠 Dropped Balls (X)
| Days dormant | Who | Subject | Last action |
|--------------|-----|---------|-------------|
| 5 | [Name] | [Subject] | You sent proposal v1, no response |

---

## ⚙️ ADMIN (X)
| Score | From | Subject | Preview | Note |
|-------|------|---------|---------|------|
| 🟡 3 | Excalidraw | Subscription renews | "..." | Time-bombed: 3 days. Keep or cancel? |

---

## 📌 FYI (X)
| Score | From | Subject | Note |
|-------|------|---------|------|
| ⚪ 0 | Apollo | [Subject] | Blocklist |

---

## ❓ UNKNOWN (X) — review and tell me how to categorize
| From | Subject | Why unclassified |
|------|---------|------------------|
| [Sender] | [Subject] | Not in CRM, no known pattern matched |

---

## Action Items for Today (top-scored only)
- [ ] 🔥 Reply within 24h: [Name] submission (HIGH ICP fit)
- [ ] 🔴 Meeting booked: [Name] on [Date] — run meeting prep 24h before
- [ ] 🟠 Followup: [proposal subject] — [Name] (5 days dormant)

Edge Cases

• Inbox empty / no unread emails: Output "Inbox clear — no unread emails in last 24h." Pass forward to daily plan.
• Gmail API fails or times out: Log "Gmail unavailable — email report skipped." Continue morning routine.
• Notion query for active clients fails: Skip the +3 sender bonus from active clients (VIP detection still works as fallback).
• Can't read full email body: Score sender + subject only. Time-sensitivity and thread state default to 0.
• Reference file missing or empty: Treat all emails as UNKNOWN (don't crash). Print "Reference file [name] is empty — fill it before next run."
• More than 30 emails in 24h: Process the 30 most recent. Note "X additional unread not scored — check inbox directly."
• Two emails from the same thread: Score both, but only generate one action item (use the most recent message).
• UNKNOWN bucket has more than 5: Show 5 most recent. Note "X more unclassified — review in inbox."
• VIP query times out: Skip VIP detection for that sender, fall back to standard scoring.
• Form parsing fails: Surface the email as INBOUND with action "Form received — couldn't parse fields, check inbox."

What good looks like

A finished email report is a decision aid, not a copy of the inbox. After reading it, you know in <60 seconds:

1. Which 1-3 emails to reply to right now (🔥 / 🔴 in INBOUND, CLIENT, REVENUE).
2. Which meetings need prep this week (INBOUND bookings).
3. Which threads you've dropped the ball on (🟠).
4. Which subscriptions are about to auto-renew (ADMIN time-bombed).
5. Whether anything new appeared that the system doesn't recognize (UNKNOWN).

The UNKNOWN section is the feedback loop — every entry there is a chance to make the skill smarter for next run.