stitch-skills: A Chinese Translation of Google's Stitch Agent Skills
The original google-labs-code/stitch-skills repository has 11,045 stars and is written in TypeScript under Apache-2.0. A Chinese translation of its documentation, yangshun2005/stitch-skills-cn, was published on 2026-11-2
The original google-labs-code/stitch-skills repository has 11,045 stars and is written in TypeScript under Apache-2.0. A Chinese translation of its documentation, yangshun2005/stitch-skills-cn, was published on 2026-11-21 and starts at zero stars. That number matters less than what the translation changes: the setup path for Stitch's agent skills now exists in Chinese, including the parts that trip people up before they write a single line of code.
What stitch-skills actually is
Stitch Design Skills is a set of agent skills and plugins for Google Stitch. It follows the Agent Skills open standard, so the same skill files can be loaded by Codex, Antigravity, Gemini CLI, Claude Code, Cursor, and OpenCode (the last one through manual installation). The official one-line description is: a library of Agent Skills designed to work with the Stitch MCP server.
The skills cover four workflows: generating designs, converting code to design, managing design systems, and turning designs into React components. Each is packaged so an agent can call it, rather than leaving you to paste prompts into a chat window and hope the output is consistent.
The hard prerequisite is Stitch MCP. The original README states plainly that these skills depend on the Stitch MCP server being configured and running in your agent environment. Skip that step and nothing else works. This is the first thing the Chinese quick-start section repeats, and it is the first thing people miss.
Why a Chinese README is not a cosmetic change
Installation for Codex is a single command with three --sparse flags:
codex plugin marketplace add google-labs-code/stitch-skills --ref main \
--sparse .agents/plugins \
--sparse plugins/stitch-design \
--sparse plugins/stitch-build \
--sparse plugins/stitch-utilities
The --sparse flags are optional and limit the checkout to the listed paths to speed up cloning. Omit them and you pull the whole repository. That distinction is buried in a tip block in the English README. In the Chinese version it sits in the quick-start section, which is the part a reader hits first.
Claude Code and Cursor use npx plugins add with different scopes:
# Claude Code — install into the current project
npx plugins add google-labs-code/stitch-skills --scope project --target claude-code
# Cursor — install into the workspace
npx plugins add google-labs-code/stitch-skills --scope workspace --target cursor
After registering the marketplace you install any combination of stitch-design, stitch-build, and stitch-utilities. Partial installs go through npx skills add google-labs-code/stitch-skills, with a warning in the original that skills often depend on each other and a selective install must include every required dependency.
None of this is conceptually difficult. It is a sequence of exact strings and flags where one wrong character produces an error message with no obvious cause. Having the sequence in your first language shortens the loop between reading and running.
The OpenCode path is where translations earn their keep
OpenCode does not support npx plugins or npx skills, and it does not use the Codex or Claude marketplace layout. Installation is manual: copy the skill folders into .opencode/skills/ and add a Stitch MCP remote entry to opencode.json or ~/.config/opencode/opencode.json, with the url field set to the endpoint from the Stitch MCP setup flow.
cp -R plugins/stitch-design/skills/generate-design .opencode/skills/generate-design
Two constraints apply. The skill name must be lowercase kebab-case and must match the directory name. Names like stitch::… have to be renamed before they load. This is the kind of detail that is easy to state in one sentence and painful to discover through trial and error. The Chinese README lists it under common pitfalls, alongside the note that some links on the Stitch documentation site may redirect incorrectly and need to be opened under the same host.
What the translation adds beyond the translated text
The repository keeps the translated README as the main body and marks it as a faithful translation of the original. Above it sits a separate section labeled as supplementary guidance, not original content, written for Chinese-speaking users. That section covers what the project is, the shortest path to a working install, environment requirements, and common pitfalls.
The environment section is honest about gaps. Operating system, hardware, and version requirements are all marked as not specified in the original. The only stated dependency is a configured and running Stitch MCP server with the appropriate environment variables and credentials, plus the ability to run codex and npx for the plugin install routes. There is also a general note that networks in mainland China may need a mirror or accelerated download, flagged as a general tip rather than something from the original documentation.
That last note is the sort of thing a translation can add without pretending it came from upstream. It is labeled as such, which keeps the boundary between original and supplement visible.
Who should read which
If you already work in English and have Stitch MCP configured, the original repository is the source of truth and the translation adds little. If you are setting up Stitch skills for the first time, or you are on OpenCode where the install is manual and the naming rules are strict, the Chinese version compresses the reading you need to do before the first successful run.
The translation date is 2026-11-21, and the project invites issues or pull requests for inaccuracies. Since the Chinese repository is new and has no stars yet, treat it as a reading aid pinned to a specific snapshot of the upstream README rather than a maintained fork. When the original changes, check README_en.md before assuming the Chinese text still matches.
I have not run the full install across all six agents, so my read on which paths are smoothest is based on the documented steps rather than hands-on testing on each one. The commands and constraints above come from the README text, unchanged.
Chinese documentation: https://github.com/yangshun2005/stitch-skills-cn
I'm an independent enterprise AI delivery consultant — private LLM deployment, RAG, agent workflows, AIOps. If your team is stuck somewhere in that stack, my inbox is open.
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.