Claude Code Best Practice
shanraisshan/claude-code-best-practiceA large reference repo of Claude Code tips, concept guides and sample agents, commands and skills; a guide to read, not a skill pack to install.
With caveats
Check first Depth 1/5
awesome-agentic-ai-zh Claude skill
A trilingual (CN/EN) learning roadmap for agentic AI with 240+ curated links, hands-on exercises and one example skill, tool-calling-tutor.
Our verdict Recommended with caveats
A well-kept learning map with one useful tool-calling debugging skill; do not expect a skill library, and the course content is mainly Chinese.
What awesome-agentic-ai-zh does
This is mainly a course-style repository, not a skill pack. It lays out a path from LLM basics to reliable agent systems in numbered stages (foundations, LLM basics, prompt design, tool use and a first agent loop, the Claude Code ecosystem, up to evaluation and reliability), with two tracks: using CLI agents such as Claude Code, Codex and OpenCode, or building your own agents. A third route covers everyday users who do not code. Each stage has explanations, curated external resources and small exercises, and the README points to outside material such as Datawhale's Hello-Agents rather than rewriting it. The site is also published as online docs.
The only agent skill is examples/stage-5/tool-calling-tutor. It activates when a tool-calling agent fails to call a tool, passes wrong arguments or loops without stopping. It triages the symptom into four branches, then guides a five-step schema repair, with 12 reference files, one script and two tests. The skill text is bilingual Chinese and English. The repo has CI checks on links and licenses, and the main branch was updated in late October 2026. If you want a library of skills, look elsewhere; if you want a structured Chinese-language curriculum, this is a good fit.
Good for
- Chinese-speaking learners starting with AI agents
- Debugging tool-calling schemas and ReAct loops with the example skill
- Teams building an internal agent curriculum
Not for
- People who want a broad set of installable skills
- Readers who need English-only instructions for the skill
What's inside depth 3/5
One 8 KB skill with 12 reference files, a script and two tests, inside a large documentation repo; a single skill is limited in scope.
- Skills
- 1
- Reference docs
- 12
- Scripts
- 1
- Tests / evals
- 4
- Templates
- 0
- Avg SKILL.md
- 7.8 KB
Counted from the repository file tree. How we score depth.
Dependencies and costs
- Python, Git and an LLM API key or local model for the exercises
Security review
Review before installingA Chinese-language awesome list and learning repo whose single installable skill, tool-calling-tutor, is a diagnosis guide with reference files and an offline eval checker. The skill has no allowed-tools frontmatter, no fetch instructions and no network or credential handling in what we have; the README tells you to copy its files into your skills folder by hand. Much of the repository was not available for review.
What we checked
The repository (636 files) on Oct 29, 2026. Not a full review: 545 of 636 files were not downloaded (large repository; we reviewed the skill folders); it relies on code outside the repository: About 545 readable repository files (documentation, GitHub templates, other examples) that are not in the local copy; the installable skill is examples/stage-5/tool-calling-tutor, whose SKILL.md, references and README were opened.
NVIDIA SkillSpector static scan: 13 pattern hits (Agent Snooping 6, Rogue Agent 4, Data Exfiltration 2, Excessive Agency 1). The 13 hits are documentation: an api.star-history.com badge URL in the README, cp commands in the tutor README that copy the skill into ~/.claude/skills, and 'retry forever' as a described failure mode in a debugging flowchart.
A review, not a certification. Skills change after we check them — read the files before installing.
License
MIT — commercial use allowed.
Can it run unattended?
Not really. The skill is an interactive diagnosis guide and the rest of the repo is learning material.
Alternatives to awesome-agentic-ai-zh
Claude Code Guide (zebbern)
zebbern/claude-code-guideA long community README guide to Claude Code plus 79 mixed skills: general developer helpers, document tools and a large group of security-testing playbooks.
With caveats
Check first Depth 2/5
Anthropic Agent Skills
anthropics/skillsAnthropic's reference skills repo: 20 skills including document tools (docx, pptx, xlsx, pdf), frontend-design, skill-creator and mcp-builder.
Recommended
Check first Depth 4/5
FAQ
What is awesome-agentic-ai-zh?
It is a learning roadmap for agentic AI in Traditional Chinese, Simplified Chinese and English, with ordered stages, 240+ curated resources and runnable exercises. It also includes one example agent skill for debugging tool calling.
How do I install the tool-calling-tutor skill?
Copy examples/stage-5/tool-calling-tutor into your agent's skills folder. It triggers when a tool-calling agent does not call a tool, sends wrong arguments or loops, and it does not cover MCP servers or production observability.
Is it only for Chinese speakers?
The README and stages are available in three language versions, so English readers can follow the roadmap. The example skill's own instructions are written mainly in Chinese with English technical terms.
Is awesome-agentic-ai-zh safe to use?
It is mostly documentation. The one skill has no network calls in the text we read, and the license is MIT.
Sources: GitHub repository (stars, license, file tree, last push Oct 29, 2026). Review based on the README and SKILL.md files as published.