AGENT SKILL

meeting-notes-ai

Use AI for meeting notes — automated summaries, action item extraction, and turning meetings into searchable knowledge.

License 许可
MIT
所属仓库
TanNhatCMS/jarvis
最近更新
2026-11-07
Stars
★ 0
Forks
1

01 前言 Front Matter

name
meeting-notes-ai
category
enterprise-communication
description
Use AI for meeting notes — automated summaries, action item extraction, and turning meetings into searchable knowledge.

03 SKILL.md 正文

Overview

AI meeting assistants transcribe, summarize, and extract action items from meetings — turning ephemeral conversations into durable, searchable records. This skill covers using them well: setup and consent, getting quality summaries, action-item hygiene, integrating notes into workflows, and the privacy considerations of recording everything.

AI meeting notes — automatic transcription, summarization, action-item extraction, and searchable archives — promise to end manual note-taking and forgotten follow-ups. The value is real but depends on setup: which meetings get recorded, how summaries are structured, where action items flow, and how privacy concerns are handled. Deployed thoughtfully, it upgrades every meeting's output.

When to use

  • Choosing or configuring an AI note-taker

  • Getting useful summaries (not just transcripts)

  • Ensuring action items don't get lost

  • Building a searchable meeting knowledge base

  • Handling recording consent and privacy

  • Integrating notes with project tools

  • Eliminating manual note-taking

  • Creating searchable meeting archives

  • Enforcing action-item follow-through

  • Standardizing notes across client-facing teams

  • Building institutional knowledge from conversations

  • Supporting async teams across time zones

Core concepts

Consent and norms. Announce recording (legally required in many jurisdictions — two-party consent states/countries), respect opt-outs (some 1:1s shouldn't be recorded), and establish team norms: which meetings get AI notes (recurring team meetings yes, sensitive HR no). Transparency builds acceptance.

Transcript vs. summary. Raw transcripts are searchable but unreadable; good summaries have: key decisions, action items (owner + deadline), open questions, and discussion highlights. Configure summary templates per meeting type — a standup summary differs from a discovery call summary.

Action item hygiene. AI extracts candidate actions; humans verify. Every action needs: clear owner, deadline, and context. Push verified actions into the task system automatically — actions living only in meeting notes die there.

Searchable knowledge. The compound value: months of meetings become a queryable corpus ("what did we decide about pricing in Q2?"). Tag meetings by project/topic, keep summaries structured, and make search easily accessible. This is the real ROI.

Accuracy limits. AI misattributes, misses nuance, and hallucinates details. Critical decisions should be confirmed in writing by participants. Treat AI notes as a draft requiring human review for anything important.

Privacy and retention. Who can access recordings? How long are they kept? What's excluded (sensitive topics)? Define retention policies and access controls — a permanent record of every word said changes how people speak.

Consent and privacy. Recording laws vary (one-party vs. two-party consent jurisdictions); company policy must be explicit; participants should be notified (verbal + visual indicator). Some meetings should never be AI-recorded (sensitive HR, legal, certain client discussions) — maintain an exclusion list and honor opt-outs gracefully. Privacy diligence is the price of admission.

Summary quality factors. Audio quality (good mics beat good AI), speaker identification (named participants vs. "Speaker 2"), agenda structure (meetings with agendas summarize dramatically better), and custom vocabularies (product names, acronyms, client names). Invest in these inputs — output quality follows.

Action-item extraction. The highest-value AI output: decisions made, tasks assigned (owner + deadline inferred or flagged when missing), and open questions. But AI misses context — establish a human review step for action items before they enter task systems. Trust but verify, especially for commitments.

Template design per meeting type. Sales calls: needs, objections, next steps, decision timeline. Standups: yesterday, today, blockers. Planning: decisions, owners, deadlines. Reviews: feedback, action items, follow-ups. One generic template produces generic notes — tailor ruthlessly. Let teams customize within guardrails; adoption follows ownership. Integration architecture. Notes → CRM (sales calls) → task manager (action items) → knowledge base (decisions) → calendar (follow-ups). Notes that do not flow anywhere get read once and forgotten. Automate the routing; humans should only review, not copy-paste. Accuracy and limitations. AI struggles with: heavy accents, crosstalk, domain jargon (without custom vocabularies), and implied commitments ("I'll try" vs. "I will"). Set expectations honestly; provide correction workflows; never use raw transcripts as contractual records.

Practical workflow

  1. Set norms. Which meetings are recorded, consent process, opt-out rules, and access/retention policies. Get team buy-in — mandated recording without discussion breeds resentment.
  2. Configure. Connect calendar, set summary templates per meeting type, define action-item routing (to task tools), and set sharing defaults (attendees automatically get notes).
  3. Run meetings well. AI notes don't fix bad meetings: agendas still matter, and clear verbal decisions ("so we're deciding X, owner Y, by Friday") produce clean summaries. Speak decisions explicitly — the AI captures what's said.
  4. Review and verify. Skim AI summaries after important meetings; correct action items and owners; confirm decisions in the notes. 2 minutes of review prevents weeks of confusion.
  5. Route actions. Verified actions → task system with owners and dates. Decisions → decision log or docs. Follow up on overdue actions from meeting notes.
  6. Leverage the corpus. Use search for prep ("what did the client say about timeline?"), onboarding (new joiners review past discussions), and audits (decision history). Review privacy settings quarterly.

Summary template: Decisions made → Action items (owner, deadline) → Open questions → Key discussion points → Next meeting date/purpose.

Rollout plan: pilot with 2–3 willing teams → configure templates per meeting type (standup: blockers + updates; planning: decisions + owners; 1:1: topics + follow-ups — never record 1:1s without explicit mutual agreement) → integrate with task tools (action items flow to assignees) → train on review habits (5-min summary check post-meeting) → expand org-wide with documented norms.

Meeting hygiene multiplier: AI notes make bad meetings searchable, not good. Pair rollout with meeting discipline: agendas required, outcomes stated upfront, and a norm that the summary — not attendance — is the deliverable for informational meetings. The combination cuts meeting load genuinely.

Adoption playbook: start with volunteers (not mandates) → showcase wins ("the notes caught an action item we'd forgotten") → expand team by team → document norms (which meetings, who reviews) → measure (time saved, action-item completion rates). Mandated rollouts breed resistance; volunteer-led rollouts breed evangelists. Quality assurance: monthly spot-checks of summary accuracy → custom vocabulary updates → template refinements → feedback channel for users to report errors. AI quality drifts as teams and topics change — QA keeps it sharp.

Common pitfalls

  • No consent process. Recording without clear notice. Legal risk + trust damage. Announce every time.
  • Trusting AI blindly. Acting on hallucinated action items or misattributed decisions. Human review for important meetings.
  • Actions stuck in notes. Extracted but never routed to task systems. Integration or manual transfer — pick one and do it.
  • Recording everything. Sensitive conversations on permanent record. Define exclusions clearly.
  • Ignoring retention. Infinite storage of every meeting. Set retention policies; delete per schedule.
  • Bad meetings with good notes. AI can't fix purposeless meetings. Fix the meeting, then capture it.
  • Access anarchy. Everyone can watch every 1:1 recording. Role-based access and sensible defaults.
  • Recording everything by default. Creating a surveillance culture. Default-off with per-meeting opt-in (or team-level norms) respects autonomy.
  • Never reviewing summaries. Auto-generating notes nobody reads. The value is in the review habit — build it into meeting close-out.
  • Ignoring accuracy limits. Treating AI transcripts as verbatim records for disputes or compliance. They are aids, not evidence — verify critical quotes against recordings.
  • Privacy afterthoughts. Recording without clear consent policies. One privacy incident undoes all productivity gains — get legal and HR aligned first.
  • Replacing human judgment. Auto-filing AI action items without review. AI suggests; humans commit — especially for client-facing promises.

04 同仓库更多 Skills

TanNhatCMS
TanNhatCMS/jarvis

web-scrape

Turn web pages into LLM-ready markdown via the Firecrawl API (renders JS, anti-bot handling). Use when browser_open/fetch fails on JS-heavy pages, or when you need clean markdown/structured extraction from a site. Requires FIRECRAWL_API_KEY.

★ 0
TanNhatCMS
TanNhatCMS/jarvis

design-taste-frontend

tasteskill v2, the anti-slop frontend skill. Art direction for websites, distilled from a study of 247 curated gallery sites (One Page Love, Awwwards Sites of the Day). Deals researched directions (idea, composition, type, palette, photographs, section blueprints) from a library, enforces a strict no-slop house style with a real-browser inspector, and runs a look-and-fix loop on screenshots. Use this whenever you build, design, redesign, restyle, polish or review a landing page, marketing site, portfolio, product page, one-pager, hero or any public-facing web page, in any stack (HTML/CSS, React, Next.js, Vue, Svelte, Astro, Tailwind) — also when the user says a page looks generic, templated, "AI-made", sloppy, or asks to make it look designed, premium, award-level or less like every other site. Not for dashboards or app screens.

★ 0
TanNhatCMS
TanNhatCMS/jarvis

test-driven-development

Use when implementing any feature or bugfix, before writing implementation code

★ 0
TanNhatCMS
TanNhatCMS/jarvis

systematic-debugging

Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes

★ 0
TanNhatCMS
TanNhatCMS/jarvis

requesting-code-review

Use when completing tasks, implementing major features, or before merging to verify work meets requirements

★ 0
TanNhatCMS
TanNhatCMS/jarvis

brainstorming

You MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation).

★ 0