nie-grassroots-logic: a Cursor and Claude Code skill for reading county-level Chinese governance

What nie-grassroots-logic is for, and who it is not for

The README frames the project as an executable AI skill built on Nie Huihua's book on the operating logic of grassroots China. Its stated job is to turn a county and village governance framework into something an agent can call repeatedly, for two purposes: explaining local news and power structure, and making personal choices about study, civil service exams, investment, elder care and starting a business.

The intended readers are listed explicitly: students and parents comparing cities, exam candidates and people already inside the system weighing postings, founders dealing with park recruitment and subsidies, buyers and regional observers watching land finance and rezoning, people considering a return to village governance, and journalists or researchers who want a causal chain instead of vague commentary.

The repository is equally direct about what it is not. It is not a PDF or a full-text copy of the book, not a reading summary, not inside information or personnel gossip, not stock or property advice, and not a general China-context essay. The README repeats that it contains no full text and asks readers to buy the legitimate edition. If what you want is the book's argument with its data and cases, this repository is the wrong object.

How the skill loads: one entry file plus lazy references

The architecture is a routing problem, not a model. A single SKILL.md holds trigger words, a decision tree and a routing table. Behind it sit 14 modules grouped into three layers plus a personal and applied group: layer I covers state institutions (dual-equilibrium, tiaokuai-and-rank), layer II covers the exercise of power (power-and-positions, village-fit), layer III covers grassroots governance (local-governance, development-dilemmas). The applied group holds life-decisions, rezoning-household, business-environment, bottom-up-project, innovation-risk-boundary, mega-project-contest, playbooks and anti-patterns-voice.

The README describes two behaviours that matter in practice. First, the agent recalls the skill automatically from scene and trigger words in the SKILL.md frontmatter, so the user does not have to name a module. Second, the agent opens only the relevant references/*.md file rather than loading everything, which keeps context from being flooded. INDEX.md is the navigation map across the whole set.

The design principles are stated as purifying structure rather than writing summaries, keeping the author's exact names (the README cites terms such as 含权量, 三座大山 and 先上车再买票), building for invocation, and a discipline it calls evidence: not inventing frameworks or numbers the book does not contain, and marking current implementing rules as pending verification. That last principle is the one to hold the project to, because the framework names are stable while the local rules they are applied to are not.

Installing the skill and confirming the agent actually reads it

The README gives two installation paths, both aimed at a skills directory. The first is a clone into the Cursor skills directory. The folder name has to match the skill's name, nie-grassroots-logic, which is why the clone target is spelled out in full.

git clone https://github.com/ayi-ai/nie-grassroots-logic.git \
  ~/.cursor/skills/nie-grassroots-logic

If you already have a clone somewhere else, the README offers a symlink instead of a second copy.

ln -s /path/to/nie-grassroots-logic ~/.cursor/skills/nie-grassroots-logic

Verification is behavioural rather than a version check. Open a new conversation and type one of the sample prompts. If the agent goes off to read the skill's SKILL.md or references/ before answering, the README treats that as a successful install.

用聂辉华的框架帮我分析：我们县为什么一边化债一边还在招商？

For a more controlled run, the README recommends naming the skill explicitly and stating the output shape you want, which is how you get a causal chain, an option comparison, risks and a list of things you must verify locally rather than a general essay.

用 nie-grassroots-logic / 基层中国的运行逻辑 skill：
[粘贴新闻 / 你的选择题]
请按框架给：因果链 + 选项对比 + 风险 + 我需要本地核实的清单

Advanced use narrows the reading list. The README's example asks the agent to read life-decisions and power-and-positions first, then answer a question about taking a township posting, and to use the red-yellow-green boundary in innovation-risk-boundary to judge whether a pilot can be pushed. Note that the README documents no uninstall step, no update command and no versioning beyond the skill ID; upgrading means pulling the clone again.

The trade-off: a toolkit with no text, no data and dated rules

The central limitation is structural. The repository holds frameworks and routing, not evidence. If a question turns on a specific figure, a case detail or a full argument, the README itself says to read the original book, and it says to cite Nie Huihua's book in academic writing with this repository only as an auxiliary tool. There is no dataset, no scraping pipeline and no live source in the layout: the top level is .gitignore, EXTRACTION.md, INDEX.md, LICENSE, NOTICE, README.md, SKILL.md, assets/, cover.png and references/.

The second limitation is currency. The README's own design principle marks current implementing rules as pending verification, which means the skill can name a mechanism such as land finance or the three mountains while the specific rule you are asking about may have moved. The README also states plainly that it is not legal advice, not investment advice and not inside information, so any output is a structured hypothesis about incentives, not a fact about your county.

The third limitation is deployment. It is built for Cursor, Claude Code, Codex and Grok style agents that read a skills directory. If your workflow is a plain chat window with no filesystem access, or a model that does not read SKILL.md frontmatter, none of the lazy loading happens and you are left with a folder of Markdown. There is also no release history in the repository, so there is no changelog to check what a given framework definition looked like earlier.

How it differs from reading the book or asking a general model

The obvious alternative is reading Nie Huihua's book directly, and the README treats these as complements rather than substitutes: use the skill for fast judgement and news analysis, use the book for complete argumentation, data and case detail. The practical difference is that the book gives you the reasoning once and the skill gives you a named framework the agent can reapply to a new province, a new posting list or a new announcement without you restating the theory each time.

The second alternative is a general-purpose model with no skill installed. There, the failure mode the README is written against is fluency without structure: the model will produce a plausible paragraph about local government incentives, but it will not route to tiaokuai-and-rank for a question about territorial responsibility, or to rezoning-household for a question about converting a county into a district. The skill's contribution is the routing table and the fixed vocabulary, including the anti-patterns-voice module the README offers for writing that avoids empty AI phrasing. Whether that discipline holds depends on the agent actually reading the reference file, which is exactly what the install check tests.

Editorial conclusion

Adopt it if you already use Cursor, Claude Code, Codex or Grok and want a named framework applied to local news, a job-choice comparison or a rezoning question, and accept that it is a methodology layer with no original text and no data. Do not adopt it if you need the book's full argument, citable page numbers or current policy detail, because the repository states it carries neither the full text nor confirmed local rules. Before relying on it, clone the repository into ~/.cursor/skills/nie-grassroots-logic, open a new conversation, and check whether the agent actually reads SKILL.md and references/ when you ask why a county keeps selling land while repaying debt.

Frequently asked questions

Official sources

1. ayi-ai/nie-grassroots-logic on GitHub
2. Issues
3. License: MIT
4. README