Turn meeting recordings into a task list with AI
Skills used in this guide (5)
项目管理专家与产品经理在做项目决策、技术评审或商业策略制定时，当需要多视角碰撞以打破思维局限，请用此技能模拟多智能体真实会议辩论，一键生成包含决策结论、核心论据与风险提示的完整会议记录，让复杂决策高效落地！
Download pack → https://github.com/anbeime/skill/tree/main/skills/multi-agent-meeting/multi-agent-meeting把会议记录转成带出处的决策、负责人与工单。
npx skills add NousResearch/hermes-agent -g从文档里摘出带出处的义务、截止时间与待办事项。
npx skills add NousResearch/hermes-agent -g加入 Google Meet 会议、实时转写字幕、可选实时发言，并在会后生成待办。适合让智能体旁听会议、记笔记、总结或当场回应。
npx skills add NousResearch/hermes-agent -g处理 Teams 会议摘要、任务回放与 Graph 订阅。
npx skills add NousResearch/hermes-agent -g
The biggest waste in meetings isn't the meeting — it's nobody remembering "who, by when, what" afterward. The point of this flow is to compress scattered discussion into a task table with owners and due dates, not to write another minutes doc nobody reads.
Who this is for / not for
For:
- Standups, design reviews, retros — meetings where a conclusion means an action
- You already have audio / transcript, or can let AI join
- You want a trackable list, not a long narrative
Not for:
- Pure brainstorming with no conclusion (AI can't extract tasks either)
- Confidential meetings (transcripts go to cloud — check data-export compliance)
- Formal proceedings needing verbatim records (that's transcription, not synthesis)
Before you start
- Material: recording file, or transcript text. Transcribe first if needed (Meet/Teams plugins do it automatically).
- Context: what kind of meeting, what output (decision / task / risk).
- (Optional) attendee → account map: so owner lands on a person, not "the product team".
Steps
1. Install the skill
multi-agent-meeting is a multi-agent listening skill. Copy its folder into your client's skills directory (same as any Agent Skill):
| Client | Directory |
|---|---|
| Claude Code | ~/.claude/skills/ |
| Codex | ~/.codex/skills/ |
| Cursor | ~/.cursor/skills/ |
| Trae | ~/.trae/skills/ |
| CodeBuddy | ~/.codebuddy/skills/ |
| Tongyi Lingma | ~/.lingma/skills/ |
2. Transcribe first (if no text)
Transcribe this meeting audio into text, keep speaker labels (Speaker A/B/C),
drop small talk and filler, but don't remove any conclusion, number, or commitment.3. Extract decisions and tasks
Give the transcript to AI and do one thing: structure it.
From the transcript, produce two tables:
1. Decisions: conclusion / proposed by / objections (if any)
2. Tasks: task / owner / due date / acceptance criteria
Rules:
- owner must be a specific person; if unknown mark [TBD]
- due date must be concrete; if unknown mark [this week]
- each task must be trackable alone; don't mergeFor lightweight cases, meeting-action-items alone outputs tasks. To back-fill obligations from a long doc, use document-to-action-items.
4. Post back and confirm
Paste the task table into the group / ticketing system and @ the owners. Don't skip this — AI's owner guesses are often ambiguous; one confirmation makes the task real.
Pitfalls
| Pit | Symptom | Fix |
|---|---|---|
| Owner is a role | "product handles it" not "Zhang" | Require a real name, else [TBD] |
| Vague due date | "ASAP" "next week" | Force concrete date, else [this week] |
| Discussion as conclusion | Tasks mix undecided talk | Separate decisions from tasks |
| Lost speakers | Can't tell who said what | Name speakers beforehand or correct after |
| Secret on cloud | Compliance risk | Use local model for internal meetings |
FAQ
No recording, only notes?
Yes. document-to-action-items extracts obligations and deadlines from docs, including rough notes.
Is live AI transcription worth it?
Yes but needs consent. Meet/Teams plugins let AI join, transcribe, follow up; announce "AI is recording" at the start.
Can tasks go straight to Jira / tickets?
Yes — have AI output in your ticket field format, then paste or wire the API.
Owners keep being wrong in multi-person meetings?
Give an "attendee → account" map beforehand and require owner by map; accuracy rises.
Do I still need minutes?
If you have conclusions, the decision + task tables are enough. For archive, ask AI for a 200-word summary from those two tables.
Related guides
Continue across these scenes:
- Write weekly reports and standups with AI
- Handle your inbox and support messages with AI
- Write PRDs and product specs with AI
相关源码
- anbeime/skill · multi-agent-meeting source
- NousResearch/hermes-agent · meeting-action-items source
- Agent Skill open format spec
一句话结论
结论：会议的价值在「谁、何时、交什么」——让 AI 把讨论压成带 owner 和截止时间的待办表，比再写一份没人看的长纪要实用得多。
References
Related Scenes
Write PRDs and product specs with AI
The pain of PRD isn't "wrong format" — it's "goal and acceptance unclear, dev guesses". The point: give background, target user, problem; let AI output "feature list + acceptance + edges", not decide "why build" for you.
Write weekly reports and standups with AI
The pain of weekly reports isn't "can't write" — it's "turning a log into outcomes". The point: give AI the raw material (what you did, what's stuck), have it close into "progress / risk / next", not invent content for you.
Translate and localize docs with AI
Machine translation's scare is "fluent but distorted" — wrong terms, wrong tone, stiff local phrasing. The point: give a glossary and target-reader tone, have it "translate + keep proper nouns", then you verify key sentences — not trust the whole blindly.
Handle your inbox and support messages with AI
The red line for email automation is "draft yes, auto-send careful". "Triage + draft + you hit send" is safest; letting AI blast-send means you own the mistake, not it.