How to Install and Use Academic DeAI for Chinese and English Academic Editing

What Academic DeAI Does

Academic DeAI is a skill for editing Chinese and English academic prose. It removes empty framing, repetitive structures, and template-like language while preserving the manuscript’s facts, citations, terminology, negation, evidence strength, causal qualifications, and authorial voice.

The project supports proofreading, language polishing, academic translation, and user-authorized structural revision. Its primary scope does not include ordinary email, blog, or marketing copy.

Academic DeAI does not choose an AI model, call a model API, or automatically chain other editors. It is an editing skill that runs within a compatible host. Version 3.0.0 follows current GPT-6 skill guidance and adapts contextual editing ideas from Humanizer 3.1.0, but it does not claim identical performance across every GPT-6 checkpoint.

Why Contextual Academic Editing Matters

Academic prose often contains phrases that add emphasis without adding evidence. Academic DeAI aims to remove such framing only when the surrounding context supports the change. It evaluates structure and meaning before treating individual words as unwanted patterns.

For example, consider this fictional Chinese passage:

值得注意的是，本研究对126名患者进行了横断面分析，结果表明睡眠质量评分与瘙痒严重程度评分相关（r=0.32，P=0.01）。这一发现不仅揭示了二者之间的复杂联系，也为今后的研究提供了重要启示。

A possible revision is:

本研究对126名患者进行横断面分析，发现睡眠质量评分与瘙痒严重程度评分相关（r=0.32，P=0.01）。

The revision removes generic framing but retains the study design, sample size, variables, correlation coefficient, and P value. This brief example illustrates the editing approach; it is not a benchmark comparison.

Key Features

• Chinese and English editing: Revise academic writing in either language at the depth requested by the user.
• Academic translation: Translate manuscripts while following supplied terminology, writing samples, structural requirements, and length constraints.
• Claim preservation: Review actors, objects, direction, time, conditions, and citation attachment at the source-claim level.
• Evidence-aware revision: Preserve uncertainty, negation, causal qualifications, ranges, negative results, and necessary comparisons.
• Structure-aware editing: Consider the manuscript’s organization before cleaning up local wording.
• Style guidance: Use author samples to guide expression without importing new facts from those samples.
• Flexible execution: Avoid mandatory rewrite counts or compulsory reports when the request is already sufficiently specific.
• Selective resource loading: Load references and tools only when they are relevant to the task.

Prerequisites

You need a host that supports skills and a valid skill location configured for that host. Current Codex documentation lists ~/.agents/skills for user-level skills and .agents/skills for repository-level skills. Existing environments may use another configured location, so follow your host’s documentation.

Python 3.10 or newer is required only for the included text utilities and validation commands. These utilities inspect content and package behavior; they do not generate or rewrite prose.

Install Academic DeAI

Fresh user installation

From the directory where you want to download the repository, run:

git clone https://github.com/heise3/academic-deai.git
mkdir -p ~/.agents/skills
cp -R academic-deai ~/.agents/skills/academic-deai

Copy the entire repository folder. Do not copy only SKILL.md, because the package uses relative references and includes optional scripts.

Use the copy command only when the destination does not already exist. After installation, reopen the host’s skill list. If Academic DeAI does not appear, restart Codex or the relevant host.

Repository-level installation

If your host supports repository skills, place the complete package under the repository’s configured skill directory, commonly .agents/skills. Confirm the exact supported path before copying files.

Upgrade an existing installation

Back up the installed Academic DeAI directory before upgrading. Replace the existing folder deliberately rather than merging unknown files from an older release. This helps prevent stale references, scripts, or configuration files from remaining in the package.

Consult the official skill documentation for current host-specific installation guidance.

Basic Usage

Revise a Chinese academic section

Identify the skill, specify the section and language, list the content that must remain unchanged, and state the desired output format:

Use $academic-deai to revise this Chinese discussion section.
Keep every substantive claim, citation, number, and causal qualification.
Return only the edited text.

Paste the discussion section after the instruction. The final sentence prevents extra commentary when you need text that can be inserted directly into a manuscript.

Translate a manuscript into English

For translation, provide terminology and style references together with explicit structural constraints:

Use $academic-deai to translate this manuscript into English.
Use the terminology list and my writing sample below.
Preserve the requested section structure and minimum length.

The terminology list can establish consistent translations for domain-specific terms. A writing sample can guide style, but it should not be treated as a source of new facts.

Request a controlled structural revision

Academic DeAI can perform structural revision when you authorize it. Define what may change and what must remain fixed. For example:

Use $academic-deai to revise this English results section.
Improve paragraph order and transitions, but preserve all headings,
statistics, comparisons, negative results, citations, and qualifications.
Return only the revised section.

This instruction distinguishes permitted organizational changes from protected substantive content. Required headings and meaningful methodological constructions, including necessary passive methods, should remain when requested.

Validate the Package

Run the package validator from the repository root:

PYTHONDONTWRITEBYTECODE=1 python3 scripts/validate_package.py .

The validator checks package structure, references, version information, and portable files.

Run the unit tests with:

PYTHONDONTWRITEBYTECODE=1 python3 -m unittest discover -s tests -v

The tests cover observable behavior of the included tools. Passing validation and tests does not prove editing quality, semantic equivalence, authorship, or detector performance.

Check two text files with content lock

The content-lock utility can compare an original document with a revision:

python3 scripts/content_lock.py original.md revised.md --strict

This utility is part of the validation workflow, not a text generator. Treat its output as an aid rather than proof that every meaning and claim has been preserved.

Working with PDF and DOCX Sources

PDF text extraction additionally requires Poppler’s pdftotext. Make sure that command is installed and available in your environment before relying on PDF extraction.

DOCX extraction reads body XML, but it does not preserve every style, comment, tracked change, footnote, or textbox. If those document features matter, use document-capable tooling and verify the editable Word file directly.

Advanced Tips for Reliable Editing

Define protected content explicitly

Do not rely on a general request such as make this sound better. Name the elements that cannot change, including claims, numbers, citations, terminology, negation, uncertainty, time frames, conditions, and causal strength.

State the permitted editing depth

Distinguish proofreading from deeper revision. For proofreading, limit changes to grammar and wording. For structural editing, explicitly authorize paragraph reordering, transition changes, or removal of redundant framing.

Protect evidence strength

Ask the skill to retain distinctions such as association versus causation, possibility versus certainty, and absence of evidence versus evidence of absence. Negative findings and qualified conclusions may look stylistically weak but can be scientifically essential.

Attach citations to the correct claims

When revising sentence boundaries or paragraph structure, ensure that each citation remains attached to the proposition it supports. This is especially important when one long sentence is divided into several shorter sentences.

Provide a terminology list

For multilingual or specialized manuscripts, supply preferred translations, abbreviations, capitalization, and terms that must remain unchanged. This reduces inconsistency without asking the editor to infer domain conventions.

Use author samples carefully

A representative writing sample can guide sentence length, level of formality, and preferred phrasing. State that it is a style reference only and must not contribute facts, citations, or arguments to the edited manuscript.

Preserve required form

If a journal, institution, or supervisor requires headings, minimum length, section order, or reporting language, include those constraints in the prompt. Academic DeAI is designed to retain required headings and structures when instructed.

Review the result manually

After editing, compare the revision against the source at the claim level. Check who performed each action, what was measured, the direction of each result, applicable time periods and conditions, causal wording, numerical values, and citation placement.

Important Limitations

• Academic DeAI does not select a model or make model API calls.
• The included tools do not write or rewrite text.
• Validation does not establish semantic equivalence or editing quality.
• DOCX extraction cannot preserve every document feature.
• The project’s design guidance is not a benchmark guarantee for every GPT-6 checkpoint.
• Installing the skill does not create or modify a remote Git repository.

Conclusion

Academic DeAI provides a structured way to polish and translate Chinese and English academic writing without intentionally weakening evidence or changing substantive claims. Install the complete package, give precise editing boundaries, preserve discipline-specific terminology, and manually compare each revision with its source. The project is available under the MIT license.