maoxuan-skill: a Claude Code skill that answers your business questions in a borrowed voice

A prompt package, not a library

The repository is unusually small for a project with a following. The top level holds four entries: `LICENSE`, `README.md`, `SKILL.md` and a `references/` directory. There is no source code, no build, no test suite, and the project publishes no GitHub releases. The primary language is not even recorded as a known one in the project metadata, because the artefacts are prose.

What it is, then, is a skill for Claude Code. The badge in the readme points at the Claude Code skill format, and the installation is a single command:

npx skills add leezythu/maoxuan-skill

After that, the skill is triggered by phrases such as 毛选, 教员, 用毛选的方式分析, 教员怎么看, and then the user asks a question in ordinary language. The readme shows three worked examples, one about a startup losing users to a larger competitor, one about two technical leaders who do not get along, and one about a backend developer feeling unqualified for a career move into AI.

The author's own framing is that this is not role play. The claim is that the useful artefact is the method: 矛盾分析法 for decomposing a problem, 统一战线 for handling internal conflict, 纸老虎论 and 农村包围城市 for competitive position, and the readme says explicitly that the point is to help you see your problem the way those frameworks see it, rather than to recite quotations.

That distinction is the whole evaluation, and it is the one thing a reader has to judge for themselves.

Seven models, ten heuristics, and the research files behind them

The distilled content is listed explicitly, which is useful because you can judge the scope before installing anything.

The seven mental models are 矛盾分析法, finding the principal contradiction as the way to break a deadlock, sourced to 《矛盾论》; 实践认识循环, the practice and knowledge cycle, that without investigation there is no right to speak, from 《实践论》; 持久战略, trading time for space when strength is unfavourable, from 《论持久战》; 农村包围城市, not attacking the strongest point first but building a base in the periphery, from 《星星之火，可以燎原》; 统一战线, from 《中国社会各阶级的分析》; 群众路线, from 《关于领导方法的若干问题》; and 纸老虎论, strategically contemptuous and tactically serious about the enemy, from the talk with Anna Louise Strong.

The ten heuristics include 没有调查就没有发言权, 抓主要矛盾, 星星之火可以燎原, 一分为二看问题, 不打无准备之仗, and 自力更生为主争取外援为辅, among others. There is also a component the readme calls the expression DNA, which is a style layer rather than a reasoning layer.

The backing material sits in `references/research/` as six Markdown files: `01-core-writings.md` for the distillation of the core works, `02-strategic-thinking.md` for the strategic analysis, `03-expression-dna.md` for the expression style, `04-external-views.md` for outside assessments and influence, `05-decisions.md` for key decision analyses, and `06-timeline.md` for a timeline.

So the repository is a written artefact with its research appendix, not a program. That is the honest description of what you are installing, and it is also the reason there is no versioning discipline to speak of.

An optional knowledge base for real citations

The skill works on its own, and the readme is clear about that: the core thinking framework is self-contained and usable after installation. What you lose without more is retrieval of the original texts and precise citations.

The optional route is a separate repository of the collected writings, cloned with:

git clone https://github.com/weiyinfu/MaoZeDongAnthology.git ~/MaoZeDongAnthology

Placed in the home directory, it is discovered automatically with no further configuration. If you want it somewhere else, clone to your path and point the skill at it:

git clone https://github.com/weiyinfu/MaoZeDongAnthology.git /你的路径/MaoZeDongAnthology
export MAOXUAN_KB_PATH=/你的路径/MaoZeDongAnthology  # 可加入 ~/.zshrc 持久化

The discovery order is documented: the skill checks the `MAOXUAN_KB_PATH` environment variable first, then the current directory, then parent directories, and finally a `MaoZeDongAnthology/` directory in the home folder. When no knowledge base is found, the skill still works and simply does not retrieve original passages.

That last sentence is the important one for an evaluator. Without the anthology, what the model produces is a paraphrase of frameworks it already knows, dressed in a particular voice. With the anthology, the same skill can quote the source. For a user who wants to check the reasoning rather than admire the tone, the second mode is the one worth having, and the difference between the two is one git clone away.

The project's own list of what it cannot do

The readme has a section titled 诚实边界, honest boundaries, and it is the strongest thing in the repository. It splits into what the skill can do and what it cannot.

It can do: decompose complex problems with the contradiction analysis method, analyse competitive situations with the protracted war and encircling the city models, guide product and organisational decisions with the mass line, map interpersonal and interest structures with the united front logic, and offer a dialectical, non-partisan perspective with 一分为二.

It cannot: replace the person, because political intuition and judgement about historical moments cannot be copied; produce concrete execution plans, because it offers an analytical framework and the specifics need to be combined with your actual situation; write poetry, because a way of thinking can be extracted but literary talent cannot be distilled; and evaluate history, because the skill focuses on methodology and does not judge specific historical events.

The line that closes the section is worth quoting in translation, because it is the correct posture for a prompt project: a skill that does not tell you where its limits are is not worth trusting.

There is a comparable honesty in the comparison with other personality skills the author names, Jobs, Musk and Munger. The argument is that their positions came with resources, capital and information advantages, whereas these frameworks were developed from a position of extreme weakness against an overwhelming opponent, and so they may apply better to a small team facing a large organisation.

Where the analogy breaks, and where it is the wrong tool

Three limits are worth stating, and they are about the artefact rather than the ideas in it.

The first is that this is a persona overlay on a general model, not a domain expert. The framework can be applied to a question, but the model doing the applying is the same model that would answer without the skill. A skill changes the framing and the vocabulary, and it can change which considerations come to mind first, but it does not add knowledge about your market, your codebase or your organisation. The readme's own boundary table says as much, in the row about not being able to produce concrete execution plans.

The second is that the mapping is a category shift. The source material is about building an insurgent organisation and fighting a war. The examples in the readme map it onto startup competition, team conflict and career anxiety, and the mapping is often illuminating. But a start-up's competitive position is not a guerrilla campaign, a senior engineer's disagreement with a peer is not a factional struggle, and reading them as such produces confident analogies that do not survive contact with a customer, a contract or a person who has to work with you on Monday. The skill helps you ask better questions, not to answer yours.

The third is the engineering side. There is no versioned release, no test, no changelog, and no issue tracker referenced in the readme. The content is prose in a directory, so an update to the repository reaches you only if you reinstall, and nothing in the artefact tells you which version of the framework you are getting. The `SKILL.md` file at the root is the entire behaviour, and it is the file to read before you trust it with a question that matters.

And the language matters. The materials are Chinese, the examples are Chinese, and the voices the skill produces are in Chinese. For a user working in another language, the output is a translation of a Chinese-language framing, which is a weaker instrument than reading the sources in the original.

Where it fits, and how the project is maintained

The readme also gives a scenario table, which is the most efficient way to judge whether this is worth your time. A startup facing a larger competitor maps to 农村包围城市 plus 持久战略 plus 纸老虎论. Deciding which of three directions to fund maps to 矛盾分析法. Internal team conflict maps to 统一战线. Being unsure about product direction maps to the practice and knowledge cycle, in other words investigate rather than guess. Facing a much stronger opponent maps to 纸老虎论 and 农村包围城市. Organisation and team management maps to 群众路线 and criticism and self-criticism. Career transitions map to the three stages of protracted war and the base of operations idea. Negotiation and cooperation map to 统一战线 and contradiction analysis.

That list is a fair description of where a framework like this is useful: a structuring conversation rather than a decision, and a vocabulary that makes the user articulate the principal contradiction rather than listing symptoms.

On maintenance, the last push to the repository was on 2026-08-19 and there are no GitHub releases, so there is no version stream to track and nothing to pin. The licence is MIT, described in the readme as free to use, change and build on, and the readme also carries a Star History badge. The author links two related skills as inspiration, 童锦程.skill for relationships and 女娲.skill for the methodology of distilling a person, and points at two other projects of their own, a Python reimplementation of the Claude Code architecture and a curated reading list on harness self-improvement.

The honest summary is that this is a well-documented, carefully scoped piece of prompt engineering with an unusually good limitations section, in a repository whose entire technical footprint is a Markdown file and a folder of research notes.

Editorial conclusion

maoxuan-skill fits a Chinese-speaking developer who wants a different analytical vocabulary for a strategy or organisational question, and who will read the generated answer as a prompt to think with rather than as advice to follow. It does not fit a team that needs domain expertise, a repeatable method, or output in a language the model reasons better in, since the artefact is a single Markdown skill file with a persona overlay and no tests. It is also the wrong tool if the question is a factual or technical one, where a retrieval-augmented answer over your own documents would be more useful than a framework. Before you install it, read the honest boundaries section, which is unusually candid, and clone the external knowledge base if you want citations from the original texts rather than paraphrase.

Frequently asked questions

Official sources

1. Issues
2. leezythu/maoxuan-skill on GitHub
3. License: MIT
4. README