Home / All Skills / Content & Media Creation / meeting-minutes-taker
English (EN)简体中文 (ZH)日本語 (JA)
Content & Media Creation · daymade/claude-code-skills

meeting-minutes-taker

Transforms raw meeting transcripts into high-fidelity, structured meeting minutes (notes / summaries). Use when (1) a meeting transcript is provided and meeting minutes, notes, or a summary are requested; (2) multiple versions of minutes must be merged without losing content; (3) existing minutes need review against the original transcript for missing items; (4) the transcript has anonymous speakers like "Speaker 1/2/3" or "发言人1" that need identifying (optionally mapped via a context.md team directory). Triggers on 会议纪要 / 会议记录 / 整理纪要 / 妙记转纪要, "write meeting minutes", "summarize this meeting", "merge these minutes", "what's missing from these notes". For fixing ASR/STT recognition errors in the raw transcript first, use transcript-fixer; this skill structures clean transcripts into minutes.

L2 Local · Restricted workspaceV1 Commit-PinnedWrite FileDual-pass Audit
Author daymadeGitHub daymade/claude-code-skills ↗Stars 1385License MIT（仓库根 LICENSE 文件；GitHub API spdx MIT；Copyright (c) 2025 daymade）Commit d5c4678cb5
Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

The core mechanism to counter the premise that 'a single pass inevitably loses content' is parallel redundant full read-throughs combined with a UNION merge: three subagents in isolated Task contexts each generate a full set of minutes from the complete transcript, followed by a union merge.

daymade-audio/meeting-minutes-taker/SKILL.md
MUST use the Task tool to spawn multiple subagents with **isolated context**, each generating **complete minutes**
What is happening here: SKILL.md explicitly opposes dividing work by sections ('❌ Narrow-focused passes… wastes tokens, causes bias') and single passes ('A single pass will absolutely lose content.'). The rationale is straightforward: different context states capture different details, and a UNION merge preserves both X and Y. Each pass writes to an independent versionN.md, preventing shared blind spots from contaminating one another.

Filesystem context offloading: Each pass writes its output to <output_dir>/intermediate/<transcript-name>/, and the merge phase reads only the required files, preventing long transcripts from exhausting the conversation context while leaving an auditable trail.

daymade-audio/meeting-minutes-taker/SKILL.md
**Critical: Write each pass output to files, not conversation context.**
Path conventions are <output_dir>/intermediate/<transcript-name>/version1..3.md + retrieval-claims.md. Emphasizes that intermediate/ should be added to .gitignore and must not be deleted after merging (needed for human review and post-hoc debugging).

Retrieval self-test (Step 3.5): A subagent in a fresh context reviews only the transcript (never the drafts) to reverse-extract 'future query claims', tests hit rates against the retrievable layer, conducts revocation scans, and outputs a quantitative report instead of a binary pass/fail.

daymade-audio/meeting-minutes-taker/SKILL.md
4. **Report, never a binary "passed"**. Output: `enumerated N / hits M / promoted K / uncertain list` and show it alongside the draft.
Motivation: The three generation subagents share the same prompt and model of 'what counts as a decision', leading to correlated errors. A UNION merge prevents content loss but cannot guard against shared blind spots. This step directly verifies the downstream property: 'If someone searches for a directive weeks later, will it be found in the Key Decisions / Action Items table?' Upgrading a miss requires a verbatim quote + timestamp along with a revocation scan; uncertainties are moved to Open Questions rather than inflating the decision table.

Speaker identification follows 'source-side labeling priority': If the transcript comes from a platform supporting manual labeling (e.g., Feishu Minutes / Tencent Meeting), the process pauses to have the user label speakers at the source and re-export, treating text-based inference strictly as a fallback.

daymade-audio/meeting-minutes-taker/SKILL.md
**stop and ask the user to label the speakers at the source, then re-export/re-ingest the labeled transcript** before generating minutes
Rationale: Platform labeling is done by humans listening to actual audio (the authoritative source). Text-based inference can only resolve individuals explicitly addressed by name and fails on diarization segments where multiple speakers are collapsed into a single label. Fallback inference (Phase A: feature profiling of word count/segment length/filler word ratio → Phase B: cross-referencing context.md team directory → Phase C) strictly requires evidence, confidence scores, and user confirmation for each mapping, prohibiting silent identity assumptions.

Multi-turn human-AI iteration drives output quality: Initial drafts must be reviewed line-by-line by a human, as terminology conflicts and domain knowledge are known only to humans. Cross-AI outputs (e.g., Gemini/ChatGPT) can be cross-referenced to reduce bias, but must be verified against the transcript before merging.

daymade-audio/meeting-minutes-taker/SKILL.md
Meeting minutes are **not one-shot outputs**. High-quality minutes emerge through multiple review cycles
Merge principle: 'UNION, never remove'. The Step 5 cross-referencing checklist differentiates between 'items we missed (add)' and 'items the other model got wrong (do not copy)'. Output requirements mandate evidenced decisions (verbatim blockquote + speaker per decision), Mermaid diagrams (ER/sequence/flowcharts based on discussion type), and images copied to meeting-media/ embedded at the top.

Intelligent file naming requires user confirmation: Uses YYYY-MM-DD-<topic>-<type>.md, where topic is a 2–4 word theme and type is review/sync/planning/retro/kickoff.

daymade-audio/meeting-minutes-taker/SKILL.md
**Pattern**: `YYYY-MM-DD-<topic>-<type>.md`
The same section explicitly notes: '**Ask user to confirm** the suggested filename before writing.'—acting as a human gate before disk write operations.

2Core Capabilities

01Three-way parallel full read-through + UNION merge generating lossless minutes (3 Task subagents running in parallel in a single message)
02Retrieval self-test: Future query claim extraction + retrievable layer hit testing + revocation scanning (enumerated N / hits M / promoted K)
03Speaker identification: Source-side labeling prioritized + fallback feature profiling + context.md team directory mapping + human confirmation gate
04Completeness self-review (completeness_review_checklist.md) + cross-AI output comparison for bias reduction
05Structured artifacts: Executive Summary / Key Decisions / Action Items / Parking Lot / Open Questions + Mermaid diagrams + meeting-media image embedding
06Evidenced quote discipline: Standalone blockquote formatting + speaker attribution + decision recognition based on structural elements (who + do-what + condition) rather than phrasing
07Smart file naming (YYYY-MM-DD-<topic>-<type>.md) written to disk only after user confirmation
08Upstream integration pipeline: doc-to-markdown (.docx/.pdf to md) + transcript-fixer (ASR error correction) → this skill → pdf-creator/ppt-creator export

3External Dependencies

KindDependency / Endpoint
climarkitdown（可选，.docx→md 转换）
cliTask/sub-agent（宿主能力，非第三方）

4Recommended Governance & Evidence

Suggested Controls for Safe Operation
For Agent Runtime
Restrict working directory
For Administrators
Retain operation logs
Evidence & Triggers (File Paths & Excerpts)
  • 本地工作目录: File read/write and artifact output: Outputs minutes and intermediate files via agent file tools into a user-specified <output_dir> (final filename confirmed with the user beforehand)
Audit Notes: Pure SKILL.md prompt orchestration: contains no scripts, external network calls, or credential access. Local read/write operations are confined to minutes and intermediate files within the user's workspace (with SKILL.md requiring filename confirmation, intermediate/ added to .gitignore, and human approval for final drafts), with no destructive operations like deletion or disk wiping. Subagent parallel generation and Mermaid/image embeddings rely entirely on host capabilities. The only considerations are standard LLM hallucination risks and conversational overhead from multi-step speaker confirmations.

5Second-Pass Verification

  • [ok] 无脚本/无资产面外文件 — glob 全量清单仅 SKILL.md + 3 references；无 tests/evals/scripts/agents。
  • [ok] 无网络外发/无凭证接触 — 目录 token 扫描零命中（URL/curl/API_KEY/secret/subprocess/os.system）；唯一命令示例为可选 markitdown。
  • [ok] 并行 3 完整 pass + UNION 机制描述一致 — Step 2 与 Checklist 均写『3 Task subagents IN PARALLEL』；Anti-Patterns 反向禁止『Sequential subagent execution』与『Divided sections』，正反一致。
  • [ok] Retrieval Self-Test 独立性（抽取器不见 draft） — Step 3.5 原文『Spawne ONE fresh-context subagent that reads ONLY the transcript — it must NOT see the draft』存在。
  • [ok] 发言人识别：源侧优先 + 不静默假设 — Phase 0 ALWAYS TRY FIRST 段落与『CRITICAL: Never silently assume speaker identity.』原文均在。
  • [ok] 文件名确认闸 — 『**Ask user to confirm** the suggested filename before writing.』原文存在。
  • [ok] 引用格式纪律与决策要素识别规则 — blockquote 格式要求、『Decisions are identified by elements, not phrasing』、不确定→Open Questions 均有原文。
  • [ok] 功能声明 vs 夸大检查 — description 四用途逐一对应工作流；『high-fidelity, evidence-based』由证据引语+自测+人审流程支撑；无越界承诺。

6Verdict & Scenarios

  • Systematically solves the 'AI drops content in a single pass' challenge as an engineering problem: redundant full passes + isolated contexts + UNION, with clear rationales for why sectioning or single passes fail.
  • Validates downstream 'retrieval semantics': The Retrieval Self-Test identifies omissions from the perspective of future queries and relies strictly on lexical anchors, uncovering shared blind spots far better than same-context self-reviews.
  • Honest and conservative speaker identification: Prioritizes human source-side labeling, requires evidence and confidence scores with user confirmation for inferences, and never silently assumes identities.
  • Built-in multi-turn human review + cross-AI benchmarking: Treats initial drafts as non-final, defers terminology conflicts and domain knowledge to humans, and enforces transcript verification for external AI outputs.
  • Zero-dependency pure orchestration + file-level context management: Prevents long transcripts from overflowing context, maintains auditable intermediate artifacts, and composes orthogonally with audio and document skills in the same repository.
    • Best for: Best for: Scenarios requiring high-fidelity, searchable meeting minutes (product/tech decision meetings, retrospectives, mid-to-large meetings with multiple speakers and action items); transcripts already cleaned upstream by transcript-fixer that require structured distillation; and teams looking to convert platform transcripts (e.g., Feishu Minutes / Tencent Meeting) into searchable knowledge assets that remain accessible weeks later.
    Not for: Not suitable for: Brief or low-stakes meetings where a single-pass summary is sufficient; poor-quality transcripts that have not been preprocessed by transcript-fixer; workflows expecting a final draft in a single AI pass without human review; real-time live meeting transcription (this skill processes completed transcripts rather than live audio); or use cases demanding verbatim transcripts rather than structured summaries.
    Install Agent-Ready
    ① Verified Mirror 2026-09-09
    Option A · Direct DownloadDownload meeting-minutes-taker.tar.gz
    sha256: 28192c17a4ef62ce…
    Option B · Copy Guide for Agent
    install.json
    Agents automatically download and verify mirrors from this JSON guide.
    ② Upstream GitHub
    GitHub Repo ↗
    Pinned to commit d5c4678cb5.