AgentSkillsCN

Voice Meeting

通过语音转录记录会议,并自动生成会议纪要

SKILL.md
--- frontmatter
context: current
model: sonnet
description: Record meetings with voice transcription and create meeting notes

/voice-meeting

Record a meeting using voice capture and automatically create a meeting note with transcript.

Usage

code
/voice-meeting start <title>
/voice-meeting start <title> for <project>
/voice-meeting stop
/voice-meeting status

Examples

code
/voice-meeting start Weekly Team Sync
/voice-meeting start Architecture Review for Project - Alpha
/voice-meeting stop
/voice-meeting status

Prerequisites

Before using this skill, ensure:

  1. The Inworld voice-agent server is running at ws://localhost:4000
  2. Start it with: cd ~/Documents/GitHub/claude-voice && npm run start:server
  3. Microphone permissions are granted

Instructions

For start action:

  1. Parse the command for:

    • title: Meeting name (required)
    • project: Project mentioned after "for" (optional)
  2. Use the voice_listen_continuous MCP tool with action "start":

    code
    mcp__voice-mcp__voice_listen_continuous({
      action: "start",
      title: "<meeting title>"
    })
    
  3. Store the session_id returned for later use

  4. Inform the user:

    • Recording has started
    • Remind them to say /voice-meeting stop when finished
    • Mention they can check status with /voice-meeting status

For stop action:

  1. Use the voice_listen_continuous MCP tool with action "stop":

    code
    mcp__voice-mcp__voice_listen_continuous({
      action: "stop"
    })
    
  2. Retrieve the full transcript and segments from the response

  3. Generate filename: Meeting - YYYY-MM-DD {{title}}.md

  4. Create the meeting note in Meetings/ folder:

markdown
---
type: Meeting
title: { { title } }
created: { { DATE } }
modified: { { DATE } }
tags: [voice-recorded]
date: "{{DATE}}"
project: { { project_link or null } }
attendees: []
summary: { { AI-generated summary of transcript } }
collections: null
recordingSession: { { session_id } }
duration: { { duration_ms formatted as HH:MM:SS } }
wordCount: { { word_count } }
---

# {{title}}

## Recording Info

- **Session ID**: {{session_id}}
- **Duration**: {{duration formatted}}
- **Word Count**: {{word_count}}

## Attendees

(To be filled in)

## Transcript

{{full_transcript}}

## AI Summary

{{Generate 3-5 bullet point summary of the meeting}}

## Action Items

{{Extract any action items mentioned in transcript, or leave template:}}

- [ ]

## Decisions Made

## {{Extract any decisions mentioned, or leave template:}}

## Follow-up

-
  1. After creating, confirm and show:
    • File path
    • Meeting duration
    • Word count
    • Brief transcript preview

For status action:

  1. Use the voice_listen_continuous MCP tool with action "status":

    code
    mcp__voice-mcp__voice_listen_continuous({
      action: "status"
    })
    
  2. Report:

    • Whether recording is active
    • Session ID
    • Current duration
    • Word count so far
    • Transcript preview (last 500 chars)

Error Handling

  • If no Inworld server connection: Remind user to start the voice-agent server
  • If stop called with no active session: Inform user no recording is active
  • If recording fails: Provide helpful error message and suggest checking microphone

Notes

  • Recording automatically includes VAD (Voice Activity Detection)
  • Maximum recording duration is 1 hour (configurable)
  • Transcripts are stored in the session until stop is called