Meeting Debrief

Capture meeting outcomes through guided Q&A and route them to the right places: action items to kanban, decisions to memory, and a structured debrief report to ~/Tandem/meetings/. Output follows ${CLAUDE_SKILL_DIR}/template.md.

For detailed action item routing procedures, see ${CLAUDE_SKILL_DIR}/references/action-item-routing.md. For detailed decision capture procedures, see ${CLAUDE_SKILL_DIR}/references/decision-capture.md.

---

Language

See METHODOLOGY.md language mirror rule. Reply in the user's language.

---

Step 1: Determine Debrief Mode

Smart routing based on user request:

If user mentions a transcript (pastes text, says "I have a transcript", references a recording or call notes):

• Redirect to meeting-notes skill: "Looks like you have a transcript -- try meeting-notes first, then run debrief to route action items."
• Stop here. Do not proceed with debrief.

If user says "debrief standup", "quick debrief", or references a routine meeting (standup, daily, 1:1, sync):

• Enter Quick Mode -- shortened flow collecting only action items and blockers
• Skip decisions, follow-ups, executive summary
• Proceed to Step 2 for meeting identification

If user says "debrief my [meeting]" or names a specific meeting:

• Enter Full Mode -- complete guided Q&A flow
• Proceed to Step 2 for meeting identification

If user is vague ("debrief"):

• Fetch calendar events (Step 2) and ask which recent meeting to debrief

---

Step 2: Identify the Meeting

Fetch recent calendar events (past 12 hours) to match the meeting being debriefed.

Parameters:

• singleEvents=true -- expand recurring events into individual instances
• timeMin = current time minus 12 hours (ISO 8601)
• timeMax = current time (ISO 8601)
• orderBy=startTime
• Fields requested: summary, start, end, attendees, description, status, htmlLink

Post-fetch filtering:

1. Remove events where status === "cancelled"
2. Find the user's own entry in each event's attendees[] array (match by email from ~/Tandem/CLAUDE.md Tools & Connectors section, or ~/Tandem/memory/preferences.md)
3. Remove events where user's responseStatus === "declined"
4. Separate all-day events from timed meetings -- all-day events are not debrief targets

Present recent meetings via AskUserQuestion:

AskUserQuestion: "Which meeting are you debriefing?"
Options: [
  "[HH:MM] -- [Meeting Title] ([attendees])",
  "[HH:MM] -- [Meeting Title] ([attendees])",
  "I'll describe the meeting manually"
]

Order meetings reverse-chronologically (most recent first).

If calendar is unavailable (no MCP connector, auth error, timeout):

• Jump to Step 2b (manual input fallback)

---

Step 2b: Manual Input Fallback

When no calendar connector is available, or user chooses "describe the meeting manually":

AskUserQuestion: "Tell me about the meeting -- who attended, what was the topic, and when was it?"
Options: [
  "I'll type the details now",
  "Let me paste the calendar invite"
]

Parse the user's response for:

• Attendee names (and companies/roles if mentioned)
• Meeting topic or title
• Time and date

Continue to Step 3 with whatever information was provided. Missing fields are OK -- the debrief will use what is available.

---

Step 3: Check for Existing Context

Before starting Q&A, silently gather relevant context. Do NOT ask permission for each source -- read quietly and compile.

3a: Prep Brief

Glob ~/Tandem/meetings/prep-{date}-{time}-*.md matching the identified meeting's date and approximate time. If found:

• Read the prep brief for context items to close the loop on
• These feed the "Prep Items Addressed" section in the report (Step 7)
• Store the prep filename for cross-linking

3b: Meeting-Notes Output

Glob ~/Tandem/documents/meeting-* and ~/Tandem/meetings/meeting-notes-* for files matching the same date and attendees. If found:

• Pre-fill debrief data from the meeting-notes output
• Present to user: "Found meeting notes from [time]. I'll pre-fill from those -- you can confirm and add anything missing."
• In full mode: pre-fill Q&A answers from extracted data (user confirms each)
• In quick mode: pre-fill action items and blockers (user confirms batch)

3c: Kanban Board

Read ~/Tandem/tasks/kanban.md for dedup reference. Store existing tasks for fuzzy matching in Step 5. If kanban.md does not exist, skip -- will create it when writing action items.

---

Step 4: Guided Q&A (Full Mode)

Collect meeting outcomes via AskUserQuestion. One question at a time, each building on previous answers. If pre-filled from meeting-notes (Step 3b), present the pre-filled answer and ask user to confirm or edit.

Q1: Decisions

AskUserQuestion: "What were the key decisions made in this meeting?"

• Free text response
• If user says "none" or skips, move on

Q2: Action Items

AskUserQuestion: "What action items came out of this meeting? Include who's responsible if known."

• Free text response
• Parse for: task description, owner name, implied deadline

Q3: Follow-ups

AskUserQuestion: "Any follow-ups needed? (emails to send, people to loop in, documents to share)"

• Free text response

Q4: Executive Summary

AskUserQuestion: "How would you summarize this meeting in 2-3 sentences? (overall outcome and sentiment)"

• Free text response
• This becomes the executive summary in the debrief report

---

Step 4b: Guided Q&A (Quick Mode)

For quick debriefs (standups, 1:1s, routine syncs), a single question captures everything needed:

AskUserQuestion: "What action items and blockers came out of this meeting?"

• Single question captures both action items and blockers
• Parse response for action items (tasks with owners) and blockers (impediments)

That is the ONLY question in quick mode. No decisions, no follow-ups, no executive summary.

---

Step 5: Action Item Processing

For detailed procedures, see ${CLAUDE_SKILL_DIR}/references/action-item-routing.md.

Overview:

1. Parse raw text into structured items: description, owner, deadline (suggest based on urgency: today, this week, next week)
2. Deduplicate against kanban.md (read in Step 3c) -- fuzzy match on key nouns, flag potential duplicates
3. Batch confirmation -- present ALL items in one list via AskUserQuestion:

AskUserQuestion: "Here are the action items I extracted. Review and confirm:"
Options: [
  "Approve all",
  "Let me edit the list",
  "Show me kanban matches first"
]

If "Show me kanban matches first": display fuzzy matches with per-item options (create new / update existing TASK-XXX / skip). 4. Write approved items to ~/Tandem/tasks/kanban.md following kanban-format.md format (increment Last Task ID, add to "To Do" column, update existing if chosen). If kanban.md does not exist, create it from kanban-format.md format. 5. Offer follow-up email after action items are processed:

AskUserQuestion: "Send a follow-up email to attendees with action items?"
Options: [
  "Yes, draft follow-up email",
  "No, skip email"
]

If yes: hand off to email-draft skill with meeting context and action item list.

---

Step 6: Decision Capture

For detailed procedures, see ${CLAUDE_SKILL_DIR}/references/decision-capture.md.

Overview:

1. Parse decisions from Q&A Step 4 Q1 into structured entries: title, rationale, date, attendees
2. Format each with meeting context (date, meeting title, attendees, rationale)
3. Append to ~/Tandem/memory/decisions/index.md (create file/directory if missing, newest first)
4. Offer people memory update for decisions that reference specific people:

AskUserQuestion: "Update [name]'s memory with this meeting context? Last met [date], discussed [topic]."
Options: [
  "Yes, update [name]",
  "No, skip"
]

If yes: append to ~/Tandem/memory/people/{name}.md with meeting date, topic, and key discussion points.

---

Step 7: Generate Debrief Report

Use the template from ${CLAUDE_SKILL_DIR}/template.md.

Full Debrief

Generate a complete report with:

• Executive summary from Q4 response (or AI-generated from other answers if user skipped Q4)
• Decisions made -- numbered list with context and rationale
• Action items -- table with owner, deadline, and kanban routing status (task ID if created/updated)
• Prep items addressed -- only when a prep brief was found in Step 3a. Show which context items from the prep were discussed vs left open. This closes the prep-debrief loop.

No "Open Questions" section -- debrief captures what happened, not what is pending.

Quick Debrief

Generate a compact report with:

• Action items -- table with owner and deadline
• Blockers -- bullet list

No executive summary, no decisions section in quick mode.

Saving the Report

Filename convention: debrief-YYYY-MM-DD-HHMM-slug.md

• Date and time from the meeting's start time in user's local timezone
• Slug: lowercase event title, spaces to hyphens, strip special characters, truncate to 40 characters
• Example: debrief-2026-03-21-1400-client-review.md

Directory: ~/Tandem/meetings/

• Create the directory if it does not exist

Cross-link to prep: If a prep brief exists for this meeting (found in Step 3a), add Related: [prep filename] in the report header and include the "Prep Items Addressed" section.

Save the report using the Write tool.

---

Step 8: Log Stats and Sync

After saving the debrief report, log activity to ~/Tandem/stats.json.

Read ~/Tandem/stats.json. If it does not exist, create it as []. Parse the JSON array, append a new entry, and write it back:

{
  "type": "meeting-debrief",
  "action": "created",
  "count": 1,
  "timeSavedMinutes": 10,
  "description": "Meeting debrief: [Title] with [attendees] ([time])",
  "timestamp": "<current ISO 8601 UTC>"
}

After appending to stats.json, follow the /sync workflow from tandem-skills/core/sync/SKILL.md to rebuild ~/Tandem/dashboard.html with updated statistics. If /sync fails, continue -- the debrief report is the primary deliverable.

---

Step 9: Confirm to User

After the report is saved, stats logged, and sync completed:

• Show the executive summary inline (full mode) or action item count (quick mode)
• Report: "Debrief saved to ~/Tandem/meetings/[filename]"
• Note how many action items were routed to kanban (with task IDs)
• Note how many decisions were saved to memory
• If prep was cross-linked: "Linked to your prep brief: [prep filename]"

No further prompts unless the user asks for changes or drill-down.

---

Task Format

When creating tasks in kanban.md, follow the canonical format in tandem-skills/tasks/task-manager/references/kanban-format.md.

Key rules:

1. Read <!-- Config: Last Task ID: NNN --> from kanban.md to get the next ID
2. Increment the counter: if last ID was 005, next task is TASK-006
3. Write the updated counter back: <!-- Config: Last Task ID: 006 -->
4. Use canonical format: ### TASK-XXX | Title (H3, 3-digit zero-padded, pipe separator)
5. Meta on ONE line: **Priority**: X | **Category**: X | **Assigned**: @user

---

Anti-Patterns

• Do NOT use the Bash tool anywhere -- all operations via allowed-tools and MCP
• Do NOT hardcode MCP tool function names -- describe the operation needed
• Do NOT process transcripts -- redirect to meeting-notes skill
• Do NOT include an "Open Questions" section in the debrief report
• Do NOT require prep to have run -- derive meeting info from calendar or user input
• Do NOT auto-create kanban items without batch confirmation
• Do NOT show per-item confirmation for action items -- batch only
• Do NOT ask about each context source -- gather silently (prep, meeting-notes, kanban)
• Do NOT create per-person subsections in the report -- group context naturally
• Do NOT show empty sections with "None" or "N/A" -- omit entirely

---

Error Handling

• Calendar unavailable: Switch to manual input fallback (Step 2b). Never fail.
• Kanban.md missing: Skip dedup. When writing action items, create new file using kanban-format.md format.
• Memory directory missing: Create ~/Tandem/memory/decisions/ and/or ~/Tandem/memory/people/ as needed.
• Stats.json write fails: Log the error but do not block -- the debrief report is already saved and that is the primary deliverable.
• Sync fails: Continue -- dashboard update is secondary.
• Meeting-notes import fails: Fall back to full Q&A (Step 4). Never fail.
• No prep brief found: Continue without cross-linking. Prep is optional.
• User provides minimal answers: Generate report with available data. Every section with no content is omitted, not shown as empty.