agentsclimarketplace

Extract youtube creator transcripts

Skill yanyi8171/extract-youtube-creator-transcripts/skills/extract-youtube-creator-transcripts

一键导出指定 YouTube 博主或频道当前公开的长视频、Shorts 和直播回放字幕,保存为带时间轴的 SRT 与清洗后的 Markdown,并支持断点续跑和增量补齐。用户提到“导出博主全部字幕”“YouTube 逐字稿”“批量转 MD”“补齐频道字幕”“更新博主字幕库”“下载 Shorts 字幕”时都应使用本 Skill;不要用于只处理单个本地字幕文件、下载视频画面或绕过会员与地区权限。Export all currently public transcripts from a YouTube creator or channel, including videos, Shorts, and stream replays. Use for bulk transcript export, SRT and Markdown generation, incremental updates, or creator transcript archives.From its SKILL.md

Install
npx -y skills add yanyi8171/extract-youtube-creator-transcripts --skill extract-youtube-creator-transcripts

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 28 days oldThe repository was created 28 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its file declares

Copied from the file, not written here

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

6.4 KB, ~1.8k tokens by cl100k_base, as published. Nobody here has run it

YouTube 博主字幕一键导出

把频道链接或 @handle 和“博主根目录”交给脚本。根目录里只产生 长视频短视频 两个可见目录;每条成功视频保存一份 SRT 和一份同名 Markdown。

开始前

  1. 确认用户给了频道链接/@handle 和保存目录。缺保存目录时只询问这一个信息。
  2. 说明将访问 YouTube;只抓字幕,不下载视频和音频。
  3. 先做工具预检。若需要安装,告诉用户下载项、大小、私有安装位置和风险,得到确认后才执行安装。用户在当前请求中已经明确同意安装时,不重复确认。
  4. 不读取系统浏览器 Cookie,不使用主账号,不修改 PATH 或 shell 配置。

选择启动脚本

<skill> 为本 Skill 目录。

Windows:

powershell -NoProfile -ExecutionPolicy Bypass -File "<skill>\scripts\bootstrap.ps1" --preflight
powershell -NoProfile -ExecutionPolicy Bypass -File "<skill>\scripts\bootstrap.ps1" --install
powershell -NoProfile -ExecutionPolicy Bypass -File "<skill>\scripts\bootstrap.ps1" --channel "<频道>" --output "<博主根目录>" --language source

macOS:

sh "<skill>/scripts/bootstrap.sh" --preflight
sh "<skill>/scripts/bootstrap.sh" --install
sh "<skill>/scripts/bootstrap.sh" --channel "<频道>" --output "<博主根目录>" --language source

预检不会安装或导出。--install 只安装私有 Deno 与 yt-dlp。若脚本返回 21,说明当前网络需要 PO Token Provider;再次说明它会下载并在任务期间启动本地服务,用户确认后运行 --install-pot,再重跑原命令。

参数规则

  • --channel:必填,接受频道 URL、频道 ID URL 或 @handle
  • --output:必填,指向博主根目录,不是 长视频短视频 子目录。
  • --language:默认 source。也可传 enzh-Hans 等语言代码;只选现有字幕,不调用翻译。
  • --cookies:仅在脚本返回 22 后使用,接受 Netscape 格式 cookies.txt
  • --pilot:默认 3;试跑通过后脚本继续完成全部项目。
  • --dry-run:只枚举频道和核对已有文件,不下载字幕。

固定工作流

  1. 先匿名运行,不主动索取 Cookie。
  2. 分别枚举频道的 /videos/shorts/streams;Shorts 以 Shorts 页成员身份为准,直播回放归入长视频。
  3. 人工字幕优先,自动字幕兜底;默认选视频原语言,不静默改用机器翻译轨。
  4. 先处理 3 条试样;之后每批 25 条,每条随机等待 5—10 秒。连续 3 次验证失败就停止,保留已经完成的文件。
  5. 只下载临时 VTT,由脚本自行生成 SRT 与 Markdown;不安装 ffmpeg。
  6. 每次写入前按视频 ID 扫描现有文件:完整配对就跳过,只有 SRT 就补 MD,同一 ID 多份则报告冲突且不删除。
  7. 输出文件通过同目录临时文件原子落盘;不覆盖任何已有文件。

Cookie 只在确有需要时使用

YouTube 官方提取器说明提醒:账号 Cookie 可能导致账号被临时或永久限制。脚本返回 22 时:

  1. 建议使用小号。
  2. 打开一个无痕窗口并登录 YouTube。
  3. 在同一标签页访问 https://www.youtube.com/robots.txt
  4. 使用开源 Get cookies.txt LOCALLY 导出 youtube.com Cookie,然后立刻关闭该无痕窗口。
  5. 把导出路径通过 --cookies 传给脚本。

Cookie 内容和路径不得写进日志、状态或最终回复。脚本不会删除用户导出的原文件;任务结束时提醒用户自行安全删除。

内容使用与版权

  • 只处理用户有权访问的公开字幕;本工具仅用于个人学习、研究,以及依法备份本人有权访问的公开字幕。
  • 未经版权方许可,不得转载、出售、商业化利用他人字幕,也不得用其制作盗版内容。
  • 使用者须遵守 YouTube 条款、版权方要求和适用的著作权法律,并自行承担使用责任。
  • 以上限制针对通过本工具取得的第三方字幕内容;本 Skill 的代码许可证仍为 MIT。

结果与失败处理

脚本结尾输出 RESULT_JSON=...,据此向用户报告:频道视频总数、成功、已有跳过、补配对、失败和报告路径。

失败必须逐条说明:

  • no_captions:视频没有可用字幕。告诉用户可选择本地 Whisper(免费但重)、付费 API 或人工处理;不要自动执行。
  • needs_pot:确认后安装私有 PO Token Provider,再重跑。
  • needs_cookiescookie_expired:按上面的无痕小号流程重新导出。
  • members_or_privateregion_restrictedremoved:不绕过权限,说明需要合法访问权限或视频已不可用。
  • networktool_outdated:先重试;仍失败时建议更新锁定工具链。

详细排查时读取 references/troubleshooting.md

验收

  • 可见输出根目录只有 长视频短视频 两个目录。
  • 每个成功视频 ID 正好一份 .srt 和一份 .md
  • Markdown 不含 -->、字幕序号或连续滚动重复句。
  • 不同视频 ID 即使字幕相同也分别保留;只在报告中标记相同内容。
  • 第二次运行不覆盖、不重复,新增为 0 时明确报告。
  • macOS 只能称为“自动化兼容测试通过”,除非确实在真实 Mac 上运行过。

What ships with it: 12 files

76.8 KB alongside SKILL.md, 7 of them executable

evals/

references/

Gives 0 of the 12 instructions most social media skills give in ~1.8k tokens

Counted across 489 of the 492 authors here whose files we hold, read 2026-08-07

  • Adapt formats and tone to each platformin 26 of 489, across 14 files
  • Build content around three to five pillarsin 25 of 489, across 13 files
  • Read product marketing context before asking questionsin 23 of 489, across 13 files
  • Respond to all comments on your postsin 21 of 489, across 9 files
  • Use the output flag to specify an output directoryin 14 of 489, across 4 files
  • Generate output logo images with white backgroundin 13 of 489, across 4 files
  • Fix failing generation scripts directlyin 13 of 489, across 4 files
  • Ask user about HTML preview after logo generationin 12 of 489, across 3 files
  • Run the download script with a URLin 12 of 489, across 3 files
  • Implement exponential backoff for 429 responsesin 12 of 489, across 3 files
  • Write the hook firstin 12 of 489, across 7 files
  • Include a single clear call to actionin 12 of 489, across 9 files

Said here and by no other author read

  • confirm channel and output directory before starting
  • run preflight checks before installing tools
  • run anonymously without prompting for cookies
  • pause after three consecutive validation failures
  • scan existing files before writing new ones
  • write output files atomically without overwriting

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 326,645. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.