huasheng13 skill: six exam modules, four shortcut methods, no licence file

Five Chinese phrases are the entire interface, and no command to remember

The skill describes itself as a systematised arsenal of solving methods, built on a teacher's publicly available teaching system, covering six modules, more than twenty question types and a claimed hundred-plus shortcut formulas.

The interface is a table of five phrases, each mapped to a mode. Asking how to solve this data analysis question selects single-question explanation. Asking how to speed up quantitative relations selects the module methodology mode. Asking for a set of past papers selects full practice. Asking how to prepare with three months left selects study planning, and asking how to allocate time in the objective section selects strategy advice. The instruction to the user is simply to send a question, ask for a method, or request a plan, with nothing to memorise.

That design has a consequence worth naming. If the routing is a table of five literal Chinese strings, then the table is illustrative rather than exhaustive, and the skill's own demonstrations use phrasings that appear nowhere in it. Scenario three opens with a question about guessing every answer in quantitative relations, which is not one of the five triggers. Whether a differently worded question still lands in the right mode is not something the page can tell you.

The badge row is equally loose. It links the national civil service examination authority's site, an agent directory, and a link to a video platform space with no account identifier in the URL, so that one cannot resolve to a channel. Neither the authority nor the directory appears in the attribution list, which names teaching videos, a question-and-answer site and a mind-mapping tool.

The module table sums to 135 objective questions and its time budget adds up

The quick reference table is the most useful page on the repository, because it is arithmetic.

Data analysis is 20 questions in a target 25 to 30 minutes. Quantitative relations is 15 questions with only 5 to 7 to be attempted. Verbal comprehension is 40 questions in 35 minutes. Judgement and reasoning is 40 questions in 35 minutes. General knowledge is 20 questions in 8 to 10 minutes. The essay section is 4 to 5 questions in 150 minutes. That is 135 objective questions in total, with shortcut methods marked as always provided for the first two modules and as absent for the other four.

The recommended order matches those targets, in one line:

常识（8min）→ 资料分析（28min）→ 判断推理（35min）→ 言语理解（35min）→ 数量关系（剩余时间选做）

Sum the stated midpoints and you get 107 of the 120 minutes an objective paper is assumed to have, leaving 13 minutes for 15 quantitative questions, of which the page tells you to spend 5 to 8 minutes picking the 5 to 7 you can do.

The four principles behind that ordering are stated in one sentence each: do the easy and the high-scoring first, since data analysis has the highest score density; do quantitative relations last because it is the worst value for time; do not agonise over general knowledge, at most 30 seconds a question; and mark the answer sheet twice, after data analysis and again after verbal and judgement reasoning, rather than at the end.

The table is internally consistent, which is worth saying plainly, because it is not the case everywhere in this README.

Guessing every C is scored at 3 points, and nothing says what the rest is worth

One of the four demonstrations answers a direct question: is guessing C on every quantitative relations question worth it? The reply works the expected value out as 15 questions multiplied by a probability of 0.2, multiplied by 1 point, giving 3 points, and concludes that it is a losing proposition.

Three guessing heuristics follow: exclude the extreme options, since the largest and smallest are usually not the answer; use the distribution of the options, on the assumption that each of the four letters appears three or four times; and when an option says approximately, pick the nearest integer.

The problem is that the trade cannot be evaluated with the numbers given. The arithmetic assumes every question is worth one point, and the page never states the point value of a question or the total available in the objective section. Without that, 3 points cannot be compared against the 5 to 7 questions the same answer tells you to solve, and the advice to walk away from a third of the paper rests on an assumed per-question value rather than a stated one.

The same answer also contradicts the module table, which recommends attempting 5 to 7 of those 15 questions, spending 5 to 8 minutes to do it. So the guidance is to invest eight minutes to rescue eight questions and to bank the rest, and the expected-value calculation exists to tell you the banking is rational. Neither half can be checked against the actual scoring, which is not on the page.

Four shortcut methods are listed under a claim of a hundred-plus formulas

The header line promises more than a hundred shortcut formulas. The method table on the same page lists four.

They are worth describing because they are the mechanism. Truncated division takes the first three digits of numerator and denominator and divides once. The 415 parts method turns a growth rate into a ratio of parts, so growth of 25 percent means the original was 4 parts and the current is 5, which is exact for that rate and 100 percent by the same arithmetic. Assumed allocation sets a base period and works backwards to the current one for multi-step growth. And the four-element system names base period, current period, growth rate and growth amount, which is the notation the rest of the data analysis material is written in.

What the table does not give is an error bound. Truncating to three digits throws away precision, and whether that matters depends entirely on the distance between the two answer options the calculation is meant to separate.

The mitigation is in the repository, one section away, in the speed diagnostic for the same module. It advises that if the options are far apart, choose directly, and only compute precisely when the gap is small, and that if column-setting is slow, truncate and keep the first three digits rather than calculating exactly. That is the correct companion rule to a lossy method, and the fact that it lives in a different section is the only thing wrong with it.

Six modules, five worked examples, and the one without an example

The examples directory holds five files, named in pinyin without a separator: one for judgement and reasoning, one for the essay section, one for quantitative relations, one for verbal comprehension and one for data analysis.

Five, against six modules. The missing one is general knowledge, and it is the module the quick reference table marks as having no shortcut methods, given 8 to 10 minutes, and solved by elimination, keyword spotting and accumulated knowledge of current affairs. It is also the module where a worked example would be hardest to write, since the questions are not template-shaped.

The naming convention differs between the two directories, which is worth knowing if you script anything against them. The examples use concatenated pinyin, while the references directory uses hyphenated pinyin with descriptive suffixes, such as a data analysis file, a data analysis shortcut file, a data analysis synthesis file, a quantitative relations file and a quantitative relations formula handbook, then a verbal comprehension file, a high-frequency idiom file, a verbal comprehension special file, a judgement file and a judgement rule handbook. Both are pinyin, and only one of them is hyphenated.

The project structure block is also cut off partway down that list, and it claims the references directory is a methodology library of twenty files, of which the visible names run to ten before the page ends.

Every answer has five parts, and the rulebook forbids three things

The output specification is the part of this skill that would survive a rewrite, because it constrains the model rather than describing it.

Every explanation of a question contains five elements. First, type identification: read the keywords and decide which question type it is. Second, the solution path: find the recognition marker, apply the formula or method, use the shortcut, and only then choose an answer. Third, the full calculation steps, shown rather than summarised. Fourth, the common mistakes for that question type. Fifth, related variants, so the same idea transfers to a neighbouring question.

The rules that follow are prohibitions. Shortcut calculations must show their working, not only the answer, and only in data analysis and quantitative relations. Method comes before calculation, so the recognition marker and the approach are explained before any arithmetic. Every question and every method must carry its common mistakes. Examples must use public past papers or data the skill writes itself, and must not copy questions from paid courses. For the essay section, the answer is a framework of stance, structure, sub-points and direction for material, not a finished model essay. And study advice has to be individual, which means asking the user about their level, their time and their goal before producing a plan.

That last rule is the one that makes this a study tool rather than a fortune teller, and the third is the one that keeps it clear of course material.

The entry file holds the triggers, and the structure list is cut in half

The project structure is small and legible. A `SKILL.md` at the root is described as the main skill file, holding the trigger conditions and the workflow, which means the routing table lives there rather than in the README. The README is the document you are reading. Then come the references directory and the examples directory.

The structure block is drawn as an ASCII tree, and it ends mid-line, partway through the tenth filename in the references listing. The comment above the directory says it is a methodology library of twenty files, so the visible portion accounts for half of what is claimed. Nothing on the page says what the other ten cover, or which module each belongs to.

Two claims in the surrounding text are also unmeasurable. The header promises more than a hundred shortcut formulas where the table shows four, and the training plan in the data analysis demonstration moves from two practice passages a day to four to a full set, with visible results in two weeks. Neither figure comes with a baseline, a metric, or a definition of what visible results means, and a two-week claim is the kind of thing that ends up quoted by people who did not test it.

For what it is worth, the skill ends on encouragement rather than a claim, and the closing line is that technique can be learned, method can be practised, and persistence is up to you.

MIT in the footer, a licence badge, and no licence file in the repository

The licensing needs a reader with a file browser.

The README's footer states an MIT licence with a copyright line naming the repository owner, and there is a licence badge at the top of the page pointing at a LICENSE file. The repository's own licence field is empty, and the top-level listing holds four entries: the README, SKILL.md, the examples directory and the references directory. There is no LICENSE file among them.

So the badge links to something the repository does not contain, the footer states terms that are not in a file, and the machine-readable field says nothing. For personal use that does not matter. For anyone who wants to reuse a skill file, the MIT line in the footer is the only statement of terms available, and it is a statement in a document rather than a grant in a file.

The rest of the state is equally short. The default branch is master rather than main. There are no GitHub releases, and the last push to the default branch was on 2026-06-22.

There is one question about the material itself that this repository takes seriously and that a reader should take seriously too. The attribution section says the content is compiled from the teacher's publicly available teaching videos, a video platform review course, a question-and-answer column and mind maps, and that all the shortcut methods are common in the examination training industry rather than exclusively copyrighted. The module ordering, the recommended answer sequence and the named system are presented as that teacher's method, so the boundary between a common technique and a particular course's expression of it is the thing to reason about before redistributing anything derived from these files.

Editorial conclusion

Take this skill if you are preparing for 行测 or 申论, you read Chinese, and you want a model to explain its reasoning rather than hand you a letter. It is unusually honest about its own boundaries, which is why it refuses to reproduce course questions and gives an essay framework instead of a finished essay, and why it asks about your level before planning. Leave it if you need a validated scoring strategy, since the module table never states what a question is worth, which makes the advice to abandon 数量关系 impossible to check. Three things to check first: whether the two-week training claim matches your baseline, whether the shortcut methods are accurate enough for the option gaps in your paper, and what licence actually governs reuse, since the footer says MIT, the badge points at a file the repository does not contain, and the license field is empty.

Frequently asked questions

Official sources

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