--- name: meeting-minutes-taker disable-model-invocation: true description: >- Transforms raw meeting transcripts into structured, evidence-based meeting minutes — merges multiple versions without losing content, reviews existing minutes against the transcript for missing items, and identifies anonymous speakers (Speaker 1/2/3, 发言人1). Triggers on 会议纪要, 会议记录, 整理纪要, 妙记转纪要, "write meeting minutes", "merge these minutes". Not for fixing ASR/STT errors first (use transcript-fixer). --- # Meeting Minutes Taker Transform raw meeting transcripts into comprehensive, evidence-based meeting minutes through iterative review. ## Quick Start **Pre-processing (Optional but Recommended):** - **Document conversion**: Use `doc-to-markdown` skill to convert .docx/.pdf to Markdown first (preserves tables/images) - **Transcript cleanup**: Use `transcript-fixer` skill to fix ASR/STT errors if transcript quality is poor - **Context file**: Prepare `context.md` with team directory for accurate speaker identification **Core Workflow:** 1. Read the transcript and any supplied participant notes or user corrections using Step 1's source contract 2. Load project-specific context file if provided by user (optional) 3. **Intelligent file naming**: Auto-generate filename from content (see below) 4. **Speaker identification**: If transcript has "Speaker 1/2/3", FIRST ask the user to label speakers on the source platform and re-export (see Step 1.5 Phase 0); infer from text only as a fallback 5. **Multi-turn generation**: Use multiple passes or subagents with isolated context, merge using UNION 6. Self-review using [references/completeness_review_checklist.md](references/completeness_review_checklist.md) 7. Present draft to user for human line-by-line review 8. **Cross-AI comparison** (optional): Human may provide output from other AI tools (e.g., Gemini, ChatGPT) - merge to reduce bias 9. Iterate on feedback until human approves final version ### Intelligent File Naming Auto-generate output filename from transcript content: **Pattern**: `YYYY-MM-DD--.md` | Component | Source | Examples | |-----------|--------|----------| | Date | Transcript metadata or first date mention | `2026-01-25` | | Topic | Main discussion subject (2-4 words, kebab-case) | `api-design`, `product-roadmap` | | Type | Meeting category | `review`, `sync`, `planning`, `retro`, `kickoff` | **Examples:** - `2026-01-25-order-api-design-review.md` - `2026-01-20-q1-sprint-planning.md` - `2026-01-18-onboarding-flow-sync.md` **Ask user to confirm** the suggested filename before writing. ## Core Workflow Copy this checklist and track progress: ``` Meeting Minutes Progress: - [ ] Step 0 (Optional): Pre-process transcript with transcript-fixer - [ ] Step 1: Read and analyze transcript - [ ] Step 1.5: Speaker identification (if transcript has "Speaker 1/2/3") - [ ] Phase 0 FIRST: ask user to label speakers on the source platform (Feishu Minutes / Tencent Meeting), re-export, use labeled transcript - [ ] Fallback only (source labeling unavailable or declined by user): - [ ] Analyze speaker features (word count, style, topic focus) - [ ] Match against context.md team directory (if provided) - [ ] Present speaker mapping with per-speaker evidence to user for confirmation - [ ] Step 1.6: Generate intelligent filename, confirm with user - [ ] Step 1.7: Quality assessment (optional, affects processing depth) - [ ] Step 2: Multi-turn generation (PARALLEL subagents with Task tool) - [ ] Create transcript-specific dir: /intermediate// - [ ] Launch 3 Task subagents IN PARALLEL (single message, 3 Task tool calls) - [ ] Subagent 1 → /intermediate//version1.md - [ ] Subagent 2 → /intermediate//version2.md - [ ] Subagent 3 → /intermediate//version3.md - [ ] Merge: UNION all versions, AGGRESSIVELY include ALL diagrams → draft_minutes.md - [ ] Final: Compare draft against all supplied sources, reconcile final states, add omissions - [ ] Step 3: Self-review for completeness - [ ] Step 3.5: Retrieval Self-Test (consumption-side verification) - [ ] Follow [Step 3.5](#step-35-retrieval-self-test-consumption-side-verification) and save its evidence - [ ] Step 4: Present draft to user for human review - [ ] Step 5: Cross-AI comparison (if human provides external AI output) - [ ] Step 6: Iterate on human feedback (expect multiple rounds) - [ ] Step 7: Human approves final version Note: = directory where final meeting minutes will be saved (e.g., project-docs/meeting-minutes/) Note: = name derived from transcript file (e.g., 2026-01-15-product-api-design) ``` ### Step 1: Read and Analyze Transcript Read all supplied sources before generation. Keep the transcript unchanged; record participant notes and user corrections separately with their author, date and source locator when known. Unknown metadata stays unknown. Give every generation pass the same source bundle. Notes need no separate reader-facing deliverable: integrate their substantive content into one current minutes document, link the transcript for verification, and retain original notes through the owning archive's source record. Follow an explicitly requested alternative delivery format. Reconcile each fact within its stated scope. A later explicit participant correction or status update can revise that fact; it does not silently replace unrelated meeting decisions. If sources remain irreconcilable, state the conflict in Open Questions. Do not invent agreement. Significant transcript-backed decisions need original quotations with speaker attribution and timestamps. Record note-only additions under **Source** with the note's author/date or source label; do not fabricate a transcript quote, speaker or timestamp for them. Bind each TODO/DONE to its exact action, owner, conditions and supporting source. Preserve unknown owners, deadlines and completion evidence as unknown. A note's author is not automatically the task owner. An invitation sent does not establish that the meeting occurred; finding an available account does not establish that someone switched accounts. Separate agreed responsibility from work delivered or a hypothesis verified. Analyze the transcript to identify: - Meeting topic and attendees - Key decisions with supporting quotes - Action items with owners - Deferred items / open questions ### Step 1.5: Speaker Identification (When Needed) **Trigger**: Transcript only has generic labels like "Speaker 1", "Speaker 2", "发言人1", etc. #### Phase 0: Source-Side Labeling (ALWAYS TRY FIRST) When the transcript comes from a platform that supports manual speaker labeling (Feishu Minutes 飞书妙记, Tencent Meeting 腾讯会议, or any tool with a diarization-editing page), **stop and ask the user to label the speakers at the source, then re-export/re-ingest the labeled transcript** before generating minutes. Send the user the source page link — it is usually in the transcript's frontmatter (`minute_url`, meeting URL). Why this beats inference: - Platform labeling is a human listening to the actual voices — the authoritative source. Text-based inference can only resolve speakers who happen to get name-called during the meeting; everyone else stays a guess. - Diarization-merged segments (multiple people collapsed into one label) are unrecoverable from text alone — no amount of inference fixes them, but source-side relabeling does. - Inference output forces `[inferred]` markers everywhere plus a per-speaker human review round; source labeling produces clean ground truth once. Fall back to Phase A–C below **only when**: (a) the user explicitly says to proceed by inference, or (b) the source cannot be labeled (raw audio file with no platform page, no edit permission). In the fallback, every mapping must carry evidence and a confidence level, unresolved labels stay as-is (never force-assign), and when the user later labels the source, go back and correct the minutes. **Fallback approach** (inspired by Anker Skill): #### Phase A: Feature Analysis (Pattern Recognition) For each speaker, analyze: | Feature | What to Look For | |---------|-----------------| | **Word count** | Total words spoken (high = senior/lead, low = observer) | | **Segment count** | Number of times they speak (frequent = active participant) | | **Avg segment length** | Average words per turn (long = presenter, short = responder) | | **Filler ratio** | % of filler words (对/嗯/啊/就是/然后) - low = prepared speaker | | **Speaking style** | Formal/informal, technical depth, decision authority | | **Topic focus** | Areas they discuss most (backend, frontend, product, etc.) | | **Interaction pattern** | Do others ask them questions? Do they assign tasks? | **Example analysis output:** ``` Speaker Analysis: ┌──────────┬────────┬──────────┬─────────────┬─────────────┬────────────────────────┐ │ Speaker │ Words │ Segments │ Avg Length │ Filler % │ Role Guess │ ├──────────┼────────┼──────────┼─────────────┼─────────────┼────────────────────────┤ │ 发言人1 │ 41,736 │ 93 │ 449 chars │ 3.6% │ 主讲人 (99% of content)│ │ 发言人2 │ 101 │ 8 │ 13 chars │ 4.0% │ 对话者 (short responses)│ └──────────┴────────┴──────────┴─────────────┴─────────────┴────────────────────────┘ Inference rules: - 占比 > 70% + 平均长度 > 100字 → 主讲人 - 平均长度 < 50字 → 对话者/响应者 - 语气词占比 < 5% → 正式/准备充分 - 语气词占比 > 10% → 非正式/即兴发言 ``` #### Phase B: Context Mapping (If Context File Provided) When user provides a project context file (e.g., `context.md`): 1. Load team directory section 2. Match feature patterns to known team members 3. Cross-reference roles with speaking patterns **Context file should include:** ```markdown ## Team Directory | Name | Role | Communication Style | |------|------|---------------------| | Alice | Backend Lead | Technical, decisive, assigns backend tasks | | Bob | PM | Product-focused, asks requirements questions | | Carol | TPM | Process-focused, tracks timeline/resources | ``` #### Phase C: Confirmation Before Proceeding **CRITICAL**: Never silently assume speaker identity. Present analysis summary to user: ``` Speaker Analysis: - Speaker 1 → Alice (Backend Lead) - 80% confidence based on: technical focus, task assignment pattern - Speaker 2 → Bob (PM) - 75% confidence based on: product questions, requirements discussion - Speaker 3 → Carol (TPM) - 70% confidence based on: timeline concerns, resource tracking Please confirm or correct these mappings before I proceed. ``` After user confirmation, apply mappings consistently throughout the document. ### Step 1.7: Transcript Quality Assessment (Optional) Evaluate transcript quality to determine processing depth: **Scoring Criteria (1-10 scale):** | Factor | Score Impact | |--------|-------------| | **Content volume** | >10k chars: +2, 5-10k: +1, <2k: cap at 3 | | **Filler word ratio** | <5%: +2, 5-10%: +1, >10%: -1 | | **Speaker clarity** | Main speaker >80%: +1 (clear presenter) | | **Technical depth** | High technical content: +1 | **Quality Tiers:** | Score | Tier | Processing Approach | |-------|------|---------------------| | ≥8 | **High** | Full structured minutes with all sections, diagrams, quotes | | 5-7 | **Medium** | Standard minutes, focus on key decisions and action items | | <5 | **Low** | Summary only - brief highlights, skip detailed transcription | **Example assessment:** ``` 📊 Transcript Quality Assessment: - Content: 41,837 chars (+2) - Filler ratio: 3.6% (+2) - Main speaker: 99% (+1) - Technical depth: High (+1) → Quality Score: 10/10 (High) → Recommended: Full structured minutes with diagrams ``` **User decision point**: If quality is Low (<5), ask user: > "Transcript quality is low (碎片对话/噪音较多). Generate full minutes or summary only?" ### Step 2: Multi-Turn Initial Generation (Critical) **A single pass will absolutely lose content.** Use multi-turn generation with **redundant complete passes**: #### Core Principle: Multiple Complete Passes + UNION Merge Each pass generates **COMPLETE minutes (all sections)** from the full transcript. Multiple passes with isolated context catch different details. UNION merge consolidates all findings. Apply source reconciliation from Step 1 before every merge, including cross-AI merges and human-feedback rounds. UNION preserves source-supported content; it does not turn superseded proposals into simultaneous active commitments. Scan all supplied sources for later reversals, modifications and conditions. Put the final effective state in Decisions / Action Items, retain useful superseded alternatives in Discussion, and route unresolved conflicts to Open Questions. Apply the same state check to diagrams. Consolidate duplicate content, retain the more detailed source-supported treatment when versions differ in depth, and maintain logical section numbering. **❌ WRONG: Narrow-focused passes** (wastes tokens, causes bias) ``` Pass 1: Only extract decisions Pass 2: Only extract action items Pass 3: Only extract discussion ``` **✅ CORRECT: Complete passes with isolated context** ``` Pass 1: Generate COMPLETE minutes (all sections) → version1.md Pass 2: Generate COMPLETE minutes (all sections) with fresh context → version2.md Pass 3: Generate COMPLETE minutes (all sections) with fresh context → version3.md Merge: UNION all versions, consolidate duplicates → draft_minutes.md ``` #### Strategy A: Sequential Multi-Pass (Complete Minutes Each Pass) ``` Pass 1: Read transcript → Generate complete minutes → Write to: /intermediate/version1.md Pass 2: Fresh context → Read transcript → Generate complete minutes → Write to: /intermediate/version2.md Pass 3: Fresh context → Read transcript → Generate complete minutes → Write to: /intermediate/version3.md Merge: Read all versions → UNION merge (consolidate duplicates) → Write to: draft_minutes.md Final: Compare draft against transcript → Add any remaining omissions → final_minutes.md ``` #### Strategy B: Parallel Multi-Agent (Complete Minutes Each Agent) - PREFERRED **MUST use the Task tool** to spawn multiple subagents with **isolated context**, each generating **complete minutes**: **Implementation using Task tool:** ``` // Launch ALL 3 subagents in PARALLEL (single message, multiple Task tool calls) Task(subagent_type="general-purpose", prompt="Generate complete meeting minutes from transcript...", run_in_background=false) → version1.md Task(subagent_type="general-purpose", prompt="Generate complete meeting minutes from transcript...", run_in_background=false) → version2.md Task(subagent_type="general-purpose", prompt="Generate complete meeting minutes from transcript...", run_in_background=false) → version3.md // After all complete: Main Agent: Read all versions → UNION merge, consolidate duplicates → draft_minutes.md ``` **CRITICAL: Subagent Prompt Must Include:** 1. Full path to transcript file - Include every supplied participant note/correction path with its source label, author/date when known, and Step 1's reconciliation and status rules; each pass receives the same bundle. 2. Full path to output file (version1.md, version2.md, version3.md in transcript-specific subdirectory) 3. Context files to load (project-specific context if provided, meeting_minutes_template.md) 4. Reference images/documents if provided by user 5. Output language requirement (match user's language preference, preserve technical terms in English) 6. Quote formatting requirement (see Quote Formatting Requirements section below) 7. Decision-recognition rule: identify decisions by **elements, not phrasing** — an instruction embedded in narrative/case-review speech is still a decision if it has an authority (who), an action (do what), and optionally a condition (until when). Precision guardrail: when unsure whether something is a real decision, route it to Open Questions instead of Key Decisions (never inflate the decision table). **Why multiple complete passes work:** - Each pass independently analyzes the SAME content - Different context states catch different details (no single pass catches everything) - Pass 1 might catch decision X but miss action item Y - Pass 2 might catch action item Y but miss decision X - UNION merge captures both X and Y **Why isolated context matters:** - Each pass/agent starts fresh without prior assumptions - No cross-contamination between passes - Different "perspectives" emerge naturally from context isolation #### Progressive Context Offloading (Use File System) **Critical: Write each pass output to files, not conversation context.** **Path Convention:** All intermediate files should be created in a **transcript-specific subdirectory** under `/intermediate/` to avoid conflicts between different transcripts being processed. **CRITICAL: Use transcript-specific subdirectory structure:** ``` /intermediate//version1.md /intermediate//version2.md /intermediate//version3.md ``` Example: If final minutes will be `project-docs/meeting-minutes/2026-01-14-api-design.md`, then: - Intermediate files: `project-docs/meeting-minutes/intermediate/2026-01-14-api-design/version1.md` - This prevents conflicts when multiple transcripts are processed in the same session - The `intermediate/` folder should be added to `.gitignore` (temporary working files) ``` // Create transcript-specific subdirectory first mkdir: /intermediate// // Launch all 3 subagents IN PARALLEL (must be single message with 3 Task tool calls) Task 1 → Write to: /intermediate//version1.md (complete minutes) Task 2 → Write to: /intermediate//version2.md (complete minutes) Task 3 → Write to: /intermediate//version3.md (complete minutes) Merge Phase: Read: /intermediate//version1.md Read: /intermediate//version2.md Read: /intermediate//version3.md → UNION merge, consolidate duplicates, INCLUDE ALL DIAGRAMS → Write to: draft_minutes.md Final Review: Read: draft_minutes.md Read: original_transcript.md + any supplied participant notes/corrections → Compare & add omissions → Write to: final_minutes.md ``` **Benefits of file-based context offloading:** - Conversation context stays clean (avoids token overflow) - Intermediate results persist (can be re-read if needed) - Each pass starts with fresh context window - Merge phase reads only what it needs - **Human can inspect intermediate files for review** - Critical for understanding what each pass captured - Supports very long transcripts that exceed context limits - **Enables post-hoc debugging** - If final output is missing something, human can trace which pass missed it **IMPORTANT: Always preserve intermediate versions in transcript-specific subdirectory:** - `/intermediate//version1.md`, `version2.md`, `version3.md` - Each subagent output - These files help human reviewers understand the merge process - Do NOT delete intermediate files after merge - Human may want to compare intermediate versions to understand coverage gaps - **Add `intermediate/` to `.gitignore`** - These are temporary working files, not final deliverables - **Transcript-specific subdirectory** prevents conflicts when processing multiple transcripts #### Output Requirements - **Chinese output** with English technical terms preserved - **Source evidence** - Apply the [source and status contract](#step-1-read-and-analyze-transcript) - **Structured sections** - Executive Summary, Key Decisions, Discussion, Action Items, Parking Lot - **Proper quote formatting** - See Quote Formatting Requirements section below - **Mermaid diagrams** (STRONG