yinguangyao/blog: a Chinese frontend engineering notebook kept in GitHub Issues

What the repository actually is

There is no source tree to install. The README is an index: a list of article titles, each one a link to a numbered GitHub Issue in the same repository, with a year marker next to it. The description says the project is a summary of JavaScript frontend development and work experience. The author identifies himself as null and lists mobile web development, the React stack, frontend engineering and web rendering as his areas. The README also points to a Zhihu column and a paid course on imooc, so the repository functions partly as a portfolio and partly as a distribution surface for writing that also appears elsewhere. That matters for how you use it: the GitHub Issues are the canonical copy of some articles, while others link out to Zhihu entirely, and a few use reference-style links defined at the bottom of the README rather than inline URLs.

The Issues-as-posts structure and what it costs you

GitHub Issues give the author comments, labels, permalinks and version history for free, and they render Markdown and code blocks without any static site generator. The trade-off is that there is no navigation beyond the README index, no tags or categories inside GitHub itself, and no search that respects the grouping the README imposes. If you want the article on KonvaJS internals, you find it in the Canvas section of the README and follow the link to issue 68. The year markers in the index are the only recency signal, and they are not sortable. A reader who wants everything about React has to scan the React section manually. This is a reasonable structure for a personal archive and a poor one for a reference site. The repository also carries topics such as agent, mcp, claude-code, konvajs and webpack, which suggest the intended scope, though the README grouping is the only real taxonomy.

The four content clusters worth knowing about

The material splits into recognisable groups. The oldest cluster, from 2018, is an eight-part walkthrough of the Underscore source, covering its overall architecture, base methods, array traversal, throttle, debounce, index lookup and template engine. The middle period, 2019 through 2022, is the largest: React state management comparison, from-scratch implementations of Redux, Mobx and React itself, React Hooks mechanics, React Diff, ES6 class and extends semantics, the module system, Promise and generator internals, decorators, Proxy, and design patterns such as strategy, state machine, AOP and IoC. The work-experience section covers Tencent Docs rendering-layer feature design, async chunked computation, mobile web development, JS Bridge, SSR, and a Nuxt isomorphic problem. Canvas has four entries, including a Leaferjs introduction and a piece on Canvas rendering engine design. The newest cluster, 2025 and 2026, is about AI: MCP explained from scratch, a nanobot and openclaw design analysis, command-line redesign for agents, five coding agents packaged into one npm package, and an eight-part series on implementing a coding agent from a single LLM call through streaming, tool calling, the ReAct loop, real coding tools, parallel tool calls, Skills, and session persistence.

The coding agent series is the part with a concrete build path

The eight-part series is the only content in the repository that reads as a guided implementation rather than an explanation. The progression in the titles is explicit: one LLM call, then streaming responses, then tool calling, then the ReAct loop, then real coding tools, then parallel tool calls, then Skills, then session persistence and multiple sessions. Each step names the mechanism it adds, which means a reader can stop at the point that matches their own project. If you only need streaming, part two is the boundary. If you are building something that manages multiple conversations, part eight is where that concern is addressed. The repository does not state which language or SDK the implementation uses, so that has to be confirmed by opening issue 93. The same caution applies to the MCP article and the nanobot piece: the titles describe the subject, not the stack.

Where this is the wrong resource

Nothing here is versioned against a dependency you can pin. A 2020 article on Babel's handling of class and extends describes a toolchain that has changed since, and the README gives no revision notes. There are no releases, so there is no changelog to tell you when an article was corrected. If you need current API documentation for React, webpack or KonvaJS, this repository will not provide it and does not claim to. It is also not a substitute for a maintained book or course: the structure is a flat list, and the depth varies by article. Some entries are full source walkthroughs, others are short opinion pieces such as the one asking whether server-side rendering is necessary. Treating the index as a curriculum will disappoint you, because the ordering is by topic and year, not by difficulty.

The alternative: a maintained book or an official guide

The closest alternative in kind is a published frontend engineering book, or the official documentation for whichever library the article covers. The difference in approach is maintenance. A book has an editor, a revision cycle and a stated edition; React's own documentation has versioned pages and migration guides. This repository has a single author, no release process and a last push date that tells you when the README was touched but not which article changed. If you want a single explanation of, say, React state management trade-offs, the repository's comparison piece may be the more direct read, but the official docs will be the one that stays correct after the next major release. Use the repository for the reasoning and the from-scratch implementations, which age more slowly than API surface, and use official sources for anything you are about to put in production.

Maintenance, licence and what to verify before you rely on it

No licence is stated in the supplied material, and the repository is not archived. That combination means the default copyright position applies: the writing is the author's, and you should not assume you can republish or translate it without asking. The README gives a WeChat contact, gioryyin, which suggests the author is reachable, but there is no stated contribution process and no issue template for corrections. The last push is dated 2026-07-11, so the repository is active as of that date, but the index contains entries going back to 2018 and the recency of the push says nothing about the recency of any individual article. Before citing or building on a specific piece, open the Issue, check its year marker in the README, and confirm that the libraries it names still behave the way the article describes.

Editorial conclusion

Adopt this as a reading list if you work in Chinese-language frontend engineering and want long-form explanations of React internals, Canvas engine design, or the step-by-step construction of a coding agent. Do not treat it as a dependency, a package, or a source of API documentation. Before relying on any single article, open the linked Issue and check its date, because the earliest entries are from 2018 and the JavaScript ecosystem has moved since. The eight-part coding agent series (issues 93 through 100) is the most current material in the repository and the most likely to still match the tools it describes.

Official sources

1. Issues
2. README
3. yinguangyao/blog on GitHub