daily-briefing

Morning Briefing

Produces a daily prioritized briefing over the user's monday CRM data and publishes it as a monday artifact — so the brief lives inside monday, not just in the chat.

Works for any monday CRM setup: pipeline users get deal buckets and pipeline totals; contacts/accounts users get relationship health and follow-up queues; lead managers get inbound lead status. If the user has no monday CRM board at all, the skill runs in standalone mode — ask for context, produce a useful brief from what they tell you.

When monday CRM features are in use, the brief automatically gets richer:

monday CRM feature	What it adds	MCP access
Notetaker	Recent meeting notes matched to open deals or contacts	get_notetaker_meetings
Activity log	Logged calls, emails, and meetings on deals/contacts (last 48h)	get_board_activity + get_updates
Tasks / follow-ups	Items with a date column = today across deals or contacts board	get_board_items_page with date filter
Calendar sync (Outlook/Google)	Today's CRM-linked meetings	Not yet in MCP — skipped; wired when available (see workset status.md)
Email sync (Gmail/Outlook)	Tracked email threads on deals/contacts	Not yet in MCP — skipped; logged emails accessible today via get_updates on contacts/deals

monday CRM's calendar and email sync features exist in the product but are not fully exposed via the MCP connector yet. The skill attempts to reach them via all_monday_api raw GraphQL. If the data isn't there, it notes the gap in the brief and skips the section — it does not error out.

Flow: Connector check → Mode → Connector discovery → Board detection → Board type → Data pull → Synthesize → Publish → Share (opt-in) → Proactive (opt-in).

Input

• Optional: board name or ID via argument.
• Optional: mode override (Default / Silent / Proactive) via project instructions or one-shot prompt.

Output

• α (default): A monday update on a pinned "My Morning" item OR a new doc in the user's personal workspace, titled Morning Briefing — <weekday>, <Mon DD>. Body carries the Generated by Claude footer + an HTML comment <!-- claude-skill-id: daily-briefing --> for idempotency. Markdown brief printed in chat as a mirror.
• β (Share mode, opt-in): Notification to manager + update on team-rollup board.
• Proactive extension (opt-in): Per-item create_update on stalled records; create_notification for commitments due today.
• Standalone mode (no board): Brief printed in chat only — no monday artifact (nothing to publish to).

Knowledge

• Board type detection + synthesis buckets per type (§ Steps 2b and 6).
• Shared artifact conventions (§ Shared patterns).

Tools (MCP)

• get_user_context — resolve user + favorite boards.
• list_workspaces — enumerate workspaces when search is ambiguous.
• search — find CRM boards by broad search terms.
• get_board_info — board schema + column IDs + labels.
• get_column_type_info — resolve columns by type when names aren't English.
• get_board_items_page — pull active records, recently closed, recent activity.
• board_insights — aggregate fallback when >200 items.
• create_update — publish α to pinned item; proactive nudges on stalled records.
• create_doc — publish α as a doc when no pinned item exists.
• create_notification — β share; proactive alert for commitments due today.
• list_users_and_teams — resolve recipient for β share.
• all_monday_api — escape hatch for doc updates / metadata lookups not in wrapped tools.

Cross-skill handoffs

• To forecast-dashboard: If briefing surfaces forecast hygiene issues, suggest /monday-crm:forecast-dashboard.
• To board-diagnosis: If >30% of records are unowned, or core fields missing on most items, suggest /monday-crm:board-diagnosis.

---

Step 0: Connector check

Goal: Fail fast if the user has no monday MCP connection.

1. Try mcp__monday__get_user_context.
2. If the tool is missing, returns auth-style error, or the user has no accessible workspace:
   • Print: "I don't see the monday connector active on this session. Install it from https://monday.com/mcp, then run this skill again."
   • Stop. Do not proceed.
3. If connector works, cache user.id and user.name.

PAUSE: do not write any create_* call before Step 6, and never before Step 0 passes.

---

Step 1: Detect mode and output config

Goal: Know write behavior and output preferences before doing anything else.

Write mode

• Default — confirm every create_* or change_* call in chat before executing.
• Silent — skip confirmations, stay on α scope.
• Proactive — ask once at the top of the session: "Proactive mode will post per-item updates on stalled records and notify you on commitments due today. (a) Approve all, (b) Review each, (c) Deny all?" Store the choice for the session.

If no mode declared, assume Default. Trigger: morning briefing silent / morning briefing proactive / mode: proactive in project instructions.

Hard safety rail regardless of mode: no deletes, no amount-column writes, no cross-workspace moves. Stage, owner, last-touch, and source column edits are allowed when bundled in a batched-confirm plan (Default) or session-level approval (Proactive).

Output config

Read from project instructions or per-run prompt. All fields are optional — defaults apply when absent.

Verbosity (default: full)

• full — all sections, per-deal detail, pipeline totals
• quick — headline numbers + top 3 action items only (fits in ~20 lines)
• tldr — single paragraph: one priority, one number, one action

Trigger phrases: quick brief / quick morning briefing → quick; tldr / tldr my day → tldr.

Sections (default: all on)

Set per-run with natural language ("skip yesterday's movements", "only show what needs attention") or in project instructions:

output:
  sections:
    pipeline_totals: true      # "Pipeline at a glance"
    yesterday_movements: true  # won/lost since last run
    meeting_today: true        # deals with a meeting today (requires notetaker or calendar)
    tasks_today: true          # tasks/follow-ups due today
    needs_action: true         # slipping, stale, follow-up buckets
    on_track: true             # hot deals
    activity_summary: true     # logged calls/emails/notes count
    new_this_week: false       # new deals added (off by default — low urgency)
    data_hygiene: true         # missing fields, unowned items

Any section with no items is automatically omitted regardless of config (don't print empty headers).

Style (default: bullets)

• bullets — markdown bullets and headers (default, renders well in monday docs and chat)
• tables — key data in markdown tables (better for scanning many deals)
• prose — short paragraph per section (better for sharing / pasting into email)

Trigger: morning briefing table format → tables; morning briefing prose → prose.

Destination (default: monday+chat)

• monday+chat — publish artifact to monday AND mirror in chat
• monday — publish to monday only, no chat mirror
• chat — chat only, no monday artifact (useful for quick checks without cluttering monday)
• doc — force a new monday doc even if a "My Morning" pinned item exists

Trigger: morning briefing chat only → chat; morning briefing doc → doc.

Config precedence: per-run prompt overrides project instructions, which overrides defaults. A per-run phrase like "quick brief" applies to this run only — it does not change stored config.

---

Step 1b: Discover monday CRM enrichment sources (MUST RUN — not optional)

Goal: Know which monday-native data sources are live before pulling CRM data. Run in parallel with Step 2.

Try each source silently — don't stop if one fails:

1. Notetaker meetings — call get_notetaker_meetings. If it returns results, cache them. If it errors or returns empty, mark notetaker: false.

2. Calendar sync — not yet available via monday MCP. Mark calendar: false. Skip. Will be wired when monday exposes calendar_events in the MCP connector (tracked in workset status.md).

3. Email sync — not yet available via monday MCP. Mark email: false. Skip. Will be wired when monday exposes crm_emails in the MCP connector (tracked in workset status.md). Note: logged emails stored as item updates on contacts/deals are accessible today via get_updates — those are handled in Step 5d.

4. Tasks board — search get_user_context → relevantBoards for a board named tasks|to-do|actions|follow-up. If found, mark tasks_board: <boardId>. If not found, skip — tasks will be inferred from deal/contact items with date columns = today.

None of these failing should block the rest of the skill. Calendar and email sync are product features not yet fully exposed in the MCP — the attempt is best-effort. Notetaker and tasks are available today.

---

Step 2: Resolve board(s)

Goal: Find the user's CRM board(s), or confirm there are none and fall back to standalone.

Broad search — cast the net wide:

1. If the user passed a board via argument:

   • Numeric → use as boardId directly.
   • String → mcp__monday__search(searchType: "BOARD", searchTerm: <arg>). One match → use it. Multiple → AskUserQuestion.

2. If no argument, scan get_user_context → relevantBoards + favorites for names matching any of:

   • Pipeline signals: deal|deals|opportunity|opportunities|pipeline|sales
   • Contact/account signals: contact|contacts|account|accounts|client|clients|customer|customers
   • Lead signals: lead|leads|prospect|prospects|inbound
   • Relationship signals: partner|partners|renewal|renewals|relationship

3. If no matches in context, run search with searchTerm: "crm" then "contacts" then "deals". Collect all candidates.

4. If still zero matches → Standalone fallback (see Step 2c below). Do not stop the skill.

5. If multiple candidates of the same type (e.g., two pipeline boards):

   • Auto-select the one with the most recent updated_at on its items — this is the board the user is actively working in. Note the selection in the brief's Data hygiene section: "Briefing on [Board Name]. To brief a different board, pass its name as an argument."
   • If updated_at is the same day or indeterminate: AskUserQuestion with board names + "all of them" option.

6. If candidates are of different types (e.g., one pipeline + one contacts board): run both paths and combine into one brief — no disambiguation question needed.

If the user explicitly picks "all": run Steps 3–7 per board and publish one combined α artifact.

---

Step 2b: Classify board type (MUST RUN — execute immediately after Step 2)

Goal: Tailor the data pull and synthesis to what the user actually has.

After resolving the board, call get_board_info and classify as one of:

Type	Signals
pipeline	status column with is_done: true labels (Won) + numbers column with $ unit (value)
leads	status column without is_done: true (no Won state) + lead-sounding labels (e.g. "New", "Contacted", "Qualified", "Disqualified")
contacts	Majority text / email / phone columns; few or no numbers columns; no Won/Lost pattern
accounts	Mix of text, numbers (revenue/ARR), status columns; company-centric item names
general	CRM board that doesn't fit above — treat like accounts

Cache the detected boardType for Steps 3–6.

If a user has both a pipeline board and a contacts/accounts board, run both paths and combine into one brief under separate sections.

---

Step 2c: Standalone fallback (no board) (MUST RUN when Step 2 finds zero boards)

Goal: Still produce a useful briefing even without a monday board connected.

Run this path if Step 2 found zero boards OR the user says "I don't have a CRM set up yet."

Ask the user two questions (one AskUserQuestion call, both together):

1. "What meetings or calls do you have today?"
2. "What deals, accounts, or relationships need your attention this week?"

Work with whatever they provide. Produce a briefing using the Standalone template (§ Step 6c). Skip Steps 3–5 and Steps 7–9 (nothing to publish). Print the brief in chat only.

---

Step 3: Read board schema

Goal: Resolve column IDs by type — global teams have non-English column titles.

1. get_board_info(boardId). Cache columns.
2. Classify columns by type:

Pipeline columns:

Concept	Type	Title hint	Fallback
Stage	status	stage, status	first status with is_done labels
Owner	people	owner, rep, person	first people column
Value	numbers	value, amount, revenue, size	first numbers with $ unit
Close probability	numbers	probability, %	first numbers with % unit
Expected close date	date	close, expected, target	first date-type column matching
Last interaction	date	interaction, activity, touch, contacted	second date-type column
Forecast value	formula	forecast, weighted	optional; computed from value × prob if absent

Contacts/accounts columns (additional):

Concept	Type	Title hint
Company	text	company, account, organization
Email	email	email
Phone	phone	phone
Last contacted	date	last contacted, last touch, last interaction
Next follow-up	date	follow-up, next contact, next step
Owner	people	owner, rep, assigned
Status	status	status, stage, health
Tags	tags	tags, segment, type

3. For ambiguous columns call get_column_type_info.
4. If a column type is missing entirely, continue without it and note in the brief under "Data hygiene".

---

Step 4: Identify terminal labels (pipeline only)

Goal: Distinguish closed records from active — without hardcoding English.

Skip this step for contacts, accounts, general board types.

From the stage column's settings.labels:

• Won = labels with is_done: true.
• Lost = labels where text (lowercased) contains lost|closed lost|cerrado perdido|verloren|perdu.
• If no is_done: true label exists, ask the user which label means "won".

Cache label index values.

---

Step 5: Pull data

Goal: Get the items needed to build the brief. Adapt the queries to the board type.

5a — Pipeline boards

Three passes:

Pass A — Active: get_board_items_page with stage not_any_of [won_idx, lost_idx], limit 200. Pass B — Recent closures: stage any_of [won_idx, lost_idx], ordered by updated_at desc, limit 50; in-memory filter to items with updated_at within last 24h. Pass C — Recent activity: reuse Pass A; sort by updated_at desc; last 24h.

Volume edges:

• Pagination: pull until cursor is null or 2,000 items. Flag sampling if capped.

• 10K items: stop at 2,000, surface sampling note.

• board_insights fallback if >200 active items for aggregate-only view.

5b — Contacts / accounts boards

Two passes:

Pass A — All active contacts/accounts: get_board_items_page, no status filter (include all non-archived), limit 200. Include: owner, last-contacted date, next-follow-up date, status, company.

Pass B — Recent activity: sort Pass A by updated_at desc; last 48h.

5c — Leads boards

Two passes:

Pass A — Open leads: get_board_items_page, filter out terminal labels (e.g., "Disqualified", "Converted"), limit 200. Pass B — Recent additions: sort by created_at desc, last 48h.

5d — Activities / updates (all board types)

This is the primary way to get logged emails, calls, and activities from monday CRM today. When users log a call, send an email (via monday email sync), or add a note on a contact or deal, it's stored as an update thread on that item. get_updates surfaces all of it.

For each resolved board:

1. get_board_activity(boardId, from: <48h ago>) — structural changes: stage moves, owner assignments, column edits. Tells you what changed on which items.
2. get_updates(boardId, limit: 100) — update threads: logged calls, tracked emails, meeting notes, manual notes added by reps. This is the activity log.

If either call returns a permission error (403 or similar): note in the brief's Data hygiene section — "Activity log unavailable — the MCP connector lacks read access to this board's activity feed. A board admin can grant it under Board Settings → Permissions → View board." Do not stop or retry; proceed without activity data.

Classify each update body by content signals:

• Contains "called", "spoke", "phone call", "left voicemail" → logged call
• Contains "email", "sent", "replied", "re:", "subject:" → logged email (may be auto-logged by monday email sync, or manually added)
• Contains "meeting", "met with", "visited", "demo", "call recording" → logged meeting
• monday Notetaker auto-posts meeting summaries as updates on the linked item — detect by checking for notetaker-style formatting (bullet summary, attendee list)
• Anything else → note

For the contacts board specifically: this is where email/activity history lives. Pull updates for all contacts that appear in your deals or that you own — not the full board. Filter by updated_at within last 48h to stay relevant.

Cache per item: {itemId, activityType, summary (first 100 chars), timestamp, author}[]. Group by deal/contact for synthesis.

5e — Notetaker meetings

If notetaker: true from Step 1b: the cached get_notetaker_meetings results contain recent meeting recordings/notes with associated items.

For each notetaker entry:

• Match to a deal or contact in the boards via item name or ID.
• Extract: meeting title, date, attendees (if present), matched item.
• Cache: matched_meeting_notes[] for synthesis.

Unmatched meetings (no CRM item linked): surface separately in the brief under "Meetings not linked to a deal" — they're an upsell opportunity.

5f — Tasks due today

If tasks_board was found in Step 1b: call get_board_items_page on that board filtered by due date = today.

If no tasks board: scan Pass A results (deals/contacts) for items where any date column = today that isn't the close date (treat as a follow-up or task date).

Cache as tasks_today[] with item name, assigned owner, linked board.

5g — Calendar events (future)

Not available via monday MCP today. Skip. When monday exposes calendar_events on the connector: pull today's events, cross-reference with open deals/contacts by crm_item link or attendee domain match, cache as crm_meetings_today[] for synthesis Step 6a. See workset status.md for tracking.

5h — Email sync (future)

Not available via monday MCP today. Skip. Logged emails stored as item updates on contacts/deals are covered by Step 5d (get_updates). When monday exposes crm_emails on the connector: pull unread emails + awaiting_reply threads, cross-reference with CRM items, surface in the brief under "Email priorities". See workset status.md for tracking.

Rate limit handling (all types)

429 → exponential backoff 3x (1s, 2s, 4s). On third failure: "monday is rate-limiting right now. Retry in 60 seconds, or narrow to fewer boards." Stop.

---

Step 6: Synthesize

Goal: Rank items by what needs action today.

6a — Pipeline synthesis

Bucket	Criteria
Meeting today	deal has a crm_meetings_today entry — always surface, regardless of other signals
Task due today	deal appears in tasks_today — surface with task description
Commit-today	close_date = today AND stage ≠ Won/Lost
Slipping	close_date in past AND stage ≠ Won/Lost
Stale	last_interaction > 14d ago AND stage in non-closed
Follow-up	last_interaction > 7d ago AND probability ≥ 50%
Hot	probability ≥ 70% AND last_interaction ≤ 7d AND close_date ≤ 30d
New	created within last 7d AND stage = earliest-index label

Priority order: Meeting today > Task due today > Commit-today > Slipping > Stale > Follow-up > Hot > New.

A deal can appear in multiple buckets — display in the highest-priority one only, but annotate lower-priority signals inline (e.g., a Stale deal that also has a meeting today → show under "Meeting today" with a 🔴 stale flag).

Totals: active pipeline Σ value; forecast Σ (value × prob/100); closing this week count + Σ value; yesterday wins Σ value; yesterday losses count.

Section header counts must match body item counts exactly. Count the items in each bucket after applying verbosity limits, then write ### Slipping (N) where N is the actual number of lines that follow. Never use a pre-limit count in the header.

Activity summary: count of logged calls, emails, meetings, and notes in the last 24h across all active deals. Surface as a one-liner in the brief (e.g., "5 activities logged yesterday — 2 calls, 2 emails, 1 note").

6b — Contacts / accounts synthesis

Bucket	Criteria
Overdue follow-up	next_follow_up date in past
Going cold	last_contacted > 21d ago AND no upcoming follow-up
Follow-up today	next_follow_up = today
Recently active	updated_at within last 24h
New this week	created_at within last 7d

Priority order: Overdue > Follow-up today > Going cold > Recently active > New.

No pipeline totals for contacts boards. Instead surface: total contacts, unowned contacts (%), contacts with no follow-up date set.

6c — Leads synthesis

Bucket	Criteria
Hot leads	status in early/qualified stages AND created within last 7d
Unassigned	owner column empty
Stale	created > 14d ago AND no status change in 7d
Recent additions	created_at within last 48h

Surface: total open leads, unassigned count, conversion rate (if terminal labels detected).

6d — Standalone (no board)

Prioritize from user-provided input:

1. Meetings today with a deal/account at stake
2. Items the user flagged as urgent
3. Deals/relationships the user mentioned as needing action

---

Step 7: Publish α

Goal: Put the brief inside monday. Skip for standalone mode (nothing to publish to).

Target selection

Respect the destination config from Step 1:

• chat → skip this step entirely; print to chat only.
• doc → skip "My Morning" search; go directly to create_doc.
• monday or monday+chat (default) → follow steps below.

1. Call search(searchType: "ITEM", searchTerm: "My Morning"). Filter results to boards the user owns or is a member of (to avoid false matches on shared workspaces). If exactly one match → create_update on that item. If multiple → take the one in the user's personal workspace (from get_user_context → personal_workspace_id). If none → proceed to step 2.
2. Else: create_doc in personal workspace with title Morning Briefing — <weekday>, <Mon DD>, <YYYY>.
3. Idempotency: before creating, search by exact title. If found, update in place. If search returns ambiguous results, add a suffix (— v2). Never duplicate.

Apply verbosity and section config before writing

Before assembling the brief body, apply the config from Step 1:

• tldr verbosity: replace the full template with: one sentence on top priority, one pipeline number, one suggested action. Skip all sections.
• quick verbosity: include only pipeline_totals + needs_action (top 2 items max per bucket, 1-line each). Skip yesterday_movements, on_track, new_this_week, activity_summary, data_hygiene. Hard cap: 25 lines total including headers and footer. (~20 is achievable when fewer buckets are populated; top 2 per bucket keeps within cap even when all three action buckets are full.)
• full verbosity (default): all sections per config.
• Section overrides: suppress any section the user turned off. Never print a section header with no content.
• tables style: render deal lists as markdown tables (| Deal | Stage | Value | Alert |) instead of bullets.
• prose style: write each section as 1–2 sentences, no headers. Use plain paragraph breaks.

If a section is configured on but has zero items, omit it silently — don't print "(none)".

Brief templates

Pipeline template:

# CRM Morning Briefing — <weekday>, <Mon DD>

<!-- claude-skill-id: daily-briefing -->

## Pipeline at a glance
- Active pipeline: $<total>K across <N> deals
- Forecast: $<forecast>K (weighted)
- Closing this week: <N> deals, $<value>K

## Yesterday's movements
- Won: <deal> ($<value>K, <owner>)
- Lost: <deal> ($<value>K, <owner>)

## Needs you today

### Closing today (<N>)
- <deal> — <stage>, $<value>K, owner <name>, last touch <X days ago>

### Slipping (<N>)
- <deal> — <stage>, $<value>K, expected <date> (<X> days overdue), owner <name>

### Going cold (<N>)
- <deal> — <stage>, no touch in <X> days, owner <name>

### Follow-up (<N>)
- <deal> — <stage>, <probability>%, last touch <X days ago>

## On track
- <deal> — <stage>, <probability>%, closing in <X> days

## Data hygiene
- <N> deals missing owner · <N> missing close date · <N> missing amount

---
Generated by Claude · <ISO timestamp> · run `/monday-crm:daily-briefing` to refresh.

Contacts / accounts template:

# CRM Morning Briefing — <weekday>, <Mon DD>

<!-- claude-skill-id: daily-briefing -->

## Relationships at a glance
- <N> contacts/accounts total · <N> owned by you · <N> unowned

## Needs you today

### Overdue follow-ups (<N>)
- <name> (<company>) — follow-up was due <date>, owner <name>

### Follow up today (<N>)
- <name> (<company>) — scheduled for today, last touched <X days ago>

### Going cold (<N>)
- <name> (<company>) — no contact in <X> days, no follow-up scheduled

## Recently active
- <name> — updated <X hours ago>

## New this week
- <name> (<company>) — added <date>

## Data hygiene
- <N> contacts missing owner · <N> with no follow-up date set
- (Run `/monday-crm:board-diagnosis` for a full audit.)

---
Generated by Claude · <ISO timestamp> · run `/monday-crm:daily-briefing` to refresh.

Leads template:

# CRM Morning Briefing — <weekday>, <Mon DD>

<!-- claude-skill-id: daily-briefing -->

## Leads at a glance
- <N> open leads · <N> unassigned · <N> added this week

## Needs you today

### Unassigned leads (<N>)
- <lead name> — <stage>, added <date>

### Hot leads (<N>)
- <lead name> — <stage>, <X> days old, owner <name>

### Going stale (<N>)
- <lead name> — <stage>, no activity in <X> days

## Recent additions
- <lead name> — added <X hours ago>, source <source>

## Data hygiene
- <N> leads unassigned · <N> with no status change in 7+ days
- (Run `/monday-crm:board-diagnosis` for a full audit.)

---
Generated by Claude · <ISO timestamp> · run `/monday-crm:daily-briefing` to refresh.

Standalone template (chat-only, no artifact):

# Morning Brief — <weekday>, <Mon DD>

## #1 Priority
**[Most important thing from what you told me]**
[Why it matters]

## Today's meetings
- [Time / Company / What's at stake]

## Needs attention
- [Item 1 — why]
- [Item 2 — why]

## Suggested next actions
1. [Action] — [why now]
2. [Action] — [why now]
3. [Action] — [why now]

Confirm-before-write (Default mode)

Print before any create_*: "I'll publish this as <target>. OK to post? (yes / no / show first)." Silent and Proactive modes skip this.

Print to chat

• monday+chat (default): print the brief in chat as a mirror + artifact URL on last line: Posted: <url>.
• monday: print one-liner only: Posted: <url> — no full mirror.
• chat: print the full brief in chat. No artifact URL.
• doc: same as monday+chat.

Publish errors

• Perms failure: degrade create_update → create_doc → print-only with banner "Couldn't publish to monday (permission error). Brief below."

---

Step 8: Share β (opt-in)

Runs only if user explicitly asked ("share with my manager", "post to the team", or Proactive mode selected Share).

1. list_users_and_teams → resolve recipient. Default to declared manager if present; else AskUserQuestion with top-5 linked teammates.
2. create_notification({ userId, itemId: <artifact>, text: "Morning briefing for <Mon DD>" }).
3. Optional: create_update on a team-rollup board (resolve by search / AskUserQuestion — don't assume a fixed name). Same footer + skill-id-comment convention.

---

Step 9: Proactive extension (opt-in)

Runs only if mode = Proactive AND session decision was "approve all" or "review each".

For each item in Stale / Going cold bucket:

• create_update(itemId, "*Claude nudge:* No activity in <X> days. What's the next step?").

For each item in Commit-today / Overdue follow-up bucket:

• create_notification({ userId: <owner>, itemId, text: "Action due today on <name>" }).

"Review each" flow: print a one-line proposal per item before writing. yes → execute; no → skip; stop → halt.

Guard rails:

• If total actions > 20: batch into one review-list doc (Proactive actions — <date>) with a single prompt.
• Never post on items the user doesn't own unless they're board admin AND gave Proactive approval this session.
• Safety rail holds: no deletes, no amount-column writes, no cross-workspace moves.

---

Step 10: Close the loop

Print: Published Morning Briefing — <date>. <N> items scanned, <N> actions posted, <N> pending.

If any write failed mid-run: list written items in chat for manual cleanup. Do not attempt compensating deletes.

---

Shared patterns

• No title prefix. Discoverability via Generated by Claude · <timestamp> footer + <!-- claude-skill-id: daily-briefing --> HTML comment.
• Idempotency before create. Search same-day artifact first. Update, don't duplicate.
• Safety rail. No deletes, no amount-column writes, no cross-workspace moves. Stage/owner/last-touch/source edits OK in batched-confirm plan.
• Type-based column resolution, not name-based — global teams have non-English columns.
• Ask once per session when entering Proactive mode.
• Board type drives output — don't show pipeline totals to a contacts user; don't show follow-up queues to a pipeline user unless they also have a contacts board.

---

Error handling

Failure	Behavior
Connector not installed	Step 0 stops; print install link
No CRM board found	Standalone fallback (Step 2c) — don't stop
Board >10K items	Cap at 2K most-recent, flag sampling
MCP tool error (retryable)	Backoff 3x; skip that strand; surface partial brief
Same-day artifact exists	Update in place
User halts mid-session	List written items in chat; no compensating deletes
429 rate limit	Backoff 3x, then stop with retry message
Publish perms failure	Degrade to doc → print-only
Multiple CRM boards	AskUserQuestion with "all of them" option
Non-English columns	Resolve by type via get_column_type_info
Proactive writes >20	Batch into one review-list doc

---

Completion criteria

• Step 0 (connector check) ran and passed.
• Board type detected and synthesis path matched (pipeline / contacts / accounts / leads / standalone).
• If board found: exactly one α artifact created or updated (no duplicate — verified via skill-id HTML comment).
• If standalone: brief printed in chat only.
• Artifact body carries Generated by Claude footer + <!-- claude-skill-id: daily-briefing --> comment.
• Chat mirror includes artifact URL on last line (non-standalone only).
• Any failed writes listed in final summary.
• Safety rail held: no deletes, no amount-column writes, no cross-workspace moves.
• Stage/owner/last-touch edits (if any) ran inside batched-confirm plan.
• If Proactive mode: single session-level decision captured, not per-item.