tjxj/z-skills

What it solves

This project bundles many small, reusable skills that let an AI agent handle common real-world tasks without reinventing the wheel each time. Instead of asking the agent to figure out how to scrape a webpage, download a video, parse a PDF, or format a Word document from scratch, you drop in a skill that already encodes the steps. It covers Chinese-focused scenarios like turning articles into comic strips, generating handwritten-style slides, converting Markdown to Word/PDF, and answering questions strictly from provided source materials.

How it works

Each skill lives in its own folder with a SKILL.md file that describes when to trigger it (keywords), what steps to run, and what output to produce. Many skills also come with ready-to-use Python scripts under scripts/ that do the heavy lifting—for example collect_web_pack.py for scraping, download_video.py for downloading videos, or extract_fingerprint.py for reading a docx template's formatting. You install skills by copying their folders into your local .agent/skills/ directory (or using npx skills add for specific ones). When you start a new session with an agent like Codex or Claude Code, the agent reads each SKILL.md and matches the trigger words in your request to the appropriate skill.

Who it’s for

It's aimed at people who use AI coding agents (Codex, Claude Code, etc.) for Chinese writing, knowledge management, and automation tasks. If you regularly need your agent to fetch web content, download videos, parse documents, generate audio, create slides, or answer questions from local files, these skills give you a predictable, tested way to get that done without writing custom prompts every time.

Highlights

• Covers a wide range: web scraping, video download (including WeChat channels and m3u8), video study pages, full-scene audio generation, document parsing, email reading, Markdown-to-Word/PDF, table extraction, docx format brushing, handwritten slides, handwritten videos, comic generation, and grounded Q&A.
• Each skill is self-contained with its own SKILL.md, scripts, tests, and evaluation examples.
• Includes specific tools like z-docx-format-brush that extracts formatting fingerprints from a template and applies them to another docx to fix messy outputs from pandoc or models.
• z-qwen-audio-studio supports multi-voice podcasts, sound effects, background music, and outputs WAV/MP3/PCM with automatic validation using ffprobe/ffmpeg.
• Video downloader supports resume with .part files and handles multiple platforms including YouTube, Bilibili, WeChat Video, and direct m3u8 links.
• The project emphasizes verifiable outputs and clear boundaries—for example, z-web-pack only collects media inventories, leaving actual downloads to z-video-downloader.