Jakeschincariol/youtube-agent-skill
Eleven Claude skills that run a YouTube channel: scripts off 21 hook formulas with a scored hook, title and thumbnail linted as one pairing, an edit decision list from your transcript, a retention reader that finds where viewers actually leave, a Shorts cutter, and a virality engine. Free, MIT.
The YouTube Agent Skill – Quick Overview
What it is: A collection of eleven Claude “skills” that help a creator plan, write, edit, and optimise YouTube videos. The tools run locally on plain Python 3 (no extra packages) and are invoked from Claude via commands such as /yt‑script or /yt‑viral. They do not publish anything automatically – they only generate text, scores, or decision lists that the user copies into YouTube.
Key capabilities
| Skill | Main function |
|---|---|
/yt‑script |
Generate a video script from a single idea, suggest five hook lines (using 21 pre‑defined formulas), score each hook, and mark where retention‑boosting beats should appear. |
/yt‑package |
Lint a title‑thumbnail pair for length, duplication and vagueness, returning a single combined package. |
/yt‑edit |
Turn a transcript (.srt) into an edit‑decision list – dead‑air, filler cues, retakes – with timestamps. |
/yt‑comment |
Analyse the comment section, group comments into four categories, draft replies in your voice and suggest which comment to pin. |
/yt‑plan |
Create a weekly production schedule that fits your available hours (one anchor video, one cheap video, three Shorts). |
/yt‑viral |
Analyse public videos in your niche, rank them by how much each outperforms its own channel’s median, and surface the patterns that seem to drive virality. |
/yt‑retention |
Parse a YouTube retention CSV export, pinpoint the exact second viewers drop off and show the spoken line at that moment. |
/yt‑shorts |
Detect short‑form clips hidden inside a long video and write a fresh opening line for each Shorts version. |
/yt‑seo |
Generate a description, useful tags, and three target search queries for a video. |
/yt‑chapters |
Create YouTube‑compatible chapter timestamps from a transcript, validating against YouTube’s formatting rules. |
/yt‑audit |
Run a high‑level audit of an entire channel and output a single most‑impactful fix. |
The six underlying Python tools (run directly from the command line) provide the heavy‑lifting for the above skills:
hookscore.py– 5‑property hook quality panel.title.py– title + thumbnail linter.deadair.py– edit‑decision list generator.chapters.py– chapter validator.retention.py– retention‑export analyser.swipe.py– niche‑viral‑trend detector.
All tools are dependency‑free and work on a clean Python 3 installation.
Installation
- Clone the repo.
- Copy the
yt‑*skill folders into~/.claude/skills/(or the repo’s.claude/skills/). - Optionally add the repo as a Claude plugin via
/plugin marketplace add …. - Populate
~/.claude/youtube/voice.md(or paste aSKILL.mdinto a chat) to give Claude your speaking style.
Limitations & caveats
- No automatic publishing – you must manually upload the generated assets.
- Hook scores are heuristics, not predictive models; a high score is not a guarantee of performance.
/yt‑viralonly reads publicly available data; it never logs into your account.- Tag suggestions are intentionally minimal because tags have limited impact.
License: MIT – free to use, modify, and redistribute.
Who might find this useful?
- Small‑to‑medium YouTube creators who want AI‑assisted assistance without relying on external APIs or paid services.
- Teams that already use Claude and prefer a local, privacy‑preserving workflow.
- Developers looking for a concrete example of how to turn Python scripts into Claude “skills”.
Bottom line: The repo is a genuine, open‑source toolkit that augments the Claude AI assistant with practical, script‑driven utilities for YouTube content creation. It stays in‑scope for AI‑assisted workflow automation while keeping everything runnable locally.
Related
- Project
- Project
- Project
- Project