TAU-HOME.COM
LOADING

Claude-Based YouTube Channel Automation Toolkit 'youtube-agent-skill'

An open-source MIT-licensed repository combining 11 Claude skills and 6 zero-dependency Python 3 scripts for YouTube scripting, title-thumbnail linting, transcr

tau · November 15, 2026

#claude-code #youtube-automation #dev-tools #open-source

Claude-Based YouTube Channel Automation Toolkit 'youtube-agent-skill'

Jakeschincariol/youtube-agent-skill is an open-source collection of 11 modular Claude skills designed to structure the entire YouTube production lifecycle, from ideation and hook scoring to transcript-based edit decision lists and retention diagnostics. Released under a free MIT license, the toolkit requires no account signup, third-party SaaS subscription, or external API keys. According to GitHub repository metadata, the project was created on September 16, 2026, and had reached 5,797 stars and 1,663 forks as of its November 13, 2026 update.

Workflow overview of youtube-agent-skill running 11 Claude skills and 6 Python tools for YouTube channel production

Image source: https://x.com/dopamine_crypto/status/2118644566987526434

Rather than relying purely on open-ended LLM prompts, the repository pairs markdown skill definitions (SKILL.md) with six standalone Python 3 scripts that run on the standard library with zero external dependencies. Deterministic tasks—such as scoring opening hooks, detecting overlapping title-and-thumbnail copy, or parsing timestamps from subtitle exports—are handled by local Python utilities, while Claude handles contextual drafting and creative iteration.

Eleven Modular Skills and Six Zero-Dependency Python Tools

The repository divides channel management into 11 slash commands (skills/yt-*). Several commands invoke companion Python scripts under the hood to validate metrics before generating recommendations.

  • Planning and Scriptwriting (/yt-plan, /yt-script): /yt-plan builds a weekly production schedule tailored to the creator's available hours, structured around one anchor video, one low-lift video, and three Shorts. /yt-script generates five opening hook candidates based on 21 hook formulas (hooks.json), evaluates their strength across five properties using hookscore.py, and drafts a spoken script annotated with retention beats.
  • Packaging and Search Optimization (/yt-package, /yt-seo, /yt-chapters): /yt-package audits the video title and thumbnail text together as a single click surface via title.py, flagging truncation, vague phrasing, and redundant wording where the title and thumbnail repeat the same message. /yt-seo generates video descriptions, relevant tags, and three target search queries, while /yt-chapters runs chapters.py on a subtitle file (transcript.srt) to produce validated YouTube chapter timestamps.
  • Post-Production and Shorts Extraction (/yt-edit, /yt-shorts): Feeding a transcript into /yt-edit triggers deadair.py, which compiles an edit decision list (EDL) pinpointing dead air, filler cues, and retakes with exact timecodes. /yt-shorts scans long-form transcripts to identify self-contained short-form segments and writes fresh opening lines for each clip.
  • Retention Diagnostics and Channel Audits (/yt-retention, /yt-viral, /yt-audit, /yt-comment): /yt-retention parses YouTube Studio audience-retention exports (retention.csv) through retention.py to isolate exact drop-off timestamps and the spoken lines immediately preceding them. /yt-viral uses swipe.py to rank public niche videos by how many multiples they outperformed their own channel's median views (--min 2.0) rather than raw subscriber count. /yt-audit reviews the broader channel to recommend one single highest-leverage fix, and /yt-comment triages viewer comments into four categories with drafted replies and a recommended pinned comment.

Installation and Configuring the Channel Voice Profile (voice.md)

The toolkit is built primarily for Claude Code and requires only a standard Python 3 installation—no pip install or package manager setup is needed.

  1. Install via Claude Code Plugin Marketplace: Run /plugin marketplace add Jakeschincariol/youtube-agent-skill followed by /plugin install youtube-agent inside Claude Code.
  2. Manual Directory Installation: Clone the repository (git clone https://github.com/Jakeschincariol/youtube-agent-skill.git) and copy the skills/yt-* directories into your global ~/.claude/skills/ folder or a project-local .claude/skills/ directory.
  3. Set Up voice.md: Copy templates/voice.md to ~/.claude/youtube/voice.md and populate it with your channel's tone, vocabulary, and delivery preferences. All 11 skills read this file so generated scripts and replies match how the creator actually speaks on camera. Creators can also share three existing video transcripts with Claude and ask it to draft their initial voice.md.

Operational Caveats and Verified Limitations

The repository documentation explicitly outlines what the toolkit does not automate and where its analytical heuristics stop.

  • No Automated Publishing: The package does not connect to the YouTube Data API or handle OAuth credentials. Every skill outputs copy-ready text blocks and leaves filming, video cutting, and final uploading entirely to the human creator.
  • Heuristic Scoring Limits: hookscore.py is calibrated against 74 real short-form hooks (the opening 15 seconds of auto-captions from top-8 and bottom-8 performing videos across five channels). While effective at catching weak or generic openings, it is a heuristic linter rather than a view-count predictor and cannot reliably separate a creator's breakout hits from average uploads.
  • Reduced Functionality in Web Chat: Individual SKILL.md files can be pasted into a standard Claude web chat session to run as prompt modes, but doing so disconnects the six local Python 3 tools that power /yt-script, /yt-retention, and /yt-edit.
  • Public-Only Niche Analysis: /yt-viral inspects public video listings only and never logs into user accounts or accesses private credentials.

Sources

Impact-Site-Verification: 9b3ebd10-6c45-44cf-a014-5b8c19b7b285