Analysis updated 2026-05-18
Collect YouTube videos and web articles on a topic into a NotebookLM notebook for research.
Score a draft article against a nine-dimension rubric and iterate rewrites to raise the score.
Publish a finished article as drafts to Zhihu, Juejin, CSDN, and WeChat public accounts.
Download YouTube videos, playlists, or audio using a yt-dlp command reference skill.
| xiaomoboy/claude-writing-skills | avbiswas/sam2-mlx | chanthruu/wind_energy_monitor | |
|---|---|---|---|
| Stars | 27 | 27 | 27 |
| Language | Python | Python | Python |
| Setup difficulty | easy | moderate | moderate |
| Complexity | 2/5 | 4/5 | 3/5 |
| Audience | writer | researcher | data |
Figures from each repo's GitHub metadata at analysis time.
Requires Python 3.9+ and each skill's own command-line dependency, such as yt-dlp for video downloads.
This repository is a set of five skills for Claude Code, aimed at people who write long-form articles and want AI help across the full workflow: gathering research, scoring drafts, rewriting, and posting to publishing platforms. The author built these from a year of personal use and extracted only the parts they actually run every week. The five skills each do one thing. The research-collector skill pulls in YouTube videos and web articles on a topic, pushes them into a NotebookLM notebook, runs analysis queries, and saves the results as markdown. The publisher-wechatsync skill takes a finished markdown article and sends it as drafts to Chinese platforms like Zhihu, Juejin, CSDN, and WeChat public accounts. The article-optimizer skill scores a single article against a fixed rubric across nine dimensions, then iterates rewrites round by round to raise the score. The score-optimizer skill works on the rubric itself, batch-scoring a labeled sample set so you can tune the evaluator when it gives scores that feel off. The yt-dlp-direct skill is a practical command reference for downloading YouTube videos, playlists, or audio using the yt-dlp command-line tool. A design principle runs through all five: each skill wraps a real command-line tool rather than trying to replace it with pure prompting. Where a CLI already does the job well, the skill adds operational knowledge around it. The boundaries between research, scoring, rewriting, and publishing are kept explicit, and skills do not cross them. Installation goes through the Claude Code plugin marketplace in two commands. Each skill has its own command-line dependency, so you only need to install the ones for the skills you plan to use. The skill body text is written in Chinese, which is the author's working language, though the descriptions that trigger the skill are in English. Python 3.9 or later is required for the bundled scripts, using only the standard library with no additional packages. The repository includes detailed install guides for macOS, Linux, and Windows, plus a separate integrations document for using these skills outside Claude Code with tools like Cursor, Aider, or the Claude API directly.
A set of five Claude Code skills for long-form writers covering research collection, draft scoring, rewriting, and publishing to Chinese platforms.
Mainly Python. The stack also includes Python, Claude Code, NotebookLM.
The README does not state a license, so usage rights are unclear.
Setup difficulty is rated easy, with roughly 30min to a first successful run.
Mainly writer.
This repo across BitVibe Labs
Verify against the repo before relying on details.