meeting-to-deal

Meeting to Opportunity

Flow: Trigger → Pull meetings → Match → Synthesize → Publish (α) → Auto-contact (β, opt-in) → Proactive next-step nudges (opt-in).

Input

• Optional: time window via argument (default = last 24h).
• Optional: pre-declared mode (Default / Silent / Proactive).

Output

• α (default): One Meeting recap update per matched deal, summarizing key points + commitments + next steps. Body carries the Generated by Claude · <ISO timestamp> footer + <!-- claude-skill-id: meeting-to-deal --><!-- meeting-id: <id> --> comments. A Meeting sync — <Mon DD> doc lists what was matched, what wasn't, and unresolved attendees.
• β (Auto-contact, opt-in): New items on the Contacts board for meeting attendees not already in CRM, linked back to the matched deal. Each new contact carries Source = Claude on a status column.
• Proactive extension: create_notification to deal owner for commitments due this week. Stage column edits and last-touch updates are also OK in Proactive (or in Default with batched confirm) when a clear transition phrase is present in the transcript ("we agreed to move to procurement review", "they signed today"). Surfaced as a single batched plan: " stage transitions detected. Apply all / review each / skip stage edits?". Amount-column writes still forbidden.

Knowledge

• Meeting → deal matching heuristic (§ Step 4).
• Update body convention (§ Step 6).
• Shared artifact conventions (§ Shared patterns).

Tools (MCP)

• get_user_context — user identity + email domain for filtering meetings and extracting internal_domain.
• get_notetaker_meetings — recent meetings + transcripts.
• search / list_workspaces / get_board_info — locate Deals + Contacts boards.
• get_board_items_page — fetch active deals to match against.
• get_column_type_info — resolve email/company columns by type.
• create_update — write the recap onto the matched deal.
• create_item — β auto-contact on the Contacts board.
• change_item_column_values — populate the new contact's email + company columns, write Source = Claude on contacts, and apply confirmed stage / last-touch edits on matched deals. Never for amount columns.
• create_notification — proactive next-step nudges.
• create_doc — sync summary doc.
• all_monday_api — escape hatch for connect-board column writes (linking contact ↔ deal). Note: connect-board writes via all_monday_api may fail silently — see Error handling reference.

Cross-skill handoffs

• From workspace-builder / daily-briefing: suggested as a way to keep deals fresh after a meeting-heavy day.
• To data-cleanup: if >30% of attendees couldn't be matched, suggest running data-cleanup to clean contact data.
• To daily-briefing: the synced updates feed tomorrow's brief.

---

Step 0: Connector check + NoteTaker availability

Goal: Confirm the monday connector works and detect whether NoteTaker is available. Also extract internal_domain for use in Steps 4–5.

1. Try mcp__monday__get_user_context. On error → standard install prompt, stop.
2. Extract internal_domain from the user's email (e.g. "kevin@monday.com" → "monday.com"). Store for Steps 4–5. If email is absent from the response, internal_domain = "monday.com" (safe default).
3. Try mcp__monday__get_notetaker_meetings({ limit: 1 }).
4. If NoteTaker returns results → set notetaker_available: true. Continue to Step 1.
5. If NoteTaker returns auth error, permission error, or empty-by-permission → set notetaker_available: false. Do not stop. The skill will collect meeting content from the user directly in Step 3.

PAUSE: do not write any create_* call before Step 6.

---

Step 1: Detect mode

• Default — bundle all writes (recaps + stage edits + last-touch updates + auto-contacts) into a single batched confirm before executing.
• Silent — skip the batched confirm; β auto-contact still requires explicit opt-in.
• Proactive — ask once: "Proactive mode will: (a) write recaps to matched deals, (b) auto-create missing contacts, (c) apply stage/last-touch edits when transcripts are unambiguous, (d) notify deal owners on commitments due this week. Amount columns stay untouched. Approve all / review each batch / deny stage edits only?". Cache the choice.

If undeclared, Default.

Hard safety rail regardless of mode: no deletes, no amount-column writes, no cross-workspace moves. Stage transitions are bundled into a batched plan — never written one-off, never silently.

---

Step 2: Resolve time window

1. If argument carries a phrase, parse: "today" / "yesterday" / "last N days" / "since YYYY-MM-DD".
2. Default = last 24h.
3. If no argument was provided AND 0 meetings are found (or on the very first run with no prior session context), print: "Using last 24h (no window specified). Add 'yesterday', 'last 7 days', or 'since YYYY-MM-DD' to change the range."
4. If parsed window > 14 days: confirm via AskUserQuestion (large windows produce noisy recaps).

---

Step 3: Pull meetings

3a — NoteTaker path (notetaker_available: true)

mcp__monday__get_notetaker_meetings({ since: <ISO>, until: <ISO>, limit: 50 }). Cache transcripts + attendees + meeting metadata.

Edge cases:

• 0 meetings in window: don't stop — fall through to Step 3b to let the user provide meeting content manually. Print: "No NoteTaker meetings found in <window>. You can paste meeting notes or describe what happened instead."
• >50 meetings: paginate up to 200, then cap and surface sampling note.
• Tool 429: backoff 3x; on third failure fall through to Step 3b.

3b — Manual input path (notetaker_available: false, OR NoteTaker returned 0 meetings)

Ask the user once via AskUserQuestion:

"No NoteTaker recordings available. How would you like to provide meeting notes?"

• (a) Paste the transcript or notes here — I'll read them from your next message
• (b) Point me to a file — provide the path and I'll read it
• (c) Describe what happened — tell me in free text (attendees, key points, commitments, next steps)
• (d) Skip — nothing to log right now

If (d) → stop with: "No meetings logged. Run this skill again when you have notes to add."

For (a) / (b) / (c): treat the user-provided content as a single meeting entry. Structure it as:

• title: first line of the notes, or ask if blank
• date: today (or extract from notes if mentioned)
• attendees: extract from notes, or ask once
• transcript_text: the full provided content

Cache as one entry in meetings[] with source: "manual" and continue to Step 4. The matching and recap flow works identically — the transcript is just from the user's input instead of NoteTaker.

---

Step 4: Resolve Deals + Contacts boards (Gather)

Resolve the boards directly — this skill is self-contained:

1. Deals board: get_user_context → scan favorites + relevantBoards for names matching deals|opportunities|pipeline|sales. One candidate → use it; multiple → AskUserQuestion; zero → list_workspaces → search("deal", BOARD). Still zero → degrade per step 4 below.
2. Contacts board: same pass for contacts|people|leads (only if β auto-contact may run). If Contacts board is missing and mode = Proactive and β was approved, surface a chat-level message before writing the sync doc: "Auto-contact was approved but I couldn't find a Contacts board. Contacts not created. Run /monday-crm:workspace-builder to set one up." Then continue with α recap only.
3. get_board_info on each. Resolve columns by type:
   • On Deals: email (if present), text company-name, people owner, status stage, numbers value, date last-touch.
   • On Contacts: email, text (or company-link board_relation), phone, people owner.
4. If Deals board missing entirely: degrade — skip α, write only the sync summary doc with "no Deals board found" header. Suggest /monday-crm:workspace-builder.

---

Step 5: Match meetings → deals (Synthesize)

Goal: For each meeting, find the single best matching deal — or flag it as unmatched.

internal_domain was extracted in Step 0. Use it throughout this step in place of hard-coded @monday.com checks.

For each meeting:

1. Extract signals: attendee emails (excluding @<internal_domain> addresses), company names from email domains, meeting title keywords.
2. Pull active deals (Pass A from daily-briefing — non-Won/Lost), limit: 500.
3. Score candidates:
   • +5 per attendee email matching a deal's email column.
   • +3 per attendee email domain matching a deal's company-name column (case-insensitive substring).
   • +2 if meeting title contains a deal name (≥3 contiguous chars match).
   • +1 if deal owner is a meeting attendee — only when ALL non-internal attendees are from @<internal_domain> (i.e. purely internal meeting). This signal is intentionally narrow: a deal owner being on an external sales call is expected and not a reliable match signal; it only carries weight for internal deal-review meetings where ownership is the strongest available signal.
4. Pick the top score if it's ≥5 AND beats the second-best by ≥3. Otherwise flag as ambiguous (multiple candidates) or unmatched (no candidate ≥5).
5. Multi-deal meetings: if top two candidates are within 2 points and both ≥5, flag as multi-match — write the recap to both deals with a header line "Meeting also discussed: ".

Edge cases:

• Internal-only meetings (all attendees @<internal_domain>) and no deal-name title match: skip (likely standup), log to summary doc as "skipped: internal-only".
• Personal-meetings filter: skip meetings tagged 1:1 (no transcript content of CRM relevance).

---

Step 6: Synthesize the recap (Synthesize)

For each matched meeting, compose a structured recap from the transcript:

• Key points (3–5 bullets, factual — what was discussed).
• Commitments made (who → what → by when, parsed from "I'll send you...", "we'll follow up by...", "next week we...").
• Open questions (statements ending with ? or signaled by "we need to figure out", "TBD").
• Stage transition signal: if transcript contains procurement / legal / contract / close phrases, propose a stage change. Mark each as confidence: high|low based on (a) explicit phrase ("we agreed to ", "signed today") = high, (b) inferred ("seems like they're ready for legal review") = low. Only high transitions enter the batched plan; low ones surface as text recommendations.

Deal-side metadata lookup: pull current stage + value + close date for context (single get_board_items_page filtered to the matched item ID).

---

Step 7: Publish α — per-deal recap (Publish)

For each matched meeting + deal pair:

1. Idempotency check: Use all_monday_api with the following query pattern to scan existing updates on the deal item:

   { items_by_id(ids: [<dealId>]) { updates(limit: 50) { body } } }
   

   Walk the body of each returned update and check for both <!-- claude-skill-id: meeting-to-deal --> and <!-- meeting-id: <id> -->. If a match is found → use all_monday_api to update that existing update's body; do not post a duplicate.

   Note: mcp__monday__search cannot search update bodies — always use all_monday_api items_by_id for this check.

2. create_update({ itemId: <dealId>, body: <recap markdown> }).

3. Update body shape:

# Meeting recap — <meeting title>
*<meeting date> · <duration>m · <attendee count> attendees*

## Key points
- ...

## Commitments
- <person> → <commitment> (due <date or "open">)

## Open questions
- ...

## Stage signal
<"applied: <old> → <new>" if the batched plan included it> OR <text recommendation if low-confidence> OR "no stage change">

---
<!-- claude-skill-id: meeting-to-deal -->
<!-- meeting-id: <noteTakerMeetingId> -->
Generated by Claude · <ISO timestamp> · run `/monday-crm:meeting-to-deal` to refresh.

Batched confirm (Default mode)

Before any write, print a single plan: "I'll write: recaps · stage edits (high-confidence only) · last-touch updates · new contacts. meetings unmatched (will list separately). low-confidence stage signals will surface as text only. Proceed? (yes / show plan in detail / no)."

If user picks "show plan in detail":

1. List all stage transitions by name: <deal name> · <old stage> → <new stage> for every planned stage edit.
2. Show one example recap (the first deal).
3. Then confirm: "Apply all? (yes / skip stage edits / no)".

Never per-deal confirms beyond that.

Auto-contact cap warning: If unmatched-attendee count > 25, add to the confirm message: " new contacts detected (> 25 cap). I'll create 25 and add the rest to a review-list doc. Proceed?" (e.g. "31 new contacts detected (> 25 cap). I'll create 25 and add the rest to a review-list doc. Proceed?")

---

Step 8: β — auto-create missing contacts (opt-in)

Goal: Don't make the user manually add every meeting attendee to Contacts.

Only runs if mode = Proactive (option a) OR Default-mode user opts in at Step 7 confirm (added prompt: "+ create N missing contacts on Contacts board?").

For each unique attendee email not found on the Contacts board:

1. create_item({ boardId: <contactsBoard>, itemName: "<First Last>", groupId: "new" }).
2. change_item_column_values to populate email, company (from email domain), and Source = Claude on the board's source/status column. If no source/status column exists, prompt the user once: "Add a Source status column on Contacts so we can flag plugin-created records? (yes / skip)". If skipped, contacts are still created but source isn't tracked.
3. Link the new contact to the matched deal via the board_relation column using all_monday_api. If this call fails (connect-board writes can fail silently), log the contact as "contact created, deal link pending — set manually" in the sync doc rather than stopping. Do not retry silently.

Cap at 25 contacts per run. Above that, batch into a review-list doc rather than creating silently. (Flag this at Step 7 confirm time — see auto-contact cap warning above.)

---

Step 9: Proactive next-step nudges (opt-in)

Only runs if mode = Proactive AND session-level approval was given.

For commitments parsed in Step 6 with a parseable due-date within next 7 days:

• create_notification({ userId: <deal owner>, text: "Commitment due <date> on <deal>: <commitment>" }).
• If create_notification requires an itemId: use the sync doc's item ID (from Step 10's create_doc return value). Do not pin the notification to an arbitrary deal item when the doc covers cross-deal commitments.

Hard rail: never write commitments to amount columns. Notifications only for proactive nudges (stage edits already handled in Step 7's batched plan, not here).

If >20 nudges queued, batch into one This week's commitments doc (with <!-- claude-skill-id: meeting-to-deal --> in body) and notify once with a link to the doc.

---

Step 10: Sync summary doc + close the loop

create_doc (idempotent — same-day update-in-place):

# Meeting sync — <Mon DD>

## Matched (<N>)
- <meeting title> → <deal name> (<deal url>)

## Multi-match (<N>)
- <meeting title> → <deal A> + <deal B>

## Ambiguous (<N>)
- <meeting title> — candidates: <deal A>, <deal B> — review and re-run with explicit deal ID

## Unmatched (<N>)
- <meeting title> — attendees: <emails> — no matching deal. Suggest creating one.

## Skipped (<N>)
- <meeting title> — internal-only, no CRM relevance

## Contacts created (<N>)
- <name> (<email>) → linked to <deal>

## Contacts pending manual link (<N>)
- <name> (<email>) — contact created, deal link pending — set manually

---
Generated by Claude · <ISO timestamp>

One-line chat summary: Synced <N> meetings to <M> deals. <K> unmatched, <L> contacts created. Doc: <url>.

---

Shared patterns

• No prefix on titles. Findability via Generated by Claude footer + <!-- claude-skill-id: meeting-to-deal --> comment in update/doc bodies; Source = Claude on contact items.
• Embedded meeting-id comment + skill-id comment for idempotency — search-and-update, never duplicate.
• Safety rail. No deletes, no amount-column writes, no cross-workspace moves. Stage edits go through the batched plan; low-confidence transitions stay as text recommendations only.
• Type-based column resolution for non-English boards.
• Batched confirms (Step 7) — never per-meeting prompts.
• Internal-only meeting filter — skips standups, 1:1s, retro meetings.
• internal_domain from Step 0 used throughout for internal/external attendee classification — not hard-coded to @monday.com.
• Skill-id comment uses legacy name meeting-to-opportunity — kept for backward compatibility with existing updates/docs carrying <!-- claude-skill-id: meeting-to-opportunity -->. The skill's invocation name is /monday-crm:meeting-to-deal.

---

Output template

Per-deal recap (create_update body)

# Meeting recap — Quarterly Review — Acme Corp
*Jun 14, 2026 · 45m · 4 attendees*

## Key points
- Discussed Q3 expansion scope — 3 additional seats requested
- Budget approved internally at Acme, pending procurement sign-off
- Timeline: targeting signed contract by end of June

## Commitments
- Sarah (Acme) → send updated SOW (due Jun 18)
- Kevin (us) → schedule legal review call (due Jun 20)

## Open questions
- Will Acme need SSO configured before go-live?
- Multi-year discount structure not yet agreed

## Stage signal
Applied: Discovery → Proposal (high confidence — "we agreed to move to procurement review")

---
<!-- claude-skill-id: meeting-to-opportunity -->
<!-- meeting-id: ntm_abc123def456 -->
Generated by Claude · 2026-06-15T09:00:00Z · run `/monday-crm:meeting-to-deal` to refresh.

Sync summary doc

# Meeting sync — Jun 15

## Matched (2)
- Quarterly Review — Acme Corp → Acme Corp — Q3 expansion (https://monday.com/boards/...)
- Demo follow-up — Beta Industries → Beta Industries — pilot (https://monday.com/boards/...)

## Unmatched (1)
- Internal team standup — attendees: all @monday.com — skipped (internal-only)

## Contacts created (1)
- Mike Chen (mike@acme.com) → linked to Acme Corp — Q3 expansion

---
Generated by Claude · 2026-06-15T09:00:00Z

Chat one-liner:

Synced 2 meetings to 2 deals. 1 skipped (internal). 1 contact created. Doc: [url].

---

Error handling reference

---

Completion criteria

• Step 0 (connector + NoteTaker) ran and passed. internal_domain extracted.
• Each matched deal got at most one new (or one updated) recap update (verified via all_monday_api items_by_id update-body scan).
• Every recap body carries <!-- claude-skill-id: meeting-to-deal --> + <!-- meeting-id: <id> --> + Generated by Claude footer.
• Sync summary doc exists (or printed to chat on failure).
• Safety rail held: no deletes, no amount-column writes, no cross-workspace moves.
• Stage edits (if any) ran inside the Step 7 batched plan; all stage transitions listed by name in "show plan in detail"; low-confidence transitions stayed as text only.
• Contacts created (if β ran) carry Source = Claude; failed deal links logged in sync doc as "pending — set manually".
• Auto-contact cap warning shown at confirm time if unmatched attendees > 25.
• Notifications (if Proactive) pinned to sync doc item ID, not arbitrary deal items.