Meeting Notes to Action Items

When to use

Triggers:

• "Turn these meeting notes into actions"
• "What did we decide in [meeting]?"
• "Clean up these meeting notes"
• "Extract action items from this transcript"
• A transcript / notes file uploaded or pasted

Don't fire for:

• Building the agenda (use meeting-agenda-builder)
• Drafting the follow-up email (use email-template-library follow-up pattern)
• THL-specific Gemini notes with Partner Tier context (use the internal thl-meeting-notes-to-followups — voice and channel rules differ)

The three things every meeting produces

Raw notes conflate these. Useful notes separate them:

1. Decisions

What: something was decided (or explicitly deferred).

Format: "We decided to [X] because [reason]. [Owner] owns the execution."

Common miss: decisions aren't written down, so six weeks later someone asks "wait, did we decide X?"

2. Action items

What: someone agreed to do something by a specific time.

Format: Task · Owner · Due date · Optional context

Common miss: "someone should follow up" — nobody owns it.

3. Open questions

What: things that came up but couldn't be resolved in the meeting.

Format: The question · Who's finding out · When we'll know

Common miss: discussion points treated as decisions when they're really unresolved.

Input types

Verbatim transcript (from Zoom / Google Meet / Teams)

Usually dumped in blocks of speaker + time + text. Lots of filler. Clean-up work is high.

Auto-summarised notes (Otter.ai, Fireflies, Gemini, Read.ai)

Already has some structure — usually Summary / Highlights / Action Items sections. Quality varies; the skill improves on these.

Hand-written notes (scrappy)

User's own notes, often bullet points or shorthand. Needs interpretation more than extraction.

Agenda + live edits

User took notes directly in the agenda. Usually has good structure already; needs cleanup + extraction.

Workflow

Step 1 — Identify the meeting type

Different meeting types produce different output shapes. Quick sort:

• Stand-up / check-in → commitments per person, blockers log, no formal decisions usually
• Decision meeting → the decision + owner + context
• Brainstorm → idea clusters, top candidates, decided next experiment
• Quarterly review → quarterly plan + commitments + metrics noted
• 1-on-1 → private notes (coaching points, feedback) + any action items to share
• Client catch-up → shared recap + internal notes

Ask if unclear from the notes.

Step 2 — Extract the three categories

Pass through the notes, extracting into three buckets:

DECISIONS
- [Decision] — [Reason] — [Owner for execution]

ACTION ITEMS
- [ ] [Task] — [Owner] — Due [Date]
- [ ] [Task] — [Owner] — Due [Date]

OPEN QUESTIONS
- [Question] — [Who's investigating] — Follow-up by [Date]

Each item must have an owner. "The team" is not an owner. "Someone" is not an owner. If you can't name a person, flag it and note it needs ownership.

Step 3 — Write the recap

A one-paragraph narrative suitable to paste into a Slack channel, an email, or a shared doc:

Recap — [meeting name], [date]

[Attendees: names or "team"]

The big thing: [1 sentence — what was the main outcome?]

Decided: [bullet or two]
To do: [high-priority action items, owner + date]
Open: [unresolved questions]

Next meeting: [date / trigger / none]

Step 4 — Cross-check with the agenda (if you had one)

If the meeting had an agenda:

• Did every agenda item get addressed? (If not, did we consciously drop it or did it fall off?)
• Are there decisions the agenda promised that weren't made? (Add as open questions.)
• Did new items come up that weren't on the agenda? (Fine, note them.)

This check catches the meetings that felt productive but actually dodged the hard stuff.

Step 5 — Flag the questionable

Things worth surfacing back:

• Unowned actions — "Action: improve onboarding" — owner?
• Fuzzy due dates — "Soon", "ASAP", "next week" without a day — ask for specifics
• Decisions without owners — "We decided to launch the new pricing" — who executes?
• Commitments that need more people — an action item owned by one person that can't be done without others — flag the coordination
• Silence on conflict — if the transcript shows tension that got glossed over, flag for follow-up (internal only, not in the shared recap)

Output template

# [Meeting name] — [Date]

**Attendees**: [names]
**Duration**: [actual time]
**Next meeting**: [date or trigger]

## Recap

[2–3 sentence narrative of the main outcome]

## Decisions

1. [Decision 1]
   - Why: [reason from the discussion]
   - Owner for execution: [name]

2. [Decision 2]
   - ...

## Action items

| Item | Owner | Due | Notes |
|---|---|---|---|
| [Task] | [Name] | [Date] | [Context] |
| [Task] | [Name] | [Date] | [Context] |

## Open questions

- **[Question]** — [Name] is investigating, follow-up by [Date]
- **[Question]** — ...

## Items for clarification (internal — not for shared recap)

- [ ] [Unowned action] — needs owner
- [ ] [Fuzzy timeline] — needs specific date
- [ ] [Mentioned but undecided] — bring to next meeting

Handling different note sources

From Otter.ai or Fireflies automatic summaries

Often the auto-generated action items miss ownership or are too vague. Use them as a starting point but re-verify:

• Cross-check against the transcript's actual quotes
• Add owners where the tool didn't detect them
• Tighten vague dates

From a Gemini / Google Meet transcript

Gemini summaries group by topic. Good for decision-meeting notes. Usually needs the "open questions" extracted separately — Gemini's good at summarising what was said but less at flagging what wasn't resolved.

From hand-written scrappy notes

The interpretation load is higher. Ask:

• What was the meeting called for?
• Who was in the room?
• What did you feel was the main outcome?

Then draft and verify against their memory.

Voice / tone for the shared recap

The shared recap (that goes to the whole room + any stakeholders) should be:

• Factual, not editorial ("we decided X" not "we wisely decided X")
• Neutral on conflict — if tension was in the room, don't amplify it in writing
• Attribute decisions to the group, not individuals (unless a specific person's decision mandate)
• Respectful of agreements — don't quote someone who might regret it in writing

Internal notes (for your own record) can be more direct about tensions, open questions, and what was avoided.

Common failure modes

• The action-item list nobody follows up on. Fix: owner + due date + a review ritual (not just the next meeting, but a weekly check-in).
• Shared recap that mischaracterises the meeting. Fix: let people edit the recap for 24 hours before locking it.
• Decisions that get re-litigated 3 weeks later because they weren't written down clearly. Fix: explicit "decided" framing + why.
• Meeting notes in 15 different places. Fix: one location per team/project. Index in one top-level doc.

Output format

Per request:

1. Shared recap — the version suitable for broadcast to everyone who was in the meeting + relevant stakeholders
2. Action-item list — clean, owner + due date + task
3. Open questions list — with investigators and follow-up triggers
4. Internal flags (not for the shared recap) — unowned actions, fuzzy dates, undiscussed tensions, follow-up reminders

What this skill does NOT do

• Send the recap. Paste or post yourself after review.
• Track action items over time. Use a project tool (Asana, Notion, Linear, Trello, or even a shared doc) for ongoing follow-through.
• Facilitate the meeting. Different skill entirely.
• Magic up decisions that weren't made. If a meeting avoided a hard decision, the recap will show that — that's useful signal, not a bug.

Tier access

Base. Universal need. Pro-tier adds integration with transcript sources (upload a .vtt from Zoom and get the structured output) + action-item tracking across weeks.

Related skills

• meeting-agenda-builder — the before-the-meeting complement
• email-template-library — for the follow-up email pattern
• thl-meeting-notes-to-followups (internal) — THL-specific with Gemini-voice + Telegram channel rules

References

• Patrick Lencioni — Death by Meeting (on making meetings actually produce output)
• Otter.ai, Fireflies, Gemini Meet notes — common transcript/summary tools