z-skills: 19 local Agent Skills for Chinese content work, from web clipping to animated SVG diagrams
A collection of reusable skills
At a glance
- What is it?
- tjxj/z-skills is a folder-per-skill collection of Agent Skills for Chinese-language creation: web clipping, video download, document parsing, Word and PDF output, audio generation and animated explainer images. Each skill carries its own dependencies, so installing the folder is only half of the setup.
- Who is it for?
- Use z-skills if you write Chinese content with an agent that reads Agent Skills and you want ready-made folders for clipping, converting and publishing. Do not expect a one-step install: each skill needs its own Python packages, system tools or cloud account, and the repository shows no license.
- Can I use it commercially?
- Not without permission. GitHub finds no licence file in the repository, and without a licence all rights are reserved by default: you may read the code but not reuse it. Check the README, or ask the authors, before using it.
- Is it still maintained?
- Yes. The repository last received commits 1 day ago.
- What is it written in?
- Mainly Python, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on November 20, 2026, and from our analysis. They are not legal advice.
Star history
- tjxj/z-skills670 ★ · +141
| Repository | 2026-09-18 | 2026-11-20 |
|---|---|---|
| tjxj/z-skills | 529 | 670 |
141 Stars in the last 63 days, measured from the daily GitHub snapshot. Compare with other projects
Editorial analysis
Installing one skill by name, because folder names and skill names differ
The repository holds 19 skills in top-level folders prefixed with z-, plus a separate z-life-hub group for family life management covering diaries, reading notes, health records, credentials and computer management. You install selectively with the npx skills installer, which needs Node.js and npm:
npx skills add tjxj/z-skills --skill z-grounded-source-qa
npx skills add tjxj/z-skills --skill z-qwen-audio-studioThe installer asks which target agent to use. A folder name is not always the skill name: the folder `z-poetry-book` registers a skill named `poetry-book`, and it has its own install path described in that skill's README. The manual route is a git clone followed by `cp -R` of one folder, or of every `z-*` folder, into the skills directory your agent reads. Copying only installs the workflow and its resources. Python packages, system tools and model APIs still have to be configured per skill, so a copied skill can load and then fail on its first run.
After installing, refresh the skills list in your host or open a new session, then describe the task in plain language or name the skill outright. The script commands throughout the README run from the root of the cloned repository, so if you installed into another directory you must adjust the script paths yourself. The `python3` on your path has to be the environment where the skill's packages were installed, otherwise the import errors will look unrelated to the install.
z-web-pack stops at a media inventory and hands videos to the downloader
The two collection skills have a deliberate boundary. `z-web-pack` collects article text, related links, images, a local reading map, a media link list and recovery notes for restricted sources. When it finds a video it writes the link only into `04-media-inventory.md`. To save the video you pass the Source URL from that inventory to `z-video-downloader`.
A single-page collection looks like this, after installing requests, beautifulsoup4 and readability-lxml in your Python environment:
python3 z-web-pack/scripts/collect_web_pack.py \
--out-root "./Clippings/Reading" \
--title "主题名" \
--max-depth 0 \
--max-pages 1 \
"https://example.com/article"The output folder is organized around six files: `README.md`, `00-research-brief.md`, `01-link-inventory.md`, `02-image-inventory.md`, `03-reading-map.md` and `04-media-inventory.md`. Setting `--max-depth 0` and `--max-pages 1` keeps the run to the one page you named, and raising them widens the crawl, so set them before pointing the skill at a site with many internal links.
Video download depends on yt-dlp, ffmpeg and a parsing service you do not control
`z-video-downloader` covers YouTube, Bilibili, Vimeo, X/Twitter, TikTok, Douyin, Instagram, Facebook, WeChat Channels, m3u8 and direct video links. It supports resumable downloads, batch lists, subtitles, covers and a download history. The Python side needs `requests`; platform videos also need `yt-dlp`, and merging audio with video needs `ffmpeg`.
python3 z-video-downloader/scripts/download_video.py \
--out-root "./Video/Downloads" \
--title "主题名" \
"https://www.bilibili.com/video/BV..."To download from a clipped pack instead, pass the inventory file:
python3 z-video-downloader/scripts/download_video.py \
--inventory "/path/to/04-media-inventory.md"The limits are stated plainly. Availability depends on login state, platform risk control and the status of the parsing service. WeChat Channels share links (`weixin.qq.com/sph/`) are resolved through an online parsing service, which needs no extra install but means the link is sent to a third party. A download that works today can stop working when a platform changes, and the skill cannot fix that.
The audio skill needs a Beijing-region Bailian account before it makes any sound
`z-qwen-audio-studio` calls Alibaba Cloud Bailian `qwen-audio-3.1-tts-next`. It handles single-voice narration, commercial ads, two-person or multi-person podcasts and radio drama, with ambient sound, action effects and background music. You can supply up to 3 reference audio clips and refer to them as `@voice1`, `@voice2` and `@voice3`. Output is WAV, MP3 or PCM at up to 48kHz stereo. After download it runs an ffprobe and ffmpeg full-decode check, and it redacts the API key, Workspace ID, reference audio Base64 and API domain from logs.
Setup has five steps: enable the model in the North China 2 (Beijing) region, create an API key valid in that region, find the Workspace ID of the workspace holding the key, install Python 3.9 or newer plus `requests`, `ffmpeg` and `ffprobe`, and set environment variables in a private terminal session:
export DASHSCOPE_API_KEY="your-api-key"
export SFM_WORKSPACE_ID="your-workspace-id"
python3 z-qwen-audio-studio/scripts/qwen_audio_studio.py doctorThe `doctor` command reports whether dependencies and variables are set without printing the credentials. A key created in another region will not satisfy this setup.
z-dongxi renders animated explainer SVGs from a spec file
z-dongxi turns content that has a process or numbers in it into a moving diagram: how RAG data flows, how an HTTPS handshake works, where money goes, where traffic goes, where a two-week schedule stalls, which of several models is stronger. You provide a topic and content, and the agent chooses a chart type and a style, writes a spec, renders it, lints it and checks screenshots of the frames.
The skill offers 35 chart types, 27 styles and 5 aspect ratios, with 3:4 as the default for WeChat official account posts. The result is a single-line SVG you can paste directly into that editor, and it can also export PNG, GIF and MP4. The install and a demo render are:
npx skills add tjxj/z-skills --skill z-dongxi
python3 z-dongxi/scripts/render.py --spec z-dongxi/demos/specs/rag.json --style sketch --out rag.svg --png rag.pngThe demo spec `rag.json` and the `sketch` style are the ones shown for a first run. This skill has its own README inside the z-dongxi folder, and that README, not the top-level one, is where its details live.
Word, PDF and book output each write to fixed places and fixed styles
`z-md-to-word` converts a local Markdown article into Word. It keeps the heading, author and date, embeds remote and local images, skips empty upload placeholders, corrects common list recognition problems, and checks afterward that the document opens and renders. The default output paths are `output/doc/<原文件名>.docx` and `output/doc/<原文件名>.doc`.
`z-md-to-pdf` lays Markdown out in three Chinese PDF styles, academic, book and modern report, with style selection, batch generation and XeLaTeX project recompilation. `z-poetry-book` builds a sourced, rebuildable LaTeX PDF from poems, classical texts, technical articles and course notes, with a source list, images, build files, a verification report and preview images. It ships six themes: `song-mist` (the default), `celadon`, `ink-zen`, `cinnabar-seal`, `moon-river` and `bamboo`. The default layout is Simplified Chinese, horizontal. Because the book route compiles with LaTeX, a machine without a working TeX setup cannot produce the PDF even though the skill folder installed cleanly.
Three smaller skills handle document plumbing. `z-smart-xparse` uses xparse-cli to turn PDFs, images and Office files into Markdown or structured results, including scanned files. `z-md-excel` pulls the tables out of a Markdown file into Excel, and `z-excel-editor` reads, edits, cleans and formats spreadsheets, including filling formulas. `z-docx-format-brush` extracts a format fingerprint from a template docx, covering fonts, sizes, line spacing, indents, table borders, cover and page breaks, and applies it to a target docx to repair messy pandoc or model output.
No license, no releases and a Chinese-language README set the adoption risks
The repository metadata lists no license, so the right to reuse or redistribute the skills is not stated. There are no GitHub releases either, which means there is no tagged version to pin; you track the main branch, and the last push was on 2026-11-02. The README and the skill table are written in Chinese, and the skill descriptions use Chinese trigger phrases such as 采集网页素材 and 转成doc. An agent prompted in English may not match those triggers, although you can name a skill explicitly.
The README also points to each directory's `SKILL.md` as the authority on capability, dependencies and execution flow, so the table is a summary rather than a specification. Some skills are demos by design: `z-expense-policy-qa` ships with a fictional demonstration policy, so its answers about reimbursement amounts and deadlines apply to that sample policy until you replace it with your own.
The retrieval skill works the same way on your own files: `z-grounded-source-qa` runs multi-expression search over any local Markdown or TXT material and answers only from that evidence, with checking against the original text and source citations. `z-mail-reader` reads mail over IMAP, downloads attachments, summarizes messages and watches for new mail, so it needs mailbox credentials from you. Both are only as reliable as the folder or account you point them at.
Editorial conclusion
Use z-skills if you write Chinese content with an agent that reads Agent Skills and you want ready-made folders for clipping, converting and publishing. Do not expect a one-step install: each skill needs its own Python packages, system tools or cloud account, and the repository shows no license. Before copying anything, read the SKILL.md of the one skill you want, check its dependencies, and confirm the license situation if you plan to redistribute.
Frequently asked questions
What is z-skills?
z-skills is a collection of 19 reusable local Agent Skills for Chinese creation, knowledge management and automation, installable into environments that support Agent Skills. A separate z-life-hub group covers household management.
How do I install a single skill from z-skills?
Run npx skills add tjxj/z-skills --skill with a skill folder name such as z-grounded-source-qa, which needs Node.js and npm. You can also clone the repository and copy one folder with cp -R into your agent's skills directory.
Does z-skills need API keys or extra tools?
Some skills do. The audio skill needs DASHSCOPE_API_KEY and SFM_WORKSPACE_ID for a Beijing-region Bailian account, and the video skill needs yt-dlp and ffmpeg. Copying a folder does not install these.
Official sources
Was this useful?