zLanqing/codex-claude-academic-skills: Three Academic Skills for Claude Code and Codex

What the three skills actually cover

The repository ships three directories at the top level: research-writing-skill, office-academic-skill and scientific-toolkit-skill. Each is a folder of instructions and reference material, not a program you run. The README describes them as complementary and gives three chained scenarios: data analysis and figures flow into paper text, which flows into a defense deck; a PDF flows into a Word reading report, which flows into a group-meeting deck; a simulation flows into methods and experiment sections.

The intended user is a Chinese-speaking researcher who already works inside an AI coding agent. The README is explicit that explanations, body text and slides default to Chinese, while paper titles, formulas, variable names, model names, software commands and bibliography entries stay in English. That split is the most concrete design decision in the repository, and it is the reason the pack is not simply a translation of an English skill set.

The three skills are unevenly specified. research-writing-skill lists rhetorical-move guides, checklists, figure templates and a brainstorming guide. office-academic-skill documents OOXML-level inspection and editing for DOCX and PPTX plus a scripts directory for thesis-defense decks. scientific-toolkit-skill is the broadest, listing roughly twenty Python libraries alongside MATLAB and Octave work.

The evidence rules are the real product

Strip away the topic labels and the shared asset is a set of constraints on what the model may assert. The README states that the skills do not fabricate data, DOIs, journal information or experimental results, and that they separate four categories of information: text or data already present, content confirmed by the user, inference, and suggested extension. It also says vague qualifiers such as "significant" and "advanced" should be replaced with measurable conditions and comparison baselines.

That is a stronger claim than most prompt collections make, and it is also the part a reader cannot verify from the README alone. Whether the four-way separation survives a long editing session depends on how the SKILL.md files phrase the instruction and how the agent weighs it against the user's own requests. A user who asks for a plausible-sounding number will often get one from any model; the skill can only make that outcome less likely, not impossible.

The same applies to source tagging. The README says claims, parameters, quantitative results, datasets and figures should carry source labels. In practice the label is only as good as the material the user supplies. If you paste an abstract without a DOI, the correct behaviour is to leave the DOI out, and that is a behaviour you have to observe rather than assume.

Installing into Claude Code or Codex

All three skills share one layout: a SKILL.md at the root, an agents/ directory and a references/ directory. The README gives copy commands for both platforms. The clone URL in the README uses a placeholder username, so substitute the real repository path before running it.

git clone https://github.com/你的用户名/claude-academic-skills.git
cp -r claude-academic-skills/research-writing-skill ~/.claude/skills/
cp -r claude-academic-skills/office-academic-skill ~/.claude/skills/
cp -r claude-academic-skills/scientific-toolkit-skill ~/.claude/skills/

After the copy, the three directories should appear under ~/.claude/skills/. The README also documents a plugin route, `/plugin install 你的用户名/claude-academic-skills`, again with the placeholder username.

For Codex the destination changes to ~/.codex/skills/, and the README additionally documents loading from a URL for the current session.

cp -r claude-academic-skills/research-writing-skill ~/.codex/skills/
codex --plugin-url https://github.com/你的用户名/claude-academic-skills

If you want the skills scoped to one project instead of your whole machine, the README says to place the directories under .claude/skills/ or .codex/skills/ in the project root. A first real use would be pointing research-writing-skill at a draft and asking for a methods section, then checking whether the output distinguishes your confirmed numbers from its own suggestions.

Where the coverage is thin or the tool is wrong

The README does not document a rollback path for edits made to an existing Word document. Because office-academic-skill advertises versioned editing of DOCX files, a user working on a manuscript should keep their own copy before letting the skill modify it. Nothing in the repository suggests an undo mechanism, and version control on binary Office files is awkward at best.

There are no releases. The last push to the default branch was on 2026-05-14, which is more than six months before today, so this is not a project to describe as actively developed. The README also does not state which Python or MATLAB versions the scientific-toolkit-skill examples target, and it does not say whether the MATLAB code paths have been exercised under Octave for the functions it lists.

The pack is also the wrong tool outside its language assumption. A researcher writing in English gets Chinese-default prose unless they override it every session, and the rhetorical-move guides are organized around Chinese academic conventions. Anyone who wants a reproducible computational pipeline rather than an agent that writes code on request should look elsewhere; the scientific toolkit generates scripts, it does not ship a tested library.

How it differs from a general scientific agent skill collection

The README credits K-Dense-AI/scientific-agent-skills, a collection of more than 134 scientific agent skills spanning MATLAB and Octave, Matplotlib, signal processing and scientific visualization. The difference is scope and language. That collection is broad and English-oriented, and a user picks individual skills for individual tasks. This repository is three skills with a declared Chinese-first convention and an explicit evidence policy shared across all three.

It also credits zouchenzhen/thesis-defense-pptx-skill, which the README describes as generating editable defense PPTX from a paper PDF or LaTeX source, preserving template style, exporting per-page PNG and checking for text overflow, and as optimized for Windows with PowerPoint. The overlap with office-academic-skill's thesis-defense-pptx scripts directory is visible in the repository layout. The practical distinction is that the defense workflow here is one part of a three-skill set rather than a standalone tool, so choosing between them comes down to whether you want the writing and computing skills alongside it.

A third credit, Gabberflast/academic-pptx-skill, is described as focusing on academic slide conventions: conclusion-style titles, one core point per slide, figure citation standards and a references page. office-academic-skill states the same rules, which suggests the conventions were adopted rather than invented here.

Maintenance, licensing and what you are actually adopting

The repository is MIT licensed, and the README restates the usual terms: free use, copying, modification, merging, publication, distribution, sublicensing and sale, provided the copyright notice and permission notice are retained, with the software provided as is and without warranty. That is the standard MIT grant, and it is permissive enough for academic and commercial use alike.

One clause matters more than the rest. The README notes that some referenced external resources, specifically scripts under scientific-toolkit-skill/references/scientific-skills/, follow their original licenses. Those licenses are not enumerated in the README, so a user who redistributes the repository as a whole is carrying files whose terms they have not read. Checking those files before redistribution is the concrete step, not a general caution.

The upgrade cost is low in the sense that there is nothing to compile and no dependency lockfile to maintain. It is higher in the sense that the skills are instructions, and instructions interact with model behaviour that changes underneath them. When the underlying agent changes how it weighs system-level skill text, the same SKILL.md can produce different output. There are no releases and no changelog in the repository to tell you when that has happened.

Editorial conclusion

Adopt it if you write papers or reports in Chinese and already run Claude Code or Codex, because the three skills share one evidence convention and can be chained in a single session. Do not adopt it if you need English-first output, a maintained package with releases, or a workflow that runs without a coding agent. Before trusting it, open research-writing-skill/SKILL.md and check whether the four information categories it claims to distinguish are actually enforced in the instructions, and verify the license of anything under scientific-toolkit-skill/references/scientific-skills/ before you redistribute it.

Frequently asked questions

Official sources

1. Issues
2. License: MIT
3. README
4. zLanqing/codex-claude-academic-skills on GitHub