| name | meeting-recap |
| category | kumite-task |
| description | Produce a recap of decisions, actions, and owners — not a transcript. Use within 24h of any meeting where decisions were made or actions assigned. For a recurring series, also produce a private behavioral readout over a set window — speaking-time, decision velocity, and action carryover as a share of the whole, each finding tied to one concrete change. Triggers: "recap the meeting", "send notes", "what did we decide", post-meeting summary, "are our standups working", recurring-meeting health, meeting-series readout. |
Meeting Recap
Recaps that read like transcripts get ignored. Recaps with three sections — Decisions, Actions, Open — get acted on.
Quick Reference
| Section | Format |
|---|
| Decisions | One line each · who decided · effective when |
| Actions | Owner · verb + artifact · due date |
| Open | Question · who answers · by when |
When to Use
- Within 24h of any decision-making meeting.
- After steerco, board, project kickoff, customer call.
- Any meeting with assigned actions.
How to Use
Prompt Copilot from the Cowork transcript or your notes:
- Extract Decisions — one line each, who decided, effective when.
- Extract Actions — owner, action, due date. No owner = no action — flag instead.
- Extract Open questions — what's unresolved and who needs to answer.
- Verify every owner against the actual attendee list. Hallucinated owners are the most common defect.
Format the recap so it can be pasted into a Teams channel or email — no preamble, no "in this meeting we discussed."
Examples
| Don't: Transcript-recap | Do: Decision-recap |
|---|
| "We discussed pricing and it was a long conversation…" | "Decision: Adopt option B pricing, effective Q4 (decided by SLT)." |
| "Action: look into pricing" | "Action: Maria — draft pricing FAQ for sales by Fri 11/14." |
| Recap pasted with attendee names invented | Owners verified against attendee list; unknowns flagged |
| 1500-word recap | 300-word recap with three sections |
Critical Rules
- Three sections only. Decisions · Actions · Open.
- Every action: owner + verb + artifact + date. No owner = flag, not assign.
- Verify owners against the attendee list. Most common hallucination.
- Send within 24h. Memory decays fast.
- No preamble. Start with the first decision.
Common Pitfalls
- Pasting the full transcript and calling it a recap.
- Vague actions ("look into pricing").
- Recaps you don't send within 24h.
- Decisions without an effective date.
- Mixing personal commentary into the decision lines.
Anti-Patterns
- Pasting the full transcript and calling it a recap.
- Vague actions ("look into pricing"). Every action: verb + artifact + date.
- Recaps you don't send within 24h. Memory decays fast.
- Hallucinated owners.
Verify Before Sending
Optional: Series Health Readout (recurring meetings)
For a recurring meeting (weekly leadership, sprint reviews, steerco, daily standup), the single-meeting recap misses the pattern. Once per cadence — monthly, or every 4-6 occurrences — produce a second, private companion: a behavioral readout of the series, for the meeting owner only, never circulated. The recap reports what happened; the readout reports whether the meeting is working.
The method is borrowed from session-usage analysis and reframed for meetings: pick a window, measure where the time and the decisions actually went, and express each finding as a share of the whole so it is impossible to wave away.
-
Set the window. Decide the range before you read anything — last 4 weeks, last 6 occurrences, this quarter. State it on the readout. A finding without a window ("Priya talks a lot") is an opinion; a finding with one ("Priya held 48% of speaking time across the last 6 standups") is a measurement.
-
Pull the signals from the transcripts in the window. Ask Copilot for each as a number, not a vibe:
| Signal | What to measure | The threshold that earns a line |
|---|
| Speaking-time share | Each attendee's talk time as a % of the meeting | One voice over 50% in a group of 6+ |
| Decision velocity | Decisions reached vs. agenda items, per occurrence | Below half the agenda closing, occurrence after occurrence |
| Action carryover | Actions reopened/rolled from a prior occurrence as a % of all actions | Carryover above ~30% — the meeting assigns but does not land work |
| Airtime vs. ownership | Who speaks most vs. who leaves with actions | High talk share, low action share (or the reverse) |
| Filler / hedge density | Hedged commitments ("we should probably"), unanswered questions | A cluster on one recurring topic — a sign it is being avoided |
| Time-to-first-decision | Minutes of the meeting before the first decision lands | Most of the meeting gone before anything is decided |
-
Write 3-5 findings, each as one line, each a share of the whole. Lead with the figure. Name the subject. Classify the line so the owner reads the shape at a glance — use plain-text labels, never colored dots or emoji:
Concern: waste or a worsening trend (the meeting is leaking time or not deciding).Healthy: a signal that the meeting is working — say it so the owner keeps doing it.Context: a neutral fact that frames the others.
Example lines:
Concern: 62% of the last 4 steercos closed under half the agenda - decisions are slipping to email.Concern: one voice held 51% of airtime in a room of 7; two attendees spoke under 4% each.Healthy: action carryover fell from 40% to 12% after owners were named in the recap.Context: 9 of 11 decisions this month landed in the final 10 minutes - the meeting decides late.
-
Tie each concern to one concrete change. A readout that only names problems is half a job. For every Concern, add one specific, testable adjustment pinned to the row it came from — the meeting analogue of an optimization tied to a specific line, not generic advice.
- Don't: "Improve participation."
- Do: "Airtime concentrated 51% on one voice -> pre-assign two agenda items to the quietest attendees next occurrence; recheck the split in 4 weeks."
- Don't: "Be more decisive."
- Do: "First decision lands at minute 38 -> move the decision item to the top of the agenda and timebox discussion to 10 minutes."
-
Keep it private and coach the meeting, not the people. This readout is for the owner to run a better meeting, never a scorecard pasted into a shared channel or used in a review. Never circulate names plus figures. If a finding implicates one person, frame the fix as a structural change to the meeting, not a callout of the individual.
Related
