Agent Skills › Illyism/transcribe-cli › summarize-transcript

summarize-transcript

GitHub

将原始SRT字幕或转录文本转化为高质量摘要，包括执行总结、关键要点和结构化笔记。支持会议记录、教育内容提取及快速速读模式，严格保留时间戳并基于原文事实生成。

skills/summarize-transcript/SKILL.md Illyism/transcribe-cli

Trigger Scenarios

用户要求总结视频内容从转录文本中提取主要观点生成会议或讲座笔记

Install

npx skills add Illyism/transcribe-cli --skill summarize-transcript -g -y

More Options

Use without installing

npx skills use Illyism/transcribe-cli@summarize-transcript

指定 Agent (Claude Code)

npx skills add Illyism/transcribe-cli --skill summarize-transcript -a claude-code -g -y

安装 repo 全部 skill

npx skills add Illyism/transcribe-cli --all -g -y

预览 repo 内 skill

npx skills add Illyism/transcribe-cli --list

SKILL.md

Frontmatter

{
    "name": "summarize-transcript",
    "description": "Transform raw SRT subtitle files generated by @illyism\/transcribe into clean, actionable executive summaries, key takeaway bullet points, and structured notes. Use whenever the user asks to summarize a video, extract main points, or generate meeting\/lecture notes from a transcript."
}

Summarize a Transcript

Transform raw .srt subtitle files or transcription text into high-value final summaries.

Process Flowchart

Walk this decision procedure to produce the exact summary format required:

What is the primary objective of the summary?
├── Executive / Meeting Notes (e.g. Zoom call, lecture, internal sync)
│   └── Output Structure:
│       1. Executive Summary (2-3 sentences)
│       2. Key Decisions & Action Items (bulleted, assigned if names present)
│       3. Main Discussion Topics (timestamped headers [HH:MM:SS])
├── Educational / Content Takeaways (e.g. YouTube video, podcast episode)
│   └── Output Structure:
│       1. Core Thesis / Hook
│       2. 3-5 Key Insights (with timestamps)
│       3. Notable Quotes (exact verbatim quotes with timestamps)
└── Quick Skim / TL;DR
    └── Output Structure:
        1. TL;DR (1 sentence)
        2. 5 Bullet Points (max 15 words per bullet)

Processing Rules

Strip SRT Boilerplate: Ignore line numbers (1, 2, ...) and timing arrows (00:00:01,000 --> 00:00:04,500) when reading the file. Focus entirely on the text content.
Preserve Timestamps for Anchor Points: When referencing key moments or quotes, compute the start timestamp from the SRT block and format as [MM:SS] (or [HH:MM:SS] if duration > 1 hour).
No Speculation: Summarize ONLY facts and statements present in the transcript text. Do not invent context or fill in external information.
Verbatim Quote Accuracy: When highlighting quotes, match the exact transcribed words—never paraphrase quoted text.

Example Output (Educational / Content)

# Summary: How to Improve Video Watch Time

**Core Thesis**: Video hooks are overrated; watch time retention depends on pacing, visual cuts, and immediate value delivery within the first 10 seconds.

### Key Insights
- **[00:04] Hook Myth**: Users watch the first 3-4 seconds even on low-effort videos; the drop-off happens at seconds 5-15.
- **[00:45] Pacing Rule**: Change visual element or camera angle every 2.5 seconds to maintain visual engagement.

### Key Quote
> "Hooks don't matter. Go and pull up your shittiest video and look at the watch time graph." — [00:01]