huasheng13-skill: a Hermes skill that packages the Huasheng Shisan civil service exam method

What huasheng13-skill actually is, and who it is for

This repository is not a program. It is a Hermes skill, a bundle of Markdown files that an agent loads to answer questions about China's civil service exam. The README describes it as compiled from the publicly available teaching system of Huasheng Shisan, plus course materials and past exam papers. The intended user is someone preparing for the exam who already has a Hermes agent and wants it to respond in a specific pedagogical style: identify the question type, name the method, show the arithmetic, then flag the common mistake.

The scope is broad. The README lists six modules: data analysis, quantitative reasoning, verbal comprehension, judgment reasoning, general knowledge, and the essay paper. It claims coverage of more than twenty question types and over a hundred speed-calculation formulas. The repository layout backs this up with twenty files under references/ and five practice files under examples/. That is a lot of surface area for a skill whose entire interface is natural language. You do not invoke a command; you ask a question, and the README's table maps phrasings like "how do I do this data analysis question" to a single-question explanation mode, and "how do I speed up quantitative reasoning" to a module methodology mode.

The framing is exam-tactical rather than academic. The README's module table pairs each subject with a target time: 25 to 30 minutes for 20 data analysis questions, 35 minutes for 40 verbal questions, 8 to 10 minutes for 20 general knowledge questions. That time pressure is the actual subject of the skill. It is for someone who can already do the problems and wants to do them faster.

How the skill is structured: SKILL.md, references, and examples

The repository has four top-level entries: README.md, SKILL.md, references/, and examples/. SKILL.md is described in the README as the main skill file holding trigger conditions and workflow. That is the file an agent reads first to decide whether this skill applies to the current request. The README does not reproduce its contents, so the exact trigger syntax is something you have to open the file to see.

references/ holds twenty methodology files. They are split by module and then by sub-topic. For data analysis there are three: ziliao-fenxi.md for the full system, ziliao-susuan.md for speed-calculation methods, and ziliao-zonghe.md for comprehensive analysis questions. Quantitative reasoning gets two files, verbal comprehension three, judgment reasoning three, the essay paper three, and general knowledge two. Three more files sit outside the module split: strategy.md for answer order and time allocation, beikao-jihua.md for study planning, and cuoti-muban.md for an error-log template and review method. A final file, zhenti-shili.md, holds worked examples.

examples/ holds five practice files, one per module. This split matters. The references are the method, and the examples are the drills. An agent answering a question can pull the method from references and the practice pattern from examples, which is a cleaner separation than a single monolithic prompt would give. The trade-off is retrieval: with twenty-five files, the agent has to pick the right one, and the README does not document how that selection happens.

Installing and a first real use

The README does not give installation steps. It says the project is a Hermes skill and links to the Hermes agent project, but it never states a command to fetch or register it. There is no package name, no version, no port, no environment variable documented anywhere in the repository. If you want to use it, the only path the README describes is cloning it and letting your Hermes agent read SKILL.md.

A clone is the one command that is unambiguous, since the repository is a Git repository on the master branch:

git clone https://github.com/WangJunqing-coder/huasheng13-skill

After that, point your Hermes agent at the directory so it can read SKILL.md. The README does not document a configuration key for this, so treat the agent's own skill-loading instructions as the source, not this repository.

Once loaded, the README says there is nothing to memorize. You ask in plain Chinese. The first useful thing to try is a single question, because that exercises the most detailed output path. The README's own worked example is a definition-matching question about pre-contractual liability, where the skill answers C and explains that the deciding factor is whether the contract had already been formed. If you want to check the speed-calculation side instead, ask about data analysis timing, which the README maps to the module methodology mode. What you should see back is a structured answer with a question-type identification, a method, the calculation steps, and a listed common error, because the README's output specification requires all four.

The speed-calculation layer: ABRX, 截位直除, 415份数, 假设分配

The most concrete thing in the repository is the speed-calculation library, and it is the part worth judging the skill on. The README names four methods. ABRX is a four-element framework for data analysis: A is the base period, B is the current period, R is the growth rate, and X is the growth amount. It is a vocabulary for setting up a problem before any arithmetic happens.

The other three are arithmetic shortcuts. 截位直除 (truncated direct division) takes the first three digits of numerator and denominator and divides directly. 415份数 (the 4-1-5 fraction method) converts a growth rate into parts: a 25 percent increase means the original is 4 parts and the new value is 5 parts. 假设分配 (assumed allocation) assumes a base period and back-solves the current period for multi-step growth rates.

These are described in the README as general methods used across the civil service exam training industry rather than proprietary techniques, and the repository states that its worked examples use public papers or self-written data rather than paid course material. That is a deliberate boundary and it is the right one, but it also caps the value: the skill cannot reproduce the parts of a paid course that are actually paid for.

The README's own example of the 415 method is worth noting for how narrow it is. It works cleanly on rates that reduce to small fractions. A rate like 25 percent maps to 4 and 5. A rate like 13.7 percent does not map to anything tidy, and the README does not say what to do then. The method is presented without a fallback.

Where the skill is thin or the wrong tool

The largest limitation is that there is no executable component. Nothing here validates an answer, times you, or tracks progress. The README describes an error-log template in cuoti-muban.md, but it is a template, not a tracking system. If you want a tool that scores your practice sets, this is the wrong repository.

Second, the answer quality depends entirely on the model reading the files. The skill supplies method and structure, not ground truth. A wrong answer produced by the underlying agent will still come back in the correct five-part format, which makes it harder to catch, not easier. The README's requirement that every explanation include a common-mistake section does not protect against a mistake in the method itself.

Third, the README's own note says study plans must be personalized: the skill is instructed to ask about your level, available time, and target before giving a plan. That is a sensible instruction, but it means the planning mode is only as good as the answers you give, and the repository contains no assessment instrument to establish your level objectively.

Fourth, language. Everything is in Chinese, which is correct for the exam but means the skill is unusable for anyone who cannot read Chinese exam terminology. There is no English layer and the README does not suggest one is planned.

Finally, the README's closing section is motivational writing about perseverance. It is harmless, but it signals that the document is written for students rather than for engineers evaluating a dependency, and the engineering information an adopter needs, such as how skill selection works, is the part that is missing.

Alternatives: a generic agent with your own notes, or a dedicated question bank

There are two realistic alternatives, and they differ from this project in kind rather than degree.

The first is using a general-purpose agent with your own notes. You would keep your own Markdown files of formulas and write your own prompts. The difference is that huasheng13-skill has already done the decomposition: twenty reference files split by module and sub-topic, plus a fixed output format of question-type identification, method, calculation, common mistake, and extension. If you build your own, you will reproduce that structure badly before you reproduce it well. What you gain is control over exactly which methods you include and no dependence on someone else's file naming.

The second is a dedicated question-bank application, the kind sold by the major exam-prep publishers. Those give you scored practice, timed sets, and progress tracking, none of which exist here. The difference in approach is that a question bank is a measurement tool and this is an explanation tool. A question bank tells you that you got 14 of 20 right. This skill tells you why the other 6 were wrong and what method you should have used. They are complementary, and the README implicitly assumes you already have the question bank, since it recommends publisher question collections as study material alongside the skill.

Maintenance, licensing, and what to verify before depending on it

The repository is not archived, and the last push was on 2026-06-22. No releases are listed. With no releases and no changelog in the repository, there is no upgrade path to speak of: you take the master branch as it stands, and any future change arrives as a commit you have to notice yourself. For a Markdown knowledge bundle that is less alarming than it would be for a library, because there is no dependency graph to break. The cost of an update is re-reading the changed reference file.

On licensing, the README carries an MIT badge and a LICENSE file is referenced, and the closing line reads MIT License with the author's handle. MIT is permissive and would normally allow reuse and modification with attribution. That said, the content is described as compiled from third-party public sources: Huasheng Shisan's public teaching videos, Bilibili review sessions, Zhihu columns, and ProcessOn mind maps. The README asserts that the speed-calculation methods are industry-general and that no paid course material is reproduced. Whether the compilation as a whole is clean under MIT depends on facts the README does not establish, and that is a question for someone qualified to answer it, not something this article can settle. If you plan to redistribute the references, that is the point to check first.

One more thing to verify: the README links to a Bilibili space and a Hermes agent site, and the badges reference a government exam portal. None of those links are described as maintained by this repository, so do not treat them as part of the project.

Editorial conclusion

Adopt it if you already run a Hermes agent and want a structured Chinese-language study companion for the civil service exam, and you accept that it is a folder of Markdown rather than a program. Do not adopt it if you need an executable tool, an English-language resource, or anything with a documented installation path, because the README gives none. Before relying on it, open SKILL.md to confirm the trigger conditions match your agent's skill format, and read references/ziliao-susuan.md to check whether the speed-calculation methods are at the depth you need. The repository's last push was on 2026-06-22, and no releases are listed, so treat the content as a snapshot rather than something with an upgrade path.

Frequently asked questions

Official sources

1. Issues
2. README
3. WangJunqing-coder/huasheng13-skill on GitHub