agentleFS
tyiiiyt01tyiiiyt01/agent-starter-kit/skills/meeting-notes/SKILL.md

meeting-notes

Skill1 starsChanged 31 days agoApache-2.0View on GitHub

把会议录音/视频转成文字并整理成结构化会议纪要(参会人、议题、结论、待办、未决事项)。当用户提供录音或视频、要求转写会议内容、整理会议记录或提炼待办事项时使用。默认本地免费方案,涉密会议不上传云端。

  • Installs packages

55 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use email-draft, status-report and 9 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • yakatotechSkill

    email-draft

    Create business email drafts. Use when asked to draft emails

    Also in 3 projects that use this

  • AnthropicSkill

    status-report

    Generate a status report with KPIs, risks, and action items. Use when writing a weekly or monthly update for leadership, summarizing project health with green/yellow/red status, surfacing risks and decisions that need stakeholder attention, or turning a pile of project tracker activity into a readable narrative.

    Also in 3 projects that use this

  • QuixiAISkill

    weekly-review

    Review the past week across goals, tasks, calendar, journal, and commitments, then propose the next week

    Also in 8 projects that use this

  • vijpatel7Skill

    cold-outreach

    Write highly effective cold emails and LinkedIn messages to recruiters, hiring managers, founders, and engineers at target companies. Based on Ben Lang's proven cold outreach framework (Next Play). Use when the user wants to: (1) write a cold email to someone at a company they want to work for, (2) craft LinkedIn messages or InMail to recruiters or hiring managers, (3) reach out about unadvertised positions, (4) follow up on outreach or applications, (5) network with founders or executives for job opportunities, (6) prepare an outreach campaign to multiple companies. Triggers on: cold email, outreach, reach out to, contact recruiter, email founder, write to hiring manager, job email, networking email, LinkedIn message, InMail.

    Also in 3 projects that use this

  • jacobraskSkill

    calendar

    Interact with Apple Calendar via AppleScript. Use when the user asks to check calendar, view events, create events, manage schedule, find free time, or list calendars. Triggers include "my calendar", "my schedule", "calendar events", "create event", "add to calendar", "what's on my calendar", "free time", "available slots", "upcoming events", "today's events". Requires macOS with Calendar.app.

    Also in 3 projects that use this

  • LangChainSkill

    competitor-analysis

    Analyze competitors in a given market segment. Trigger on: competitive landscape, competitor analysis, market comparison, competitive positioning.

    Also in 3 projects that use this

The file

Files
SKILL.md

会议录音转写与纪要整理

0. 先确认三件事

  1. 音频从哪来:录音文件、视频文件,还是要从云端下载?格式是 wav / mp3 / m4a / mp4 里的哪种?
  2. 能否上传云端:会议内容是否涉密?涉密会议只能走本地方案,绝不上传
  3. 要不要区分说话人:多人讨论要标「谁说的」;单人汇报不需要

1. 环境准备

python3 -m pip install faster-whisper
# 非 wav / mp3 的音频先转格式(whisper 对 16k 单声道最友好)
ffmpeg -i input.m4a -ar 16000 -ac 1 output.wav

ffmpeg 没装会报 command not found,先确认系统里有(macOS:brew install ffmpeg)。

2. 本地转写:faster-whisper

pip 包名 faster-whisper,导入名 faster_whisper(下划线),别写混。

from faster_whisper import WhisperModel

model = WhisperModel("small", device="cpu", compute_type="int8")
segments, info = model.transcribe(
    "output.wav",
    language="zh",
    word_timestamps=True,
    vad_filter=True,          # 掐掉长静音,明显提速
)
for seg in segments:          # segments 是生成器,遍历时才真正开始计算
    print(f"[{seg.start:6.1f}-{seg.end:6.1f}] {seg.text}")

要点:

  • segments 是生成器:transcribe() 返回后不遍历就不会计算,别以为它没跑
  • CPU 用 compute_type="int8" 省内存提速;有 NVIDIA 显卡可改 device="cuda"
  • 首次运行会自动下载模型,之后缓存在本地

3. 模型大小与取舍(最重要的决策)

模型体积说明
tiny约 75MB最快,质量差,只够试跑
base约 142MB快,准确率一般
small约 466MB常用均衡点
medium约 769MB更准,明显更慢
large-v3约 1.5GB最准,CPU 上很慢,建议有显卡再用

⚠️ 中文效果提醒(务必如实告诉用户):whisper 系列在中文上偏差较大,small 级别字错率约 22%,纯中文会议转出的稿子需要大量人工校对。

中文场景更好的选择:阿里 FunASR 的 SenseVoice-small,量化后约 250MB,字错率约 8%,可经 sherpa-onnx 纯 CPU 运行、不需要 Python。中文为主的会议优先试它。把「错字率 vs 速度 / 显卡」摆清楚,让用户自己选。

4. 需要区分说话人:whisperX

多人会议要分「谁说的」,用 whisperX(在 whisper 基础上做说话人分离 + 时间对齐)。

  • 它会额外拉取说话人分离模型,下载量大、依赖较重,先确认网络与磁盘
  • 说话人只会标成 SPEAKER_00 / SPEAKER_01 这类编号,具体是谁仍需人工对照

5. 本地 vs 云端

本地(faster-whisper / SenseVoice)云端(OpenAI / 阿里云 / 讯飞)
费用免费需 API Key,按量付费
隐私数据不出本机音频要上传
速度慢(尤其 CPU)快
效果取决于模型与硬件通常更好

选择原则:涉密 / 敏感 → 本地;追求效果且不敏感 → 云端。

6. 纪要结构化模板

转写完不要直接把流水稿丢给用户,按下面结构整理:

参会人:
时间:
议题:
  1. ...
结论:
  - 就 X 达成一致:...
待办:
  - [ ] 谁 · 做什么 · 何时完成
未决事项:
  - X 尚无结论,待 Y 后确认

7. 从纪要提取待办

  • 每条待办必须有责任人和时间点:缺谁就问、缺何时就写「时间待定」
  • 原文含糊(「尽快」「回头弄」)的,标成「待确认」,不要替用户臆造截止日期
  • 结论与讨论要分开:讨论过程不等于已定结论

8. 隐私与合规(硬提醒)

  • 会议录音可能含商业机密、个人信息,默认走本地方案
  • 上传云端前必须确认合规:取得参会人 / 公司许可,确认云服务的数据条款
  • 转写稿和纪要也要注意存放位置,别随手放进共享目录

9. 常见坑自检

  • 音频已转成 wav / mp3,采样率与声道合适
  • 模型已选对(中文优先考虑 SenseVoice-small)
  • 遍历了 segments 生成器,确实拿到了文本
  • 转写稿人工校对过(中文错字率不低)
  • 待办的责任人和时间都写清或标了「待确认」
  • 已确认是否允许上传云端

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.