Claude Code Skills Chinese Collection
Chinese-language catalog of 438+ Claude Code skills, agents and plugins, plus 20 install-ready original skills.
- Source repo
- laolaoshiren/claude-code-skills-zh
- Stars
- ★ 904
- Last updated
- today
- License
- MIT
- Primary language
- Python
- FA score
- 36/100 · Major gaps
At a glance
- How it runs
- Works with
- Portable with changesClaude CodeCodex (Partial support)
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- A Chinese-speaking developer joining an unfamiliar codebase wants a Chinese-language review report via zh-code-reviewer
- Not a fit if
- Teams that need English-only documentation and catalogs
- Teams wanting one audited, maintained skill package
- Organizations unwilling to vet third-party skill sources
- Source review
- 36/100 · Major gaps
What does this agent do, and when should you use it?
This is a Markdown-first curated GitHub repository, not a runnable program. It organizes 438+ Claude Code Skills, Agents and Plugins into scenario buckets — popular, platform operations, developer efficiency, content creation, academic research, AI agents, finance/business and Chinese-specific — and adds 20 original skills of its own, with zh-code-reviewer, zh-readme, api-tester, refactor-advisor and perf-profiler featured as the starting five. The original skills ship as SKILL.md directories; installation means copying skills/ into ~/.claude/skills/ so Claude Code can load them in session. There is no runtime, server or model-calling code in the repo: the value is a Chinese-readable selection list plus a small set of immediately usable skill files. It is aimed at developers working in Claude Code, with Codex, Gemini CLI and Cursor mentioned as adjacent targets.
The core artifact is a large categorized table in README: each row gives a skill name, GitHub link, a one-line Chinese description and sometimes a star count. Original skills live under skills/ (for example skills/zh-code-reviewer/, skills/zh-readme/, skills/api-tester/), each a directory containing SKILL.md that Claude Code reads to acquire domain instructions. Users clone the repo or download the ZIP, then run cp -r claude-code-skills-zh/skills/* ~/.claude/skills/ to install into the global skills directory, after which they trigger tasks in a Claude Code session — producing a Chinese code review report, drafting a Chinese README, or generating and validating API tests against a confirmed contract and authorized environment.
- A Chinese-speaking developer joining an unfamiliar codebase wants a Chinese-language review report via zh-code-reviewer
- An open-source maintainer needs a Chinese README and uses zh-readme to analyze the project first, then write
- A backend engineer with a settled API contract wants api-tester to generate and verify tests in an authorized environment
- A team maintaining a legacy service uses refactor-advisor to surface code smells and get an actionable refactoring list
- A developer profiling a slow path uses perf-profiler to locate bottlenecks from a reproducible baseline and profiler evidence
- A newcomer wants to browse the high-star catalog table by scenario before installing individual skills
How do you install or deploy this agent?
Requires Claude Code and git installed locally. Native GitHub path:
git clone https://github.com/laolaoshiren/claude-code-skills-zh.git
cp -r claude-code-skills-zh/skills/* ~/.claude/skills/If GitHub is slow, use the documented mirror:
git clone https://gh-proxy.com/https://github.com/laolaoshiren/claude-code-skills-zh.git
cp -r claude-code-skills-zh/skills/* ~/.claude/skills/Or skip git and download the ZIP:
# https://github.com/laolaoshiren/claude-code-skills-zh/archive/main.zip
# mirror: https://gh-proxy.com/https://github.com/laolaoshiren/claude-code-skills-zh/archive/main.zipHow do you use this agent?
After the copy, trigger the skills conversationally inside Claude Code: ask for a Chinese code review of the current repo, a Chinese README, API tests generated and run against a contract, refactoring suggestions, or a performance analysis. api-tester confirms environment and authorization before generating and validating tests; perf-profiler requires a reproducible baseline before trusting profiler evidence. The repo exposes no CLI entry point itself — the workflow is copy into ~/.claude/skills/ and drive it through Claude Code dialogue.
What are this agent's strengths and limitations?
- Chinese-first documentation across the whole catalog, removing the translation step of English roundups
- Ships 20 original skills as actual SKILL.md directories rather than links only
- Covers Chinese platform scenarios (WeChat Official Accounts, Xiaohongshu, Feishu, WeCom) that large English lists typically omit
- Install is one git clone plus one cp command — no build step, no server
- MIT licensed, so copying and internal redistribution are uncomplicated
- Star counts across the 438+ entries drift over time; the README itself calls them scale references, not rankings
- Third-party rows are link plus one sentence with no shared security audit trail, so provenance and authorization must be checked per skill
- Only 20 original skills, and they assume the Claude Code skills directory convention, so porting to other agents needs rework
- No versioned releases, tests or CI validation; following updates means tracking main
- The README is extremely long, so discovery depends on Ctrl/Cmd+F or the external site
How does this agent compare with similar options?
The README names its alternatives directly: anthropics/skills is the official English general-purpose set; ComposioHQ/awesome-claude-skills and VoltAgent/awesome-agent-skills are 1000+ scale English curated lists; hesreallyhim/awesome-claude-code is broader Claude Code resource navigation; alirezarezvani/claude-skills offers 380 English skills split across engineering, product and marketing. This repo differentiates on Chinese documentation, China-platform skills, and its 20 original Chinese skills, and the README frames them as complementary rather than competing.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Claude Code Skills Chinese Collection This agent | 36 · Major gaps | Agent plugin / skillFree + model costs | ★ 904 | today | Python | Claude Code |
| Polyphonic Decision Engine | 26 · Major gaps | Self-hosted service | ★ 123 | today | HTML | — |
| Flock AI Dev Team | 67 · Some gaps | Self-hosted serviceFree + model costs | ★ 501 | 21d ago | Go | Claude Code · Claude API |
| OrchestKit AI Development Toolkit | 86 · Good | Agent plugin / skillFree + model costs | ★ 300 | today | TypeScript | Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The repository is primarily a curated list plus skill descriptions; no original SKILL.md content is present in the evidence, so least privilege, user confirmation, data flow and sensitive-data handling for the actual skills cannot be assessed. The workflows do show some least-privilege discipline (contents: write scoped to auto-commit, persist-credentials: false, askpass file with umask 077 and trap cleanup), hence least_privilege 1. Dependency security scores 2 because all Actions are pinned to commit SHAs, a verifiable good practice; deducted because recommended third-party skills and the gh-proxy.com mirror carry no security guidance. External effects and rollback score 1: the auto-commit rewrites README/site with no documented rollback or human review. Source attribution scores 2: every third-party entry carries a GitHub link.
Internal inconsistency: the title and badge say 440+, the fact summary says 438+, and the comparison table says 438+ curated plus 20 original — the scale is not stated consistently, capping this at 1. Dependency availability: install relies on git, cp and ~/.claude/skills with no version or prerequisite notes, 1. Failure messages: workflows use set -euo pipefail and reject non-branch refs, but no user-facing install failure guidance exists, 1.
Audience and scenarios are clear (Chinese developers on Claude Code/Codex/Gemini CLI/Cursor) with detailed scenario categories, 2. Capability boundaries: no statement of where each skill applies, where it does not, or how it differs from official skills, 1. Trigger precision: no SKILL.md trigger descriptions are in evidence, so invocation behavior cannot be judged, 1. Environment fit: only macOS/Linux-style paths and a China mirror are given; Windows and version differences are unaddressed, 1.
Information architecture is clear (quick nav, category anchors, comparison table), 2. Install notes cover clone/cp/ZIP/mirror, 2. Naming stability: entries mix Chinese/English and repo styles and the scale numbers drift, 1. Examples and FAQ: a 30-second quick start and five recommended skills exist, but no FAQ or sample outputs, 1. Known limitations: entirely absent, 0. License: full MIT text present, 2. Versioning/changelog: no CHANGELOG or version number, only a badge date, 0. Maintenance responsibility: a MAINTENANCE.md link and auto-maintenance workflow exist, but publisher identity is unverified and the responsible party is unclear, 1.
Output usability: usable as a navigation list, but the original skill content is not in evidence, so output quality cannot be confirmed, 1. Marginal value: heavy overlap with anthropics/skills and awesome-claude-skills; Chinese and domestic-platform coverage is differentiating but the evidence does not show a large increment, 1. Cost-benefit: copy-and-install is cheap, but users must vet 440+ entries for quality and safety themselves, so the payoff is uncertain, 1.
Claim traceability: star counts are labeled as point-in-time but carry no timestamp or method, and 440+/438+ cannot be checked, 0. Cross-source corroboration: README is the only source with no independent confirmation, 0. Fact/inference separation: promotional claims ('most practical', 'best resource site') are mixed with facts without distinction, 0.
- No original SKILL.md content is in evidence, so permissions, triggers and actual output quality of the skills cannot be verified; review each before installing.
- Scale figures are inconsistent (440+ / 438+ / 438+ curated plus 20 original) and star counts are point-in-time with no timestamp; do not treat them as quality signals.
- The install path includes the third-party mirror gh-proxy.com, and the many listed third-party skills are not security-reviewed, creating supply-chain risk.
- The auto-maintenance workflow commits README/site changes directly to main with no documented human review or rollback.
- Publisher identity is unverified and the responsible maintenance party is unclear.