meeting-recorder

name	meeting-recorder
description	Records meetings and retrieves transcripts for post-processing. Use when the user mentions recording a meeting, standup, sync, call, 1:1, retro, or any audio. Also handles transcript retrieval when the user returns and asks to grab, get, or process their transcript.
argument-hint	[meeting name]
allowed-tools	Bash

Meeting Recorder Skill

When to Use

Activate this skill when the user wants to:

• Record a meeting, standup, sync, call, 1:1, retro, or any audio capture
• Retrieve a transcript (phrases like "grab the transcript", "get my notes", "process that meeting")

Source Selection Heuristics

Infer the audio source from context clues in the user's request:

User says	Source	Rationale
"in-person", "standup", "at my desk"	mic	Local microphone captures room audio
"Teams", "Zoom", "video call", "remote"	system	System audio via BlackHole loopback
"hybrid", "conference room with remote folks"	both	Captures both room and remote participants
No context provided	mic	Safe default; mic works without extra setup

Starting a Recording

1. Determine the source using the heuristics above.
2. Call the start_recording MCP tool:
   • source: one of mic, system, both
   • meeting_name: optional, use whatever the user called the meeting (e.g., "standup", "sprint planning")
   • Omit output_path to let the server auto-generate one.
3. The tool returns {status: "recording_started", output_path, source, message}.
4. DO NOT poll, wait, or loop after starting. The recording runs in the background.
5. Tell the user:
   • Recording has started
   • What source is being used
   • To come back when the meeting is done and say something like "grab the transcript"
   • They stop the recording by clicking the menu bar icon

Retrieving a Transcript

When the user returns and asks for their transcript:

1. Call the get_transcript MCP tool.
   • Omit output_path to automatically use the most recent session.
   • Only pass output_path if the user specifies a particular file.
2. Handle the response based on status:

status: ready

Transcript text is in the transcript field. Proceed to post-processing (see below).

status: recording

The meeting is still being recorded. Tell the user:

• "The recording is still running. Click the menu bar icon to stop it, then ask me again."

status: processing

Audio is being transcribed. Tell the user:

• The current processing step (from current_step field)
• To wait about 30 seconds
• Offer: "Want me to check again in a moment?"

If the user says yes, call get_transcript again.

status: error

Something failed. Tell the user:

• Which step failed and the error details (from the error field)
• Suggest re-recording if the error is unrecoverable

ToolError (exception)

No session found. Ask the user:

• "I couldn't find a recent recording session. Do you have the file path for the transcript?"
• If they provide a path, call get_transcript with that output_path.

Post-Processing (When Transcript is Ready)

Once you have the transcript text:

1. Analyze the transcript

• Summarize the meeting in 3-5 concise bullet points
• Extract action items with owners in bold (if identifiable from the conversation)
• Extract key decisions that were made
• Infer meeting metadata: type, attendees, workstreams, whether it is recurring

2. Write to Obsidian vault

Use the Obsidian MCP write_note tool to create a note in the 20-meetings/ directory.

Frontmatter schema (all fields required)

---
type: meeting
created: YYYY-MM-DD
updated: YYYY-MM-DD
workstreams:
  - (infer from transcript content)
status: active
tags:
  - meeting
  - (infer type: standup, sync, retro, 1:1, planning, etc.)
meeting-type: standup|sync|retro|1:1|planning|other
attendees:
  - "[[Person Name]]"
recurring: true|false
source: meeting-recorder
---

Key rules:

• created and updated use today's date in YYYY-MM-DD format
• attendees must use wikilink syntax: "[[Person Name]]"
• meeting-type is one of: standup, sync, retro, 1:1, planning, other
• workstreams and tags are inferred from the transcript content
• recurring is true for standups, syncs, retros; false for ad-hoc meetings

Note structure

# Meeting Title

## Summary
- Bullet point 1
- Bullet point 2
- Bullet point 3

## Action Items
- **Owner Name**: Description of action item
- **Owner Name**: Description of action item

## Key Decisions
- Decision that was made
- Another decision

## Raw Transcript
<details>
<summary>Full transcript</summary>

(paste the complete transcript text here)

</details>

• The H1 title should be the meeting name if provided, or inferred from the transcript content
• If no action items or decisions are identifiable, include the section with "None identified."
• The raw transcript goes in a collapsible <details> block to keep the note scannable