stormzhang/ai-coding-guide: A 92-Article Chinese Course for Codex and Claude Code

What ai-coding-guide actually is, and who it is written for

This repository is not a program. It is a Chinese-language course about two AI coding CLIs, Codex and Claude Code, published as 92 articles totalling roughly 520,000 characters according to the README. The stated audience is people with zero background: the README says the material suits "0 基础学习", starting at installation and continuing to engineering practice. The Codex track has 39 articles and is described as the main recommendation; the Claude Code track has 53 and is kept alongside it.

The problem it addresses is real and specific. Codex and Claude Code both ship fast, and their documentation is written for people who already know what a CLI, a sandbox, an approval prompt and an MCP server are. A beginner in the Chinese-speaking world faces two gaps at once: the vocabulary, and the fact that most walkthroughs assume English-language tooling. The README claims every feature, command and default behaviour was checked against the official Codex and Claude Code documentation rather than third-party guesses, and that each concept is rewritten in a three-part pattern of scenario, everyday analogy, then the actual case. That claim is the repository's central promise, and it is also the thing a reader cannot verify from the repository alone, since no source citations per article are described.

The material is text and images, not code. The top level holds claude-code/, codex/ and deepseek-harness/ directories, two README files, a licence and one og.png. There is no build system, no test suite and no package manifest described in what is available. Treat the repository as a book with a Git history, not as a library you install.

The Codex track: four entry points, AGENTS.md, sandbox approval and config.toml

The Codex half is organised around mechanisms rather than around a feature list. Article 01 covers what Codex is and its four entry points; article 02 covers core concepts; article 03 handles installation and login on Mac, Windows and Linux. From there the track moves into AGENTS.md, sandbox approval, config.toml, Chronicle memory, Worktrees, and a dedicated article on migrating from Claude Code.

That ordering is the useful part. Sandbox approval and config.toml are exactly the two areas where a new Codex user gets stuck, because they determine what the agent is allowed to touch and how its defaults are overridden. Putting them after installation and before the practical tasks means the reader meets the constraints before writing prompts, not after something has already run against the wrong directory. The presence of a migration article is a telling choice: it assumes a reader who already used Claude Code and now wants to move, which is a more realistic starting point than a blank slate.

The track also covers connecting third-party models, including DeepSeek, in article 05, and subscription and billing in article 04. The deepseek-harness/ directory at the repository root suggests supporting material for that path, though the README does not describe its contents. What the README does not state is how often individual articles are revised as Codex changes. The last push to the repository was on 2026-09-02, which tells you the repository was touched recently, but a push can be a README edit as easily as a rewrite of the sandbox article.

The Claude Code track: 53 articles from /init to Agent SDK and GitHub Actions

The Claude Code half is broader and reads like a reference sequence. It opens with what Claude Code is, installation, how it works, and the API configuration choice between subscription login and an API key. It then covers third-party and domestic model access, the Coding Plan and billing, and a first run.

Integration surfaces are covered in separate articles: VS Code, JetBrains, the desktop app, and a web and cloud article about running it in a browser and on a phone. Project-level concerns get their own entries: /init and CLAUDE.md generation, project structure, context management, permissions, and a security article that asks the blunt question of whether you should trust an AI with your code. The extension mechanisms follow in order: MCP, subagents, plugins, memory, Agent Skills, skill-creator, and agent teams for multi-session work. Article 30 is a decision guide comparing CLAUDE.md, Skill, Hook, MCP and Subagent, which is the article most likely to save a reader from over-engineering a setup.

Configuration and control come next: settings.json at user and project level, output styles, hooks, a CLI reference, modes, slash commands, checkpoints, a plugins reference, environment variables and Git workflow. Automation appears in the GitHub Actions article, and programmatic use in the Agent SDK article. The track closes with a capstone project, best practices, anti-patterns, a troubleshooting FAQ and a glossary. An anti-patterns article and a troubleshooting article are the two entries that most tutorial collections omit, and their presence here is a genuine differentiator.

How to use ai-coding-guide: reading it, not installing it

There is nothing to install from this repository. The README points readers to the online version at coding.stormzhang.ai, described as a dark terminal-style reading experience, and that is the intended path. The repository itself is the source of the articles and the images.

If you want a local copy, the standard Git route applies. This clones the default branch and gives you the two article directories:

git clone https://github.com/stormzhang/ai-coding-guide.git
cd ai-coding-guide
ls claude-code codex

The README does not document a build step, a static site generator or a preview command, so expect the directories to contain article sources and assets rather than a runnable site. For the actual first use, follow the README's own routing rather than browsing the file tree: it sends beginners to the Codex track first.

The README gives the entry points as two URLs, one per track:

https://coding.stormzhang.ai/codex/
https://coding.stormzhang.ai/claude-code/

Start at the Codex index if you are new. If you already use Claude Code and want to move, the Codex track's migration article is the intended shortcut, and the README does not describe a reverse migration path.

Where ai-coding-guide is the wrong tool

The first limitation is language. Everything described is in Simplified Chinese, with README.en.md as the only English entry point. If you cannot read Chinese, the 92 articles are inaccessible regardless of their quality, and a translated README does not change that.

The second is that this is a tutorial, not a specification. The README claims the content was verified against official Codex and Claude Code documentation, but the repository does not describe per-article source links, a changelog, or a versioning scheme for the articles. There are no releases. When Codex changes a sandbox default or Claude Code changes a settings.json key, nothing in the repository structure tells you which article is now stale. The last push was on 2026-09-02, which shows activity but not coverage.

The third is pace. A 92-article course is a commitment. If you need to answer one question, such as which permission mode to use for a specific repository, the official documentation is faster and more authoritative. This collection earns its place when you want the sequence and the beginner-level rewriting, not when you want a single fact.

Finally, the repository is documentation about tools that are themselves fast-moving and commercial. Billing, subscription tiers and model access are covered in both tracks, and those are the sections most exposed to change. The README cannot make pricing stable, and neither can any tutorial.

Alternatives: official docs, and what changes when you switch

The direct alternative is the official documentation for each tool. The README itself links to developers.openai.com/codex and code.claude.com/docs/zh-CN as its fact sources. The difference in approach is fundamental: official docs are maintained by the vendor, versioned with the product, and authoritative on defaults and flags, but they assume you know the vocabulary and they are not organised as a learning path. This repository inverts that: it sequences the material for a beginner and rewrites the concepts, at the cost of being a second-hand account that you must re-check.

A second comparison is the deepseek-harness/ directory inside this same repository, which points at a different concern: running these CLIs against non-default models. That is a configuration problem rather than a learning problem, and it is covered in article 05 of each track rather than as a standalone course.

For English-speaking readers, the honest alternative is a general English-language guide to AI coding CLIs, or the official docs plus experimentation. The trade-off is identical to any translated technical material: you gain a curated path, you lose the ability to check the original wording without opening the vendor docs anyway. Given that the README's own selling point is fidelity to official sources, keeping the official page open in a second tab is not a compromise of the method. It is the method.

Maintenance, licence and the cost of keeping up

The repository is not archived and the last push was on 2026-09-02. That is recent enough that the material is likely to reflect the current shape of both CLIs, but the repository publishes no releases, so there is no version boundary to pin your reading to. If you fork it or quote it, record the commit you read.

The licence is MIT, stated in the README badge and present as a LICENSE file at the repository root. MIT is permissive: reuse and adaptation are allowed subject to the licence terms. If you plan to republish the articles, translate them, or fold them into internal onboarding material, read the LICENSE file and the attribution expectations it sets. That is a description of the licence, not legal advice.

The upgrade cost is the interesting part. There is no dependency to bump and no migration to run. The cost is editorial: every time Codex or Claude Code changes a default, an approval flow or a config key, one or more of the 92 articles may need a rewrite, and the repository gives no signal about which. A reader who wants to stay current has to re-read the specific article for the feature they use, and check it against the official docs. That is a low ceiling on maintenance effort and a high ceiling on reader diligence.

Editorial conclusion

Adopt it if you or your team read Simplified Chinese and want a structured path through Codex or Claude Code rather than scattered English blog posts; the Codex track is what the README recommends as the main line. Skip it if you need English-language material, or if you want a runnable tool rather than prose. Before relying on any page, open the official Codex or Claude Code documentation alongside it and check the specific command, flag or config key you intend to use, because the repository carries no release history and no changelog for the 92 articles.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. Project website
4. README
5. stormzhang/ai-coding-guide on GitHub