meeting-notes — Turn Raw Notes into Structured Summary + Action Items

Single-file Claude skill that turns a wall of raw meeting notes (or a transcript) into a clean structured doc: TL;DR, decisions, action items with owners, open questions, and follow-ups. Drop one file in. No zip.

meeting-notes takes the messy text you paste after a meeting — bullet stream, transcript dump, or scribbled notes — and turns it into the document everyone actually reads: a four-line TL;DR, the decisions made, action items with owners and dates, open questions, and a list of who’s expected to do what before the next meeting.

One file. One copy. Restart. Ready.

At a glance

Field Value
Name meeting-notes
Version 1.0.0
Files 1 file (SKILL.md)
Runtime None — pure prompting
Triggers on “clean up these meeting notes”, “summarise this meeting”, “extract action items”, “what did we decide”, “who’s doing what after this call”
Does NOT trigger on writing meeting agendas, transcribing audio (use a different skill)

Step 1 — Create the folder

bash
mkdir -p ~/.claude/skills/meeting-notes

Windows:

powershell
New-Item -ItemType Directory -Force -Path "$env:USERPROFILE\.claude\skills\meeting-notes" | Out-Null

Step 2 — Paste this single file

Path: ~/.claude/skills/meeting-notes/SKILL.md

markdown
---
name: meeting-notes
description: Use this skill whenever the user pastes raw meeting notes, a transcript, or a bullet dump and asks for a clean version, summary, action items, decisions, or follow-ups. Trigger on "clean up these meeting notes", "summarise this meeting", "extract action items", "what did we decide", "who's doing what after this call", or when the user pastes obvious meeting-style notes (timestamps, speaker names, "we discussed", "AI: …"). Do NOT use for writing a meeting AGENDA before a meeting, or for transcribing audio — those are different jobs.
version: 1.0.0
---

# meeting-notes

Turns a wall of raw meeting text into the document everyone actually reads after the call.

## When invoked

Output ONE markdown document with these five sections, in order. If a section has nothing, write `_None._` rather than omitting it — readers scan for the headings.

### 1. TL;DR

Three to five lines. Plain English. The kind of summary you'd give a colleague who skipped the meeting and asks "what'd I miss?"

### 2. Decisions

Each decision = one bullet. Format: **decision** — _rationale in half a sentence_. If the decision is conditional ("we'll do X if Y by date Z"), make the condition explicit.

### 3. Action items

Markdown checklist. Each item:

    - [ ] Owner — verb + deliverable — due YYYY-MM-DD

If the owner is unclear in the source notes, write `Owner: TBD` and call it out at the bottom. If the date is unclear, write `due TBD`. Never invent owners or dates.

### 4. Open questions

Bullets, each one a sentence framed as a question. These are the things the meeting did NOT resolve. If they were resolved, they belong in section 2.

### 5. Follow-ups before next meeting

A short list of "before we meet again, X should happen". Differs from action items in that these are typically pre-conditions for the next meeting, not standalone tasks.

## Style rules

- Names: capture exactly as they appear in the source (don't normalise "Sarah" → "Sarah K." unless source did).
- Acronyms: keep as-is. If the source spells one out the first time, do the same.
- Numbers and dollar amounts: keep verbatim.
- Quotes: preserve only when the wording matters (commitments, controversial calls). Otherwise paraphrase.
- Voice: third-person, past tense ("the team agreed", not "we agreed").

## When the input is messy

- **Transcript with timestamps and speakers.** Strip them. Group by topic, not by who said what.
- **Bullet stream from one note-taker.** Don't preserve their bullet structure if it's chronological — re-order by topic for sections 2 & 3.
- **Two competing accounts (e.g., user pastes their notes + a colleague's).** Reconcile silently where they agree. Where they conflict, list the conflict in section 4 (Open questions) — do not pick a winner.
- **Hebrew or mixed-language input.** Output in the same language(s) the user used. Don't translate unless asked.

## When NOT to use

- Writing a meeting AGENDA in advance — different skill (proposal-style writing, not summarising)
- Transcribing audio to text — different skill (audio I/O)
- Drafting follow-up emails based on the notes — possible follow-up step, but don't auto-do it. Ask first.

## Output

Just print the markdown to the user. Don't save to disk unless asked. The whole point is paste-back-into-Notion-or-email.

Step 3 — Restart Claude Code

Quit, re-open. Skill registers on session start.

Step 4 — Try it

Paste any meeting notes:

text
Clean up these notes:

10:02 Sarah: kicked off, quick review of last week's numbers
10:05 we agreed to pause the EU rollout until legal signs off
  - Yossi to email legal today, due Friday
10:11 dan brought up the API rate limit issue, no decision
10:14 sarah said we should ship the search redesign next sprint
  - tania owns it
  - blocked on figma handoff from design
10:20 quick aside on hiring, 2 candidates this week
10:23 wrap, next meeting tuesday

The skill produces a clean TL;DR, the EU pause + search ship as decisions, three checklist action items with owners, the API rate limit as an open question, and the figma handoff as a pre-meeting follow-up.

Customising

  • Different output language by default. Add a line under “Style rules”: “Default output language: Hebrew” (or whatever).
  • Always save to disk. Edit the “Output” section to read: “Save as ./meeting-YYYY-MM-DD.md and print path to user.”
  • Different section order. If your team scans for action items first, swap the order of sections 1 and 3.

For more downloadable skills, see the Skill Library. For the mechanics behind SKILL.md, see Anatomy of a Skill.