youtube-agent-skill README

The project's README from github.com/Jakeschincariol/youtube-agent-skill. As one markdown file: README.md.

The YouTube agent skill

Eleven Claude skills that run a YouTube channel. Free, MIT, no signup, no API key, nothing to connect.

One of them writes your script off 21 hook formulas and scores the hook before you waste a take on it. One lints the title and the thumbnail as a single pairing, because writing them separately is why half of your click surface says the same thing twice. One reads your audience-retention export and tells you the exact second people left and what you were saying when they did. One turns a transcript into an edit decision list. One finds the Shorts already hiding inside a long video. One goes and finds what is working in your niche and ranks it by how far each video beat its own channel, not by how big the channel is.

Nothing gets published until you do it. These skills write. You upload.

Install

Paste this into Claude:

https://github.com/Jakeschincariol/youtube-agent-skill

Install this skill, then confirm /yt-script works.

Or do it yourself, in Claude Code:

git clone https://github.com/Jakeschincariol/youtube-agent-skill.git
cp -r youtube-agent-skill/skills/yt-* ~/.claude/skills/

Or as a plugin:

/plugin marketplace add Jakeschincariol/youtube-agent-skill
/plugin install youtube-agent

Project-local instead of global: copy the same folders into your repo's .claude/skills/. No Claude Code at all? Paste any single SKILL.md at the top of a chat and it runs as a mode. You lose the six Python tools, which is most of the point of /yt-script, /yt-retention and /yt-edit, but the rest works.

Then spend ten minutes on templates/voice.md. Copy it to ~/.claude/youtube/voice.md and fill it in, or send Claude three of your own videos and say "write my voice.md from these". Every skill reads that file. It matters more here than anywhere else, because you have to say the words out loud.

The eleven

command what it does
/yt-script One idea into a script. Five hooks off 21 formulas, scored, then the spoken script with the retention beats marked.
/yt-package Title and thumbnail as one pairing, linted for truncation, duplication and vagueness.
/yt-edit A transcript into an edit decision list: dead air, filler cues, retakes, with timecodes.
/yt-comment The comment section triaged into four piles, then replies in your voice. Says which one to pin.
/yt-plan A week that fits the hours you actually have. One anchor, one cheap one, three Shorts.
/yt-viral What is working in your niche, ranked by multiple over each channel's own median.
/yt-retention Your retention export read properly: hook leak, the cliffs, the slide, and what to change.
/yt-shorts The Shorts already inside a long video, with a new first line written for each.
/yt-seo The description, the tags that are worth having, and the three queries this should win.
/yt-chapters Chapters from a transcript, validated against YouTube's own rules so they render.
/yt-audit The whole channel, ending in ONE fix rather than twenty.

The six tools

Every one of these runs on a clean Python 3 with no dependencies. They are the reason the skills are not just prompts.

python3 skills/yt-script/hookscore.py --hook "one line"        # 5-property hook panel
python3 skills/yt-package/title.py --title "..." --thumb "..." # title + thumbnail linter
python3 skills/yt-edit/deadair.py transcript.srt               # edit decision list
python3 skills/yt-chapters/chapters.py transcript.srt          # validated chapters
python3 skills/yt-retention/retention.py retention.csv         # where they left, and why
python3 skills/yt-viral/swipe.py collected.json --min 2.0      # outliers by own-channel multiple

The fine print

The part most packs leave out.

It does not publish. YouTube's Data API would allow it with your own OAuth, which makes this pack different from the LinkedIn and Instagram ones, where posting is not available at all. It is still not built here: every skill ends in a block you copy and a question — ship it, or change it? Anything that claims to run your channel unattended should be read with that distinction in mind.

hookscore.py is a heuristic, not a predictor. The panel was built and calibrated against 74 real short-form hooks (the first 15 seconds of auto-captions, top-8 and bottom-8 by views across five channels). It separates deliberately bad hooks from real ones well. It separates a given creator's hits from their own misses barely at all. A low score is a reason to look again; a high score is not a promise.

The formula classifier is about the words, not the result. When /yt-viral says a title used The Statistic, that is a judgement about the title you can see, not a claim about why the video got its views.

/yt-viral reads, it does not scrape. Public listings only. It never logs in as you and never touches your credentials.

Tags barely matter and this pack says so instead of selling you a tag generator.

Nothing invents a number. If a skill wants a figure it does not have, it asks you for it or writes the line without it.

Licence

MIT. Use it, change it, ship it.

youtube-agent-skill (Jakeschincariol/youtube-agent-skill)

Searching Database…

1.6K ·GitHub
Youtube-agent-skill
这是什么？一套包含十一个 Claude Skills 的技能包，处理运营 YouTube 频道的日常工作：根据钩子公式写脚本、检查标题/缩略图搭配、读取留存率导出数据、构建剪辑决策列表、从长视频中截取 Shorts，以及评估你的细分领域中什么内容有效。它免费、采用 MIT 许可，无需注册或 API 密钥。
这只是一套 AI 提示词，还是包含真正的工具？两者都是：每个技能都是一个供 Claude 读取的 SKILL.md，但其中六个还附带小的 Python 3 脚本（比如钩子评分器），没有额外依赖，这正是它们不只是提示词的原因。
怎么安装这个技能包？把它粘贴到 Claude 对话中、在 Claude Code 里安装，或者作为插件添加；如果只想在单个项目中使用，把同样的技能文件夹复制到你仓库的 .claude/skills/ 目录即可，不必全局安装。
使用它需要 Claude Code 或 API 密钥吗？完全不需要注册或 API 密钥。运行 /yt-script 和 /yt-retention 等命令背后的六个 Python 工具需要 Claude Code；没有它，你仍然可以把单个 SKILL.md 粘贴到任何 Claude 聊天中，作为纯提示词模式使用，只是没有那些由脚本驱动的命令。
它真的能帮我发布视频吗？不能——任何东西都不会自动发布。它不会碰 YouTube 的发布 API；那需要你自己配置 OAuth，而这个技能包不做这件事，所以每个技能最终都只产出一份由你亲自批准的草稿。
/yt-viral 会登录其他频道来抓取数据吗？不会——/yt-viral 只读取你所在细分领域的公开视频列表；它绝不会以你的身份登录，也绝不会碰你的凭据。
怎么让这些技能用我自己的风格写作？手动填写 templates/voice.md，或者把你自己的三个视频发给 Claude，让它替你起草这个文件，然后保存到 ~/.claude/youtube/voice.md——每个技能在写作之前都会读取该文件。
/yt-script 给出的钩子评分有多可靠？hookscore.py 只是用 74 个真实短视频钩子校准过的启发式方法，所以它能可靠地标记明显偏弱的钩子，但几乎分不清创作者自己的爆款和自己的失败之作——高分只能当作一种可能，而不是保证。
这十一个 /yt- 命令都覆盖什么？脚本创作（/yt-script）、标题/缩略图检查（/yt-package）、剪辑决策列表（/yt-edit）、评论分诊（/yt-comment）、每周规划（/yt-plan）、细分领域爆款排名（/yt-viral）、留存率分析（/yt-retention）、Shorts 截取（/yt-shorts）、SEO（/yt-seo）、章节生成（/yt-chapters），以及只给一条修改建议的频道体检（/yt-audit）。
频道体检会不会给我甩出一大堆修改建议？不会——/yt-audit 的设计是最终只给出一条针对整个频道的推荐修改，而不是二十条零散的建议。
What are similar projects?linkedin-agent-skill, ai-marketing-skills, marketing-agi, marketingskills, open-steps, guizang-product-video-skill, herk-edit-kit, instagram-agent-skill
Which topics is it in?seo, social-media, claude-skill, video-editing, claude-code-plugin, youtube, content-creation, agent-skill, skill, video, skills, claude
Creator
Jakeschincariol
Size
<0.1 MB
Stars
1.6K
Updated
CommitSep 16
Age
1mo
Commits · 90 Days