Anti-AI Writing Kit

A portable writing kit for drafting, editing, rewriting, polishing, and reviewing prose without common AI-writing tells.

Click here for Simplified Chinese Version (简体中文版)

This is a kit, not a single prompt:

• SKILL.md contains the default operating instructions and writing rules.
• operations/kit-operations.md handles customization, reset, and rule maintenance.
• skill-customized.md, when created, is a complete standalone copy with the user's changes.

Test report

We tested Anti-AI Writing Kit, Humanizer, Stop Slop, and a no-skill baseline on English direct-writing tasks. Each condition used the same prompt set, model settings, and evaluation method. All generations used gpt-5.6-sol with medium reasoning.

Lower AI-smell rates are better.

Condition	Stage 1	Stage 2	Combined
Anti-AI Writing Kit	0.16%	0.38%	0.26%
Stop Slop	0.55%	0.43%	0.50%
Humanizer	0.89%	0.95%	0.92%
No skill	3.09%	2.75%	2.94%

The report tests whether our approach works and whether the skill loads completely. It does not establish a fixed ranking among peer skills. Detailed results and methodology are in the full report.

Scope and limits

This kit focuses on helping AI write and edit without common AI-writing patterns. It is not an AI detector, a personal-voice trainer, or a set of platform-specific templates.

The kit gives AI clear rules and checks, but no rule file can make every model follow every instruction perfectly.

Rule overview

The complete runtime rules, explanations, phrase lists, and examples are in SKILL.md:

1. Hard bans
2. Positive defaults
3. Word and phrase cleanup
4. Claims and evidence
5. Structure and formatting
6. Rhythm and repetition
7. Final-check preferences
8. Additional user preferences, used only in a customized file

Use with webpage AI

Download SKILL.md. You can edit the file before uploading it to your chat.

Use with agentic AI

Use the whole folder with an agentic AI that can load skills and run files.

Getting started

1. Customize it once. Type customize and the agent will create your local preference file.
2. Use it for drafting or editing. Send a topic, brief, draft, or finished piece.
3. Add personal rules when you find a new AI-writing habit. The agent will save each rule in skill-customized.md.

Install

Codex

Choose one location.

User-level install, available in every Codex project:

mkdir -p ~/.codex/skills
git clone https://github.com/evelynyaxueke/anti-ai-writing-kit.git ~/.codex/skills/anti-ai-writing-kit

Project-level install, available only inside one project:

mkdir -p .agents/skills
git clone https://github.com/evelynyaxueke/anti-ai-writing-kit.git .agents/skills/anti-ai-writing-kit

For a private repository, the machine must be signed in to an account with access.

After installing in the Codex app, open the command menu with Cmd+K or Cmd+Shift+P, then choose Force Reload Skills. If the skill does not appear, start a new Codex task or restart Codex. A turn that began before installation cannot see a newly installed skill.

Other agents

Use the agent's supported skill, plugin, or custom-instruction location and point it at the whole folder. SKILL.md alone is a manual fallback when relative scripts are unavailable.

Model choice

Use a current model with dependable instruction following for long or evidence-sensitive work. Smaller models can still benefit from the kit, but they are more likely to miss cross-sentence contrast, rhythm problems, evidence gaps, or the semantic final pass.

Customize

Type:

customize

The agent creates skill-customized.md only when requested, then walks through the numbered writing sections. With Node.js, scripts/print-customization.mjs extracts each category and subsection from one active rule source, so Sections 3.1, 3.2, and later subsections appear one at a time in full rather than being summarized or combined. The file contains the complete operating instructions and all eight writing sections, so it can be loaded as one standalone customized skill.

Older compact and legacy customized files remain supported. The agent edits an existing file in place and does not overwrite or silently convert it.

Rough replies are enough during customization. You can add a fragment, give an example, remove a rule, or say the section is fine.

To inspect the script-driven flow:

node scripts/print-customization.mjs --plan
node scripts/print-customization.mjs --step 3
node scripts/print-customization.mjs --step 3.1
node scripts/print-customization.mjs --next 3.1

Add a rule

Every rule added during normal use goes to skill-customized.md. If the file does not exist, the agent creates it first. SKILL.md stays unchanged.

In a new session, the safest instruction names both the skill and the action:

Use Anti-AI Writing Kit. Add this to my rules: never use "X" as a punchline.

If the skill is already active in the session, the shorter version is enough:

Add this to my rules: never use "X" as a punchline.

If you complain about an AI-writing habit while the skill is active, the agent should understand the context and ask whether you want to save it. A complaint is not saved without your confirmation. In a new session, name Anti-AI Writing Kit so the agent knows which rules you mean.

The agent first checks whether the rule already exists. New rules go in the smallest applicable section of the customized file.

Reset

Type:

reset

The agent deletes only skill-customized.md, then returns to the complete default SKILL.md.

File layout

anti-ai-writing-kit/
├── SKILL.md
├── README.md
├── reports/
├── scripts/
├── tests/
├── operations/
└── agents/

Generated local files:

• skill-customized.md, created only when a user requests personal customization
• .DS_Store, macOS folder metadata

Both are ignored by .gitignore.

License

MIT License. Copyright (c) 2026 Evelyn Ke.