| name | meeting-notes |
| description | Summarize and organize a transcript (meeting, lecture, interview, etc.) into a structured .md file. Use when the user provides a transcript or text to organize. User can specify the title. |
| disable-model-invocation | true |
| argument-hint | [title] [transcript text] |
Transcript Summarization Skill
When the user provides a transcript (meeting, lecture, interview, presentation, etc.), extract key information and organize it into a structured markdown document.
Input
The user passes text via $ARGUMENTS. The first line or quoted phrase may be the title. If no clear title is given, infer one from the content. If $ARGUMENTS is empty or too short, ask the user to paste the transcript.
Output Rules
1. Filename Convention
- Format:
YYYY-MM-DD-topic-keywords.md
- Date: Use today's date unless the user specifies otherwise
- Topic: Use the user-specified title, or extract 2-4 core keywords from the content, joined by hyphens
- Korean is allowed in keywords
- Examples:
2026-03-11-도메인-제품-온보딩-1차.md, 2026-03-15-결제-시스템-개선-논의.md
2. Save Location
- Save to the current working directory
3. Document Structure
Use the following structure as a base, adapting flexibly to fit the content:
# [Title]
> 날짜: YYYY-MM-DD
## 요약
- 전체 내용을 2~4문장으로 핵심 요약
---
## 1. [Topic Section]
- Key points in logical flow
- Important context and background included
## 2. [Next Topic Section]
...
---
## 결정 사항
- Confirmed decisions (if any)
## 액션 아이템
- [ ] 담당자: 할 일 (if any)
## 미결 사항 / 추가 논의 필요
- Unresolved items (if any)
4. Core Principles
- Context-driven organization: Don't just list points — understand the overall flow and structure logically by topic and narrative arc.
- Detailed yet accurate: Include all important content without omission, but remove chatter, repetition, and fillers. When omitting non-essential content, verify that context is not damaged.
- Faithful to source: Do not distort speakers' intent or nuance. Mark areas where inference was necessary.
- Term correction: Fix speech-recognition errors using context clues (e.g., "엠에스에이" → MSA, "에이피아이" → API).
- Structurize: Split into topic-based sections. Make cause-effect, comparisons, and relationships explicit.
- Bold emphasis: Highlight key concepts, important numbers, and decisions with bold.
- Use tables: Comparisons, classifications, and conditional handling are best presented as tables.
- Speaker attribution: When distinguishable, note who said what.
- Decisions vs Open: Clearly separate decided items from unresolved items.
- Use dividers: Separate sections visually with
--- horizontal rules.
- Number sections: Number each section for easy reference.
5. Section Flexibility
- Omit sections (결정 사항, 액션 아이템, 미결 사항) if they don't apply to the content.
- Add sections as needed based on the content's nature (e.g., Q&A, key takeaways, references).
Execution Flow
- If the user provides a title, use it. Otherwise infer from content.
- Analyze the transcript to identify topics and key content.
- Generate an appropriate filename (date + title/keywords).
- Write the markdown document following the rules above.
- Save the file and inform the user of the file path with a brief summary.
Important
- The output document should be written in the same language as the transcript (typically Korean).
- This skill instruction is in English for token efficiency, but the output must match the input language.
