Matt Pocock skills zh-CN: a translation mirror with a fixed skill vocabulary

A mirror that syncs content but not git history

This repository is not an ordinary fork. It holds a Simplified Chinese localization of `mattpocock/skills`, and the translation policy is narrow on purpose. Natural-language descriptions get translated. Directory names, skill names, frontmatter keys, commands, code blocks, paths, URLs, package and tool and API identifiers, and labels that carry behaviour all stay exactly as upstream wrote them. The stated reason is that rewriting any of those would break install and runtime behaviour.

Sync runs by content refresh rather than by history. The page says the mirror carries neither the upstream Git history nor the upstream repository management status, so you cannot read the upstream commit trail through it, and the maintenance rules live in `.skills/translate-skill/SKILL.md`. The translation work is executed by Codex and reaches `main` through pull requests opened by the maintainer. Requested and already decided renderings are collected in `TRANSLATION-GLOSSARY.md`, and `CONTRIBUTING.md` is where you claim a term or a translation PR. The user-visible install path stays `vinvcn/mattpocock-skills-zh-CN` in every method the page gives.

Two install paths, and using both duplicates every skill

The page groups the routes into two postures and tells you to pick only one per agent.

npx skills@latest add vinvcn/mattpocock-skills-zh-CN

The skills.sh route copies the skill files into your project so you can edit and customise them, which the page frames as making them your own. That is the fork-shaped option, and it puts the update burden on you.

/plugin marketplace add vinvcn/mattpocock-skills-zh-CN
/plugin install mattpocock-skills@mattpocock

The Claude Code plugin route updates itself, automatically by default. The same two operations exist for the shell:

claude plugin marketplace add vinvcn/mattpocock-skills-zh-CN
claude plugin install mattpocock-skills@mattpocock

Choosing both is the documented mistake. Install one and you own editable files, install the other and you get version tracking, install both and every skill ends up on disk twice.

Every skill is invoked as /zh-<name>, which does not line up

For the DeepSeek Harness plugin install, the page says any skill is invoked as `/zh-<name>` and gives `/zh-grilling` as the example. The names in the translated body are the upstream ones: `/setup-matt-pocock-skills`, `/grill-me` under `skills/productivity/`, `/grill-with-docs` under `skills/engineering/`, plus `/triage` for the label pass.

That gap is worth checking before a command goes into a script or a habit. The example on the page, `/zh-grilling`, matches neither of the grilling skill names the same page links, which sit at `skills/productivity/grill-me/SKILL.md` and `skills/engineering/grill-with-docs/SKILL.md`. Read the real invocation name out of the directory you installed rather than deriving it from the prefix rule.

The DeepSeek Harness route has a third shape and a separate package name:

dsh plugin --profile web add @vinvcn/dsh-mattpocock-skills-zh

That path installs as a plugin against a named profile, and `dsh-plugin/README.md` holds its details.

setup-matt-pocock-skills forces three decisions on first run

One selection is mandatory on a first install: `/setup-matt-pocock-skills`. Run it inside your agent and it walks three decisions.

The first is the issue tracker, and the page names GitHub, GitLab, local files, or any other method you describe. The second is which labels to use while triaging issues, with the explicit note that `/triage` consumes those labels, so a careless answer changes what the triage skill does later. The third is where the docs it creates should be saved.

The page says to run the step once per repository, so a multi-repo workspace repeats the whole sequence rather than inheriting one global answer.

Ordering matters as well. The setup skill has to be selected during the install that brings the other skills in, and it has to run before those skills do anything, because the tracker, the labels and the docs location are inputs to later runs.

Shared language is a file, not a longer prompt

The second failure mode the set targets is verbosity. An agent dropped into a project gets asked to learn the terms while it works, then explains in twenty words something one word covers. The fix offered is not a prompt trick but an artifact: a shared language document that decodes the project's vocabulary. The page puts the point in a before and after pair taken from a `CONTEXT.md` in `course-video-manager`. Before: "There's a problem when a lesson inside a section of a course is made 'real' (i.e. given a spot in the file system)". After: "There's a problem with the materialization cascade".

That shortening is folded into `/grill-with-docs`, which runs as a grilling session while also building the shared language and writing decisions that are hard to explain into ADRs. The listed payoffs are concrete: variable, function and file names become consistent because everything uses the shared language, agents navigate the codebase more easily, and agents spend fewer tokens thinking because the language is shorter. The sibling skill `/grill-me` is the one the page points at for non-code situations.

The debugging skills assume tests you can actually execute

The third failure mode is code that will not run, and the page frames it as feedback rather than as competence. It quotes The Pragmatic Programmer: "The rate of feedback is your speed limit."

Codex and DeepSeek Harness have separate plugin routes

Agent coverage is uneven across the documented routes. Codex now has its own plugin route (`codex plugin marketplace add vinvcn/mattpocock-skills-zh-CN`, then `codex plugin add mattpocock-skills@mattpocock`), as do Claude Code and GitHub Copilot, Gemini CLI installs with `gemini skills install`, and other agents use the skills.sh installer. DeepSeek Harness users get a third one that installs as a plugin under its own package name, `@vinvcn/dsh-mattpocock-skills-zh`, run against a named profile.

The tree shows where those pieces live. `.claude-plugin/` and `dsh-plugin/` hold the two plugin wrappers, `.skills/` holds the translation maintenance rules, and `skills/` is the translated content itself. An `.out-of-scope/` directory is also present. `AGENTS.md`, `CLAUDE.md`, `GLOSSARY.md`, `SCOPE.md` and `SYNC.md` sit at the top level with `docs/`, `assets/` and `scripts/` beside them, and both `LICENSE` and `LICENSE.zh-CN.md` are present.

Activity is current: v1.1.0 published on 2026-07-23, v1.2.3 on 2026-09-14 and v1.3.1 on 2026-10-08, with the last push on 2026-10-28. What that cadence buys is translated wording catching up with upstream, not new skill logic.

Editorial conclusion

Use this mirror when your model or your team works in Chinese and you would rather not translate the upstream skills yourself, and install exactly one path so nothing lands on disk twice. Before you rely on it, open the directory you installed and confirm the real invocation name, since the page's /zh-grilling example for dsh installs does not line up with the upstream skill names. Then check the tracker, label and docs answers your /setup-matt-pocock-skills step produces. If you intend to edit the skills, take the skills.sh copy; if you only want to follow releases, take the plugin bundle.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. README
4. Releases
5. vinvcn/mattpocock-skills-zh-CN on GitHub