wlzh/skills

What is wlzh/skills?

It's a GitHub project that bundles many small, independent tools called skills. Each skill is a script or set of scripts that can be used on its own or combined with others, typically through the Claude Code assistant (or directly via the command line).

The project is written in Chinese (README is in Chinese), but the skills themselves are universal.

The skills (as listed in the README)

🎙️ youtube-to-xiaoyuzhou

Automatically downloads a YouTube video/audio and publishes it as a podcast episode to Xiaoyuzhou (小宇宙). Features: AI-generated cover image, keyword filtering, voice changing, scheduled publishing, and cookie-based login.

🎵 voice-changer (v1.5.0)

Audio voice changer using RVC AI models – real timbre conversion (not just pitch). Handles long audio by splitting it into chunks to avoid memory issues. Supports multiple voice presets (Chinese voices like 御姐, 七妹, etc.) and can be called by other skills.

🎬 video-voice-changer (v1.1.1)

Extracts audio from a video, applies voice changing (reusing voice-changer), then re-merges the audio with the original video. Keeps original video quality, never overwrites the source file (creates a new one), and cleans up temp files automatically.

✂️ audiocut-keyword (v1.1.0)

Audio keyword filter. Uses FunASR Paraformer for accurate transcription (character-level timestamps), identifies keywords from a config, and cuts them out using FFmpeg. Works with Python 3.10. Use cases: ad removal, content cleanup, YouTube-to-podcast pipelines.

🎙️ text-to-speech (v1.1.0)

Text-to-speech with high-quality voices (based on Microsoft Edge TTS, 18+ Chinese voices). Can parse podcast scripts (removes comments/markup), handle SSML emotion tags, optionally integrate voice-changer for post-processing, and is highly configurable.

📹 video-downloader (v1.1.0)

Downloads YouTube videos with full control over quality (best/1080p/720p/480p/360p) and format (MP4, WebM, MKV). Also supports audio-only MP3 download. Auto-installs dependencies and shows video info. Note: it's derived from an external source.

🎨 image-generator (v1.1.0)

Generic image generation skill supporting multiple AI models (ModelScope, Gemini, etc.). Flexible prompt/parameters, auto-saves images, and is designed to be called by other skills.

🔥 code-roaster (v1.1.0)

Reviews code in a funny, Gordon Ramsay style – gives a sarcastic but practical code review report covering code smells, bugs, security, performance, and best practices. Three modes: mild / standard / brutal.

📹 youtube-to-blog-post (v3.5)

Converts a YouTube video into an SEO-optimized Hexo blog post. Extracts real content from the video description (including code examples), generates a 160-character meta description and 5-8 high-quality keywords, uses the YouTube thumbnail as cover, removes AI-writing traces, and produces an English kebab-case file name.

📦 quark-mswnlz-publisher (v1.3.0)

One-click publishing pipeline: bulk-saves files from Quark (夸克网盘) to a GitHub resource repo, auto-generates share links (permanent + encrypted + random extraction code), categorizes resources into 10+ repos (books, movies, etc.), writes monthly markdown files and updates README indexes, auto-commits and pushes to GitHub, sends Telegram notifications (single channel for each repo, one aggregated message for groups), and triggers a website build on GitHub Actions. Includes a template for promotional files that get copied into every resource folder.

📺 youtube-tracker

Periodically checks a list of YouTube channels for new video uploads. Only outputs when a new video is found (good for cron job pushing to group chat). Supports adding channels via @handle, URL, or channelId. Verifies API key, and sets up a baseline on first run to avoid flooding you with old videos.

📺 youtube-publisher (v1.2.0)

Uploads videos to YouTube with full metadata control: title, description, tags, category, thumbnail (local path or URL), subtitles (SRT/VTT), privacy (public/unlisted/private), playlist assignment, and Shorts flag. Has a dry-run mode to preview without actually uploading. Also auto-converts > and < characters in title/description to 》 and 《.

📋 wlzh-invoice-scanner (v3.8.0)

Scans directories for all types of invoices (transport, accommodation, dining, etc.), extracts key info, and generates a categorized summary report. Uses Python for amount verification (to avoid LLM math errors). Handles 5 categories: city transport, long-distance transport, accommodation, ticket refund fees, dining. All railway tickets count as long-distance. Refund fees are separate. Outputs a structured summary card and can append to an existing report. Supports batch scanning multiple directories at once, with mixed modes.

🐦 x-fetcher (v1.1.0)

Fetches tweets and X (Twitter) long articles, saves them as Markdown files. Also downloads media (images/videos), grabs engagement data (likes, retweets, views, bookmarks), and organizes by username. Has profile-level and user-level config, and a smart first-run guide. Based on an external project.

📰 wespy-fetcher (v1.1.1)

Fetches articles from WeChat public accounts, generic web pages, and Juejin. Can list WeChat album contents (--album-only) and batch-download full albums (--max-articles). Default output is Markdown, but also supports HTML/JSON/ALL. If the upstream repo is missing, it auto-clones it to a unified directory (~/Documents/QNSZ/project). Based on an external project.

🔒 vps-security-hardening (v1.0.7)

Automatically secures a VPS via SSH. Performs 7 hardening steps: create a sudo user and disable root password login, change SSH port (supports Ubuntu versions), install/configure Fail2ban, enable SSH key login, optional SSH login notification, configure UFW firewall, and show Docker security reminders. Auto-detects Ubuntu version for SSH socket handling. Outputs a full security report.

Installation & usage

• Each skill is independent – you can use one or combine several.
• Basic usage:

  python3 ~/.claude/skills/<skill-name>/scripts/script.py [arguments]
  # or via Claude Code:
  /<skill-name> [arguments]
  

• Dependencies differ per skill – see each skill's README.
• Example combinations:
  • youtube-to-xiaoyuzhou <url> --filter-keywords --change-voice female_3 (download → filter → voice change → publish)
  • text-to-speech script.txt --voice female_3 (TTS → voice change)

Related project

prd-manager (v2.0.0) – separate repo

A skill for managing versioned requirement documents (PRD, design, dev, ops, test, change log). Helps with Vibe Coding by enforcing test-driven development, automating ops (deploy, monitor, rollback), and recording incidents for learning.

License & contribution

• MIT License (see LICENSE file).
• Contributions welcome – issues and pull requests are accepted.