codex-academic-writing-skills

Codex Academic Writing Skills

Codex skills for Chinese biomedical researchers: academic translation, scientific writing refinement, reviewer responses, manuscript logic auditing, LaTeX–Word format cleanup, journal cover letters, and reference integrity checks.

This is a small, practical skill pack for authors who already have real research content and want repeatable local workflows inside Codex. It is intentionally not a paper-writing bot, a publication guarantee, or a substitute for authors, clinical experts, statisticians, or journal editors.

中文说明 · Quick demos · WeChat launch article · Personal site

Why This Exists

Many academic AI tools stop at a generic prompt: "polish this", "reply to reviewers", or "check references". In real biomedical writing, the hard part is not only fluency. The hard part is keeping LaTeX intact, preserving P values and gene names, avoiding causal overclaims, not inventing references, and making revision logic auditable.

This repository turns seven high-frequency academic writing tasks into reusable Codex skills with explicit boundaries and validation scripts.

What Makes It Different

Compared with	Difference
Generic chat prompts	Local, reusable Codex skills with clear triggers, constraints, and repeatable behavior.
General writing polishers	Designed to preserve scientific evidence strength, statistics, citations, LaTeX, gene names, drug names, and figure/table references.
Large prompt collections	Small and focused: seven skills for common submission and revision tasks.
"AI writes the paper" workflows	Supports responsible author-controlled writing, audit, and cleanup; it does not fabricate content or bypass peer review.
English-only academic tools	Includes workflows that are especially useful for Chinese biomedical, clinical, pharmacy, bioinformatics, and AI-in-medicine researchers.

Quick Start

git clone https://github.com/zl9099/codex-academic-writing-skills.git
cd codex-academic-writing-skills
./install.sh

Then ask Codex in natural language, for example:

Use reference-integrity-checker to audit whether these citations really support the claims.

See Quick demos for short before/after examples.

Who This Is For

• Researchers writing manuscripts in English, Chinese, LaTeX, Word, or Markdown.
• Biomedical, clinical, pharmacy, bioinformatics, and AI-in-medicine teams.
• Authors preparing revised manuscripts and reviewer responses.
• Codex users who want reusable local skills instead of repeatedly pasting long prompts.

Typical Use Cases

• Translate a Chinese manuscript paragraph into English while preserving LaTeX, citations, abbreviations, P values, and gene names.
• Polish an abstract without changing association into causation.
• Draft a point-by-point response to reviewers while separating accepted revisions from polite rebuttals.
• Audit a manuscript for overclaims, missing controls, weak figure logic, and reviewer-risk points.
• Clean text damaged by moving between LaTeX, Word, Markdown, PDF, and submission portals.
• Draft a journal cover letter without inventing editor names, impact factors, journal policies, or acceptance claims.
• Check whether references and citation claims need manual verification before submission.

Skills Overview

Skill	Purpose
academic-translation-latex	Accurate academic Chinese-English translation while preserving LaTeX, citations, equations, gene names, drug names, units, P values, and figure/table references.
scientific-style-polisher	Scientific prose polishing without changing evidence strength or inventing results.
reviewer-response-builder	Point-by-point reviewer and editor response drafting with polite rebuttal and revision summaries.
manuscript-logic-auditor	Logic, claim, controls, novelty, figure-story, and reviewer-risk audit before submission.
latex-word-roundtrip-cleaner	Cleanup for text damaged by moving between LaTeX, Word, Markdown, PDF, and submission portals.
journal-cover-letter-builder	Journal cover letters, presubmission inquiries, editor-facing submission notes, and novelty statements.
reference-integrity-checker	Reference metadata, DOI/PMID consistency, citation-to-claim alignment, duplicate references, and bibliography risks.

Repository Structure

codex-academic-writing-skills/
├── skills/
│   ├── academic-translation-latex/
│   ├── scientific-style-polisher/
│   ├── reviewer-response-builder/
│   ├── manuscript-logic-auditor/
│   ├── latex-word-roundtrip-cleaner/
│   ├── journal-cover-letter-builder/
│   └── reference-integrity-checker/
├── examples/
├── scripts/
│   └── validate_skills.py
├── install.sh
├── uninstall.sh
├── README.md
├── README.zh-CN.md
├── LICENSE
├── CHANGELOG.md
├── CONTRIBUTING.md
└── SECURITY.md

Installation

One-Command Installation

From the repository root:

./install.sh

By default, skills are installed into:

~/.codex/skills/

The installer only installs the seven whitelisted skills in this repository. It does not copy or modify other local skills.

Manual Installation

Copy the seven directories under skills/ into your Codex skills directory:

mkdir -p ~/.codex/skills/
cp -R skills/academic-translation-latex ~/.codex/skills/
cp -R skills/scientific-style-polisher ~/.codex/skills/
cp -R skills/reviewer-response-builder ~/.codex/skills/
cp -R skills/manuscript-logic-auditor ~/.codex/skills/
cp -R skills/latex-word-roundtrip-cleaner ~/.codex/skills/
cp -R skills/journal-cover-letter-builder ~/.codex/skills/
cp -R skills/reference-integrity-checker ~/.codex/skills/

Installer Options

./install.sh --dry-run
./install.sh --force
./install.sh --target ~/.codex/skills/

• --dry-run prints intended actions without copying files.
• --force backs up existing target skill directories before replacing them.
• --target installs into a custom Codex skills directory.

Uninstall

./uninstall.sh

To skip confirmation:

./uninstall.sh --yes

The uninstaller removes only the seven whitelisted skills and never removes the entire ~/.codex/skills/ directory.

Natural-Language Trigger Examples

• "Use academic-translation-latex to translate this LaTeX abstract into journal-ready English."
• "Use scientific-style-polisher to reduce AI-like phrasing without changing the science."
• "Use reviewer-response-builder to draft a response to Reviewer 2."
• "Use manuscript-logic-auditor to identify overclaims and missing controls."
• "Use latex-word-roundtrip-cleaner to clean this Word-pasted LaTeX text."
• "Use journal-cover-letter-builder to draft a cautious cover letter for this manuscript."
• "Use reference-integrity-checker to audit whether these citations support the claims."

Typical Input and Output Examples

Detailed examples are available in:

• academic-translation-latex
• scientific-style-polisher
• reviewer-response-builder
• manuscript-logic-auditor
• latex-word-roundtrip-cleaner
• journal-cover-letter-builder
• reference-integrity-checker

Example input:

Please polish this sentence without changing the evidence strength:
"Our results prove that pathway X prevents disease Y."

Expected output principle:

Replace unsupported causal wording with cautious, evidence-aligned language.

Example output snippet:

Our results suggest that pathway X is associated with disease Y and may represent a hypothesis for further validation.

More examples:

• Quick demos
• academic-translation-latex
• reviewer-response-builder
• reference-integrity-checker

How These Differ From General Chat Prompts

General prompts are often single-use and easy to drift from. Codex skills are reusable local instructions with explicit trigger scenarios, boundaries, and quality checks. These skills are designed to preserve technical content, reduce accidental overclaiming, and make common academic writing tasks more consistent.

Academic Integrity and Use Boundaries

These skills are designed to support responsible writing, not to bypass authorship or peer review.

They do not:

• generate fictional references
• fabricate data or research results
• guarantee acceptance or publication
• replace authors, clinical experts, statisticians, methodologists, or journal editors
• remove the need for final human review

For confidential manuscripts, patient-related text, controlled-access data, or unpublished results, use a local environment that complies with institutional and journal policies.

Privacy and Data Security

Do not commit or share private manuscripts, patient data, unpublished findings, API keys, tokens, cookies, passwords, or private repository URLs.

This repository contains only generic skill instructions and safe examples. The validation script checks for common absolute paths and suspicious secrets, but human review is still required before public release.

Known Limitations

• These skills do not verify the truth of references unless paired with a literature-search workflow.
• They cannot judge all field-specific norms without expert review.
• They do not replace statistical review or clinical review.
• They may preserve ambiguous terminology if the source text is unclear.
• They are not intended to generate full manuscripts from scratch.

Validation

Run:

python3 scripts/validate_skills.py
bash -n install.sh
bash -n uninstall.sh

The validator checks skill structure, frontmatter, duplicate names, suspicious secrets, local absolute paths, placeholder text, Markdown links, file size, and install/uninstall safety boundaries.

Contributing

See CONTRIBUTING.md.

Version

Current public release: v0.2.0.

See CHANGELOG.md.

Acknowledgements

This project was inspired in part by ideas and workflow patterns from awesome-ai-research-writing and AI-Research-SKILLs. The skills in this repository were independently adapted and rewritten for Codex-based academic writing workflows.

Reference projects:

• awesome-ai-research-writing: https://github.com/Leey21/awesome-ai-research-writing
• AI-Research-SKILLs: https://github.com/Orchestra-Research/AI-Research-SKILLs

License

MIT License. See LICENSE.