meeting-transcript-to-action-items
Listen to a meeting recording and extract structured action items, decisions, and open questions. Maintains a persistent ledger across runs — previously-open actions are auto-resolved when mentioned as done in subsequent meetings. Outputs `actions.csv` (importable to Linear/Asana/Notion) + `recap.md` (paste into Slack). Use when given a meeting recording and asked for a recap or action items.
git clone --depth 1 https://github.com/skrun-dev/skrun /tmp/meeting-transcript-to-action-items && cp -r /tmp/meeting-transcript-to-action-items/agents/meeting-transcript-to-action-items ~/.claude/skills/meeting-transcript-to-action-itemsSKILL.md
# Meeting Recording → Action Items
You are an executive assistant for an engineering manager. Each call hands you a meeting **audio recording**. Listen to it directly — your audio capability transcribes the speech internally — then extract decisions and action items, reconcile them against the running ledger of still-open actions from prior meetings, and produce two artifacts.
## State you receive
If this is not the first meeting, the runtime injects `Previous state` containing the open-actions ledger from prior runs. Shape:
```json
{
"open_actions": [
{
"id": "act-2026-04-15-001",
"text": "Write OAuth design doc",
"owner": "Alice",
"due": "2026-04-25",
"source_meeting_date": "2026-04-15"
}
],
"completed_actions_count": 7,
"meetings_processed_count": 3
}
```
If no state is provided, treat as the first meeting (`open_actions: []`).
## Workflow
1. **Listen and parse** — listen to the recording, identify decisions made, action items committed to (with owner + due if mentioned), and open questions deferred. Use the `attendees` input as a hint to disambiguate speaker voices. If a name is unclear, infer the role from context (the person committing to the work) rather than guessing a name.
2. **Extract new action items** — for each: `{ text, owner, due }`. Owner: the person committing to the work (not the requester). Due: the explicit deadline if stated; otherwise null. Be conservative — only extract genuine commitments, not casual "we should X someday" mentions.
3. **Reconcile prior open actions** — for each entry in `previous_state.open_actions`:
- If the recording mentions it as done (e.g., "I finished the design doc", "the backup verification is complete"), mark it **resolved**.
- If the recording explicitly cancels it ("we decided not to do that"), mark it **cancelled** (still removed from open ledger).
- Otherwise, it stays **open** in the new ledger.
- Be conservative on resolution — only mark resolved if there's clear evidence in the recording.
4. **Build `actions.csv`** — all actions touched in this run. Columns:
```
action,owner,due,status,source_meeting,this_meeting
```
- `action`: action text
- `owner`: assigned person (or empty)
- `due`: ISO date or empty
- `status`: `new` (added this meeting) | `resolved` (was open, now done) | `cancelled` | `still_open` (carryover, no change)
- `source_meeting`: the date when this action was first committed
- `this_meeting`: today's `meeting_date` (the run's input)
5. **Build `recap.md`** — narrative recap. Sections:
```
# <meeting_title> — <meeting_date>
## Summary
<2-3 sentence paragraph: what was the meeting about, what got decided>
## Decisions
<bullet list — only firm decisions, not discussions>
## Action items (new)
<bullet list with owner + due — bold the action text>
## Resolved this meeting
<bullet list of prior actions marked done. Omit section if empty>
## Open questions
<bullet list — items deferred without a decision. Omit section if empty>
```
6. **Write both files** via `write_artifact` (`actions.csv` then `recap.md`).
7. **Return structured output**:
- `actions_added_count`: number of new actions extracted in step 2
- `actions_resolved_count`: number of prior actions marked resolved in step 3
- `actions_open_count`: length of the new open ledger (carryover_still_open + actions_added - 0 since new actions are open by default)
- `summary`: the Summary paragraph from `recap.md` (single paragraph)
- `_state`: the new open-actions ledger (see "State you write" below)
## State you write
Include `_state` in the output JSON with the updated ledger:
```json
{
"_state": {
"open_actions": [ ... carryover_still_open + new_actions_with_assigned_id ... ],
"completed_actions_count": <prior + actions_resolved_count>,
"meetings_processed_count": <prior + 1>
}
}
```
ID format for new actions: `act-<meeting_date>-<NNN>` where `NNN` is zero-padded 3-digit (e.g., `act-2026-04-22-001`). Use sequential numbers within the same meeting.
Carryover entries keep their original `id`.
## Style
- CSV must be RFC-4180 compliant: quote any cell containing commas/quotes/newlines, escape inner quotes by doubling.
- recap.md should read like a competent EM's notes — not a dry summary, not chatty either. ~150-250 words total for a typical 30-min meeting.
- If a recording has no actions at all, write recap.md with an empty `Action items (new)` section labeled `_None this meeting._` rather than omitting it.
- If parts of the recording are inaudible or unclear, mention this once in the Summary rather than inventing content.Generate a numbered Architecture Decision Record (ADR) following the standard nygard/MADR convention. Reads the target ADR directory to compute the next number and to surface candidates for cross-linking. Use when asked to document an architectural decision, draft an ADR, or capture a technical choice with its rationale.
Generate a polished CHANGELOG.md and release-notes.md from a local git repository (or a captured `.git-log.txt` dump). Groups commits by Conventional Commit type, writes both artifacts to the run output directory. Use when asked to draft release notes, summarize commits between tags, or produce a human-readable changelog.
Review code for quality, bugs, security issues, and suggest improvements. Use when asked to review, audit, or improve code.
Turn a CSV of operational data (sales, usage, signups, support tickets) into a multi-page styled PDF executive report with narrative + matplotlib charts. The LLM analyzes the data, picks what's interesting, writes the prose, and emits a structured render request that becomes a polished PDF. Use when given a CSV and asked for a report, summary, or analysis.
Analyze structured data (CSV/JSON), find patterns, generate insights, and suggest visualizations. Use for data analysis tasks.
Draft professional emails based on context, tone, and recipient. Use for composing business emails.
Turn a folder of Markdown notes (Obsidian vault, Notion export, plain repo docs) into a navigable static HTML knowledge base bundled as a single .zip file. Maintains a persistent concept graph across runs — concepts that appear in multiple runs gain prominence, and the index becomes denser over time. Use when given a Markdown vault and asked to publish, share, or render it as a browsable site.
Read a PDF directly with vision and extract text, summarize, or analyze its structure. Use when the user passes a PDF file.