yt-audit

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

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.