# A Chinese translation of everything-claude-code whose install commands still fetch the original

> xu-xiang/everything-claude-code-zh translates a large Claude Code configuration bundle into Chinese, covering agents, skills, hooks, commands, rules and MCP entries. The prose is translated and the wiring is not, so every documented command still clones and installs the upstream repository.

**xu-xiang/everything-claude-code-zh** — A Chinese translation of everything-claude-code: a complete collection of Claude Code configs (agents, skills, hooks, commands, rules, MCPs), sourced from an Anthropic hackathon winner's real-world setup, to help Chinese-speaking engineers understand and use Claude Code efficiently.

- Repository: https://github.com/xu-xiang/everything-claude-code-zh
- Website: https://oneskill.one
- Stars: 2,001 · Forks: 324
- Language: JavaScript
- License: MIT
- Published: 2026-10-14 · Updated: 2026-10-14 · Language: en
- Canonical page: https://hysenlabs.com/projects/xu-xiang-everything-claude-code-zh

## Every install command points at affaan-m, not here

This repository presents itself as the Chinese translation of a Claude Code configuration set, covering agents, skills, hooks, commands, rules and MCP entries. What it adds to the original layout is translation furniture: a second README at `README.zh-CN.md`, a `docs/` directory, and `translation_workdir/`, a staging area that never shows up in the manifest's published file list.

What did not get translated is the wiring. The marketplace line, the clone URL and the manifest's `repository`, `homepage` and `bugs` fields all name affaan-m/everything-claude-code, and the manifest is published under the name `ecc-universal`. Follow the quick start to the letter and you install the upstream English bundle. The difference lives in the prose inside `agents/`, `commands/`, `skills/` and `rules/`, which the v1.4.0 note says was completed across 80+ files.

For an engineer who drafts prompts faster in Chinese, that is the entire proposition, and it is not nothing. For anyone who wants a fork they can pin, patch or publish under their own name, the manifest hands them a foreign repository URL instead, which is a thin basis for either.

## The plugin install stops before the rules directory

Installation is two steps, and the second exists because the first is incomplete by design. The plugin route adds a marketplace and installs the bundle:

```bash
/plugin marketplace add affaan-m/everything-claude-code
/plugin install everything-claude-code@everything-claude-code
```

After those two lines you are told you can reach 13 agents, 56 skills and 32 commands. What you cannot reach is the rules directory, because the README states that Claude Code plugins cannot distribute `rules` automatically, so those files need a clone and a script of their own. Command names also change shape between the two routes, and a plugin install needs the namespaced form while a manual install accepts the short one:

```bash
/everything-claude-code:plan "添加用户身份验证"
/plan "添加用户身份验证"
/plugin list everything-claude-code@everything-claude-code
```

Every hook and script was rewritten in Node.js, and the plugin is stated to support Windows, macOS and Linux, so portability rests on the runtime rather than on shell choice. That is a real answer to the usual objection about shell hooks, but it also means the bundle carries a JavaScript runtime into whatever you install it into.

## Language arguments decide which rule directories land

Rules used to be flat files. The v1.4.0 note says they were refactored into `common/`, `typescript/`, `python/` and `golang/` directories so you install only the languages you write, and the script takes language names as arguments:

```bash
git clone https://github.com/affaan-m/everything-claude-code.git
cd everything-claude-code
./install.sh typescript python golang
./install.sh --target cursor typescript
```

The `--target cursor` flag is the only documented route to a host other than Claude Code, and that is a narrow seam. The tree carries `.agents/`, `.codex/`, `.cursor/` and `.opencode/` directories, and the manifest ships the `.opencode/` set down to `.opencode/index.ts` and `tsconfig.json`, yet no install command is given for any of them.

Because rules are copied rather than unpacked as a package, collisions are the hazard worth thinking about, which is what the `configure-ecc` skill from v1.4.0 is aimed at, with guided setup and merge and overwrite detection. How that detection resolves when a local rule file already differs from the shipped one is not spelled out anywhere in the documentation, so treat the first run as something to inspect rather than something to trust.

## Package manager choice walks six sources in order

Hooks that shell out have to guess which package manager the host project uses, and this bundle makes that guess explicit instead of silent. Resolution runs through six steps: the `CLAUDE_PACKAGE_MANAGER` environment variable, a project file at `.claude/package-manager.json`, the `packageManager` field in `package.json`, a lock file match among `package-lock.json`, `yarn.lock`, `pnpm-lock.yaml` and `bun.lockb`, a global file at `~/.claude/package-manager.json`, and finally the first manager found on the machine.

```bash
export CLAUDE_PACKAGE_MANAGER=pnpm
node scripts/setup-package-manager.js --global pnpm
node scripts/setup-package-manager.js --detect
```

The same script writes the global and project variants, and `/setup-pm` does the same job from inside Claude Code. The last link in that chain is the one to watch, since the fallback is the first available manager: on a machine missing the project's manager, the bundle proceeds with a wrong choice, and nothing in the documentation says how the mismatch gets surfaced to you.

## Instinct import once dropped everything after frontmatter

The learning loop in this bundle stores what it extracts as instincts, laid out as frontmatter followed by Action, Evidence and Examples sections. A defect in that loop deserves more attention than most, because the v1.4.1 entry records that `parse_instinct_file()` silently discarded everything below the frontmatter whenever `/instinct-import` ran, reported by the community contributor @ericcai0814 and closed through issue 148 and pull request 161.

Silent truncation is the dangerous shape of this failure. Nothing errors, the import reports success, and the evidence and examples that make an instinct worth reusing are the first thing to disappear. The v1.4.1 note names the fix without naming a regression test, and the machinery around those files kept growing: v1.2.0 had already shipped a second generation of continuous learning built on instincts with confidence scoring, import and export, and evolution, and `skill-stocktake` arrived with v1.6.0 as one of seven new skills.

If you adopt any of it, read an instinct file that came out of an import before you let an agent act on it.

## Six languages are claimed but only two are navigable

The header of this translation offers three reading paths: English, Simplified Chinese, and Traditional Chinese at `docs/zh-TW/README.md`. A second switcher lower down adds Japanese at `docs/ja-JP/README.md`, and the headline block claims six languages of support, a figure the v1.6.0 note repeats when it credits contributions from 30 contributors across 6 languages. Two of those six have a named path in the file header, and both live under `docs/`. The rest have no route at all, so the language claim is larger than anything a reader can reach from this file.

The version headings tell a similar story about record keeping. The repository is not archived, the last push was on 2026-03-05, and because it publishes no GitHub releases, the entries under What's New, running from v1.2.0 to v1.7.0 with every one of them dated February 2026, are the only release record on offer. The manifest string 1.7.0 matches the newest heading.

The guides are split the same way. The README sends readers to two posts on x.com for the shorthand and longform guides, covering token optimization, memory persistence, continuous learning, validation loops, parallelization with git worktrees, and subagent orchestration, while the top-level tree also carries `the-shortform-guide.md`, `the-longform-guide.md`, `the-security-guide.md` and `the-openclaw-guide.md`. Nothing under What's New records whether those documents were translated.

## Codex, Cursor and OpenCode get different mechanisms

Claiming support for Claude Code, Codex, Cowork and other agent frameworks is cheap; the tree shows how much of it is real. The v1.3.0 entry added a full OpenCode port with 12 agents, 24 commands and 16 skills, hooks wired through OpenCode's plugin system across more than 20 event types, three native custom tools named `run-tests`, `check-coverage` and `security-audit`, plus an `llms.txt`. The v1.6.0 entry then added a Codex CLI route through a `/codex-setup` command that generates a `codex.md` file, and v1.7.0 widened that into Codex app support built on `AGENTS.md`.

Three targets, three levels of effort. OpenCode gets a compiled plugin with its own tools and event surface, Codex gets a generated markdown document, Cursor gets a flag on a shell script. Only one of them ships an install command in the quick start, and none of the counts transfer between targets: 13 agents is what a plugin install yields, 12 agents is what the OpenCode port carries, so a number quoted for one says nothing about another.

Security tooling arrives through a separate door. The `/security-scan` skill runs AgentShield from inside Claude Code, and AgentShield itself carries 1282 tests and 102 rules behind it; the same entry put an ECC Tools GitHub app on github.com/marketplace/ecc-tools with free, professional and enterprise tiers. Read that as an external dependency with a commercial path behind it, not as another skill you can inspect in the tree.

## The skill catalogue drifted from engineering to content work

Read what v1.6.0 and v1.7.0 added and the audience shifts underneath those notes. The skills introduced in v1.7.0 are `article-writing`, `content-engine`, `market-research`, `investor-materials` and `investor-outreach`, alongside `frontend-slides`, a dependency-free HTML presentation generator that also carries a PPTX conversion guide. The seven from v1.6.0 sit closer to engineering craft: `search-first`, `swift-actor-persistence`, `swift-protocol-di-testing`, `regex-vs-llm-structured-text`, `content-hash-cache-pattern` and `cost-aware-llm-pipeline`, plus `skill-stocktake`.

That mix says the bundle stopped being a developer toolkit and turned into a founder's toolkit. Language-specific depth arrived alongside it, with Django patterns, security, TDD and verification, the same quartet again for Java Spring Boot, and six commands for spreading work across services, from `/pm2` through `/multi-plan` to `/multi-workflow`. A `/sessions` command for browsing session history sits with them.

For a team, the effect is that skill count is a poor measure of usefulness here. Fifty-six skills reads like coverage; the five that write investor materials are dead weight for a backend team, while the two about token cost and cache design are the ones that compound over months.

## Conclusion

Adopt this repository if you draft prompts faster in Chinese and want the agents, skills and rules text in that language, since the translation covers 80+ files and leaves the plumbing alone. Skip it if you need a self-contained install or a version marker beyond the manifest string 1.7.0, because every documented command clones affaan-m/everything-claude-code. Before committing, clone that repository, run install.sh for one language, and diff the rule files it writes against your own, because an overwrite there is silent.

## FAQ

### Is everything-claude-code-zh a standalone project or a translation of another repository?

It is a translation. The quick start adds the marketplace affaan-m/everything-claude-code, the clone URL points at the same account, and the manifest lists the upstream repository, homepage and bugs URLs under the name ecc-universal. The top-level tree adds README.zh-CN.md, a docs/ directory and translation_workdir/ on top of that layout.

### Why do the rules still need a manual install after the plugin is installed?

The README states that Claude Code plugins cannot distribute the rules directory automatically, so rules need a separate clone followed by install.sh with your language names. Agents, skills and commands arrive through the plugin, rules do not.

### Does everything-claude-code-zh work with Codex, Cursor and OpenCode?

The README names Claude Code, Codex, Cowork and other agent frameworks as targets. The v1.3.0 entry added an OpenCode port with 12 agents, 24 commands, 16 skills and hooks across more than 20 event types, v1.6.0 added a /codex-setup command that generates codex.md, and Cursor is reached through the --target cursor flag on install.sh.

### How many agents, skills and commands do I get after installing?

The quick start ends with 13 agents, 56 skills and 32 commands on the Claude Code plugin route. Those figures do not carry over: the OpenCode port is described separately with 12 agents, 24 commands and 16 skills.

### What was the instinct import bug fixed in v1.4.1?

parse_instinct_file() silently dropped everything after the frontmatter, which removed the Action, Evidence and Examples sections during /instinct-import without raising an error. The fix came from the community contributor @ericcai0814 through issue 148 and pull request 161.

### How do I control which package manager the hooks use?

Resolution walks six sources: CLAUDE_PACKAGE_MANAGER, .claude/package-manager.json, the packageManager field in package.json, a lock file match, ~/.claude/package-manager.json, then the first manager available. You can write the choice with scripts/setup-package-manager.js using --global, --project or --detect, or set it from Claude Code with /setup-pm.

## Sources

- [Issues](https://github.com/xu-xiang/everything-claude-code-zh/issues)
- [License: MIT](https://github.com/xu-xiang/everything-claude-code-zh/blob/main/LICENSE)
- [Project website](https://oneskill.one)
- [README](https://github.com/xu-xiang/everything-claude-code-zh/blob/main/README.md)
- [xu-xiang/everything-claude-code-zh on GitHub](https://github.com/xu-xiang/everything-claude-code-zh)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/xu-xiang-everything-claude-code-zh