# nie-grassroots-logic turns a book about county government into a routing table, not a summary > An agent skill that distills the governance frameworks from a Chinese academic book into fourteen named modules, routed by trigger words and lazily loaded so a question pulls in only the references it needs. The interesting engineering is the packaging rather than the content: a skill file that decides which reference to open, and a design principle that forbids inventing frameworks or numbers the book does not contain. **ayi-ai/nie-grassroots-logic** — 聂·基层运行逻辑 · Agent Skill：基于聂辉华《基层中国的运行逻辑》的方法论工具箱（不含原书全文） - Repository: https://github.com/ayi-ai/nie-grassroots-logic - Stars: 560 · Forks: 65 - Language: HTML - License: MIT - Published: 2026-09-17 · Updated: 2026-09-17 · Language: en - Canonical page: https://hysenlabs.com/projects/ayi-ai-nie-grassroots-logic ## The book's structure is distilled, and the summary is deliberately absent This is an executable skill built from one Chinese-language book on how county, township, and village government actually operates. The repository is emphatic about what it is not: not a digitised copy of the book, not a reading summary or a review, not insider information or personnel gossip, not stock or property recommendations, not legal advice, and not a general recipe of national conditions. What it offers instead is named frameworks, the situations in which each applies, and how to use it. The stated method is to distil structure rather than write summaries, to keep the author's own precise names for each framework, and to build things that can be invoked rather than read. The fourth principle is the one that matters most for trusting output: do not invent frameworks or figures the book does not contain, and mark present-day rules as still to be verified. ## A skill file routes, and the references load only when needed The packaging is the part worth studying if you build skills yourself. The book's three-layer logic is split into an entry point plus a set of lazily loaded references. One skill file holds the trigger words, a decision tree, and a routing table; everything else lives in separate reference documents that the agent opens only when a question needs them. The stated reason is context economy, since opening everything at once would fill the window with material the question does not require. Recall is automatic and comes from the scenarios and trigger words declared in the skill file's front matter, so the routing decision is made by metadata rather than by the agent reading a directory listing. A separate index document acts as the master map, which means there is a human-maintained table of contents for the machine's benefit rather than your own. ## Fourteen modules in three groups, plus a fourth for personal decisions The routing table divides the material into three analytic groups and one applied group. The first covers national institutions and holds the dual equilibrium between internal and external stability, alongside the framework where vertical lines and horizontal blocks combine with the block dominant, which is where responsibility for local accountability problems tends to land. The second covers how power operates, including a formula for the power content of a post expressed as rank multiplied by whom you manage multiplied by how much money you control, plus a module on fit for village work. The third covers grassroots governance and its development dilemmas. The fourth is the applied group: life decisions covering education, civil service exams, investment, retirement and starting a business, rezoning effects on households, business environment, bottom-up project promotion, an innovation risk boundary with red, yellow and green zones, mega-project contests, a set of fast-path playbooks, and a module on avoiding formulaic writing. ## Installation targets a skills directory, and the folder name has to match There is no package to install. The repository is cloned, or symlinked, into the skills directory your agent reads: ```bash git clone https://github.com/ayi-ai/nie-grassroots-logic.git \ ~/.cursor/skills/nie-grassroots-logic ``` An existing clone can be linked instead of copied, which is what you want if you intend to pull updates. One hard requirement is stated plainly: the folder name has to be identical to the skill's declared name, nie-grassroots-logic, or the agent will not find it. That single string is the join between the repository and the runtime, and there is no validation step that tells you it is wrong. The documentation describes installation for Cursor, Claude Code, and comparable tools, and the description names Codex and Grok as additional callers. ## Verification is behavioural: ask a question and watch what the agent opens There is no install confirmation command, and the test is whether the agent reaches for the files. The documented procedure is to open a new conversation and type a specific question, phrased as a real scenario rather than a test string: ask it to analyse with the author's framework why a county is resolving debt while still pursuing investment. If the agent reads the skill file or something in the references directory, the installation worked. If it answers from general knowledge instead, the folder is in the wrong place or the name does not match. That is a weaker signal than a version check, and it is a better one, because it verifies the thing you actually care about, which is whether the routing fires. ## Six prompting styles, from plain language to naming a module The usage section is the longest part of the documentation, and it opens by saying you do not need to memorise module names. Asking in plain language is the recommended path, and a table maps five example questions to the frameworks that usually get pulled in: why a county is selling land maps to the land finance loop and the three pressures, choosing between a township post and a ministry maps to power content and promotion, a county becoming a district maps to the rezoning module, negotiating a park deal maps to the government-enterprise quadrants, and an empty station after a high-speed line opens maps to the mega-project module. Four stronger styles follow: naming the skill explicitly, restricting to one dimension so the answer stays focused, naming specific modules to read first, and supplying context. That last one is the quality lever, and the four things it asks for are the administrative level of the place, your own role, the material you have, and the form of output you want. ## The frameworks keep the author's names, and two carry red-line warnings A quick reference table names each framework in one line and pairs it with the situations it applies to. Among them: the dual equilibrium between internal and external stability; vertical lines combined with horizontal blocks, block dominant; resource allocation by rank, summarised as people following resources and resources following power; the power content of a post as rank times whom you manage times how much money; the three pressures on local officials, namely maintaining stability, attracting investment, and taking on debt; the pressure-driven system with attention allocation, covering indicator inflation and single points of veto; the government-enterprise quadrants of cooperation, separate governance, collusion and harm; the land finance loop where cheap industrial land cross-subsidises expensive commercial and residential land; and the trade-off between urban dynamism and order. Two entries carry warnings in the table itself: the Kunshan three axes must be used with red lines added, and the west-to-east gas pipeline sequence is presented as a nine-step script rather than a rule. ## A MIT repository with a NOTICE and an extraction record beside it The tree is small and shaped like a skill rather than an application: a skill file, an index, an extraction document, a notices file, a licence, an assets directory, and the references directory that holds the lazily loaded material. The extraction document is the interesting one, because it is the audit trail for a project whose main risk is fabrication, and it sits next to a notices file that would otherwise be redundant under a plain MIT licence. The language the repository reports is HTML, which comes from how the documentation is laid out rather than from the skill's own structure. The last push was 2026-08-10 and there are no releases, so there is no version to cite and updates arrive as commits to the default branch, which matters most for the symlink install path since that one tracks the working copy. ## Conclusion Use it as a thinking checklist rather than an oracle. The strongest thing in it is the routing discipline, since a skill that names when a framework applies and refuses to invent evidence is more useful than one that answers everything, and the project is explicit that it is neither legal nor investment advice. Two constraints to hold onto. It is a methodology, not the book, so anything needing full argument, data, or cases means reading the original. And its own stated practice is to mark present-day rules as needing verification, which means you are expected to check the current detail locally before acting. ## FAQ ### What is the nie-grassroots-logic skill? A methodology toolbox distilled from a Chinese academic book on how county, township, and village government operates, packaged as an agent skill for tools such as Cursor, Claude Code, Codex, and Grok. It is used both to explain local news and power structures and to frame personal decisions about education, civil service exams, investment, retirement, and starting a business. ### How do I install nie-grassroots-logic? Clone the repository into your agent's skills directory, for example the Cursor one, or symlink an existing clone into that same path. The folder name must be identical to the skill's declared name, nie-grassroots-logic, or the agent will not load it. ### Does nie-grassroots-logic include the original book? No. It is a methodology skill and not an electronic version of the book, and it contains no full text. The stated method is to distil structure rather than write summaries, and the documentation asks you to buy the legitimate edition to support the author. ### Which frameworks does nie-grassroots-logic use? Named frameworks from the book, including the dual equilibrium between internal and external stability, vertical lines combined with horizontal blocks with the block dominant, resource allocation by rank, the power content of a post as rank times whom you manage times how much money, the three pressures of stability, investment and debt, the government-enterprise quadrants, the land finance loop, and the top-down and bottom-up co-governance path. ### What are the limits of nie-grassroots-logic? It is explicitly not the original text, not a summary, not insider information, not stock or property recommendations, and not legal advice. Its design principles include not inventing frameworks or figures the book does not contain and marking present-day rules as still to be verified, and for academic citation you are told to cite the original book rather than this repository. ## Sources - [ayi-ai/nie-grassroots-logic on GitHub](https://github.com/ayi-ai/nie-grassroots-logic) - [Issues](https://github.com/ayi-ai/nie-grassroots-logic/issues) - [License: MIT](https://github.com/ayi-ai/nie-grassroots-logic/blob/main/LICENSE) - [README](https://github.com/ayi-ai/nie-grassroots-logic/blob/main/README.md) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/ayi-ai-nie-grassroots-logic