Guide · Agent skill
How to use the Write: Cut the AI Taste agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
Write: Cut the AI Taste is an agent skill from the tw93/Waza project (plugins/waza/skills/write, MIT licence). It tells an AI agent to edit text you already have so it sounds less like a chatbot, while keeping your meaning, facts and voice. This guide explains what agent skills are, how to install this one, how to trigger it, and what it needs and touches on your machine.
Write: Cut the AI Tastetw93/Waza · MIT Download the skill Quick start
- 1Download the skill. Download the ZIP for the Write: Cut the AI Taste skill from this page.
- 2Unzip it into a skills folder. Place the skill folder in ~/.claude/skills/ for personal use, or in a project's .claude/skills/ folder for one project.
- 3Or upload it in the Claude apps. In the Claude apps, open Settings, then Capabilities, then Skills, and upload the skill.
- 4Check the requirements. Make sure Python 3 is installed for the punctuation checker. Bash is only needed for the optional shell wrapper.
- 5Ask for an edit in plain language. Paste your text and say what you want, for example polish a release note or remove AI tone from a draft, and name the audience.
- 6Review the output. Check facts, numbers and technical terms against your original before you publish.
What is an agent skill and what does this one do?
An agent skill is a folder of instructions and reference files that an AI agent loads when a task matches the skill's description. This skill is defined by a SKILL.md file plus reference files for different kinds of writing.
When installed, the agent edits your existing text to remove stock AI phrasing and over-polished tone. It is designed to keep the author's meaning, factual claims and structure intact, and it avoids changing sentences that already read naturally. It handles Chinese and English prose, technical instructions, release notes, social posts and product copy, and loads a matching style reference for each.
- Removes AI-style wording and performative polish rather than swapping in fancier vocabulary
- Works on Chinese and English text, including bilingual Chinese/English copy
- Runs a punctuation checker on the output for Chinese/English mixing and punctuation errors
- Does not write code comments or commit messages; the skill excludes them
How do you install the Write skill?
Download the ZIP for this skill from this page and unzip it. You have two ways to make it available to your agent.
For Claude Code, unzip the skill folder into your user skills directory ~/.claude/skills/, or into a project's .claude/skills/ folder if you want it only for that project.
In the Claude apps, you can upload the skill instead. Go to Settings, then Capabilities, then Skills, and upload the skill there.
- Personal use across projects: ~/.claude/skills/
- One project only: the project's .claude/skills/ folder
- Claude apps: Settings → Capabilities → Skills
How do you trigger the skill with natural requests?
You do not need a special command. The skill's description covers drafting, editing, localizing and cutting AI tone, so ask for those things in plain language. The skill also lists trigger phrases such as 改稿, 润色, 去AI味, proofread, polish, rewrite, tweet and release notes.
Give the agent the text itself or name the file it lives in. Say who the audience is when it is unclear. The skill asks for the audience before editing if it cannot infer it from the text, because a blog post, an RFC and an email need different edits.
- Polish this release note for users, not developers: [paste text]
- Remove the AI tone from this blog draft and keep my examples as they are
- Proofread this Chinese and English product copy and check the punctuation
- Rewrite this installation section so each step is clear
- Edit this GitHub reply to the reporter; keep it short
What happens during an edit?
The skill treats its catalog of AI patterns as a set of things to recognize, not a checklist to run in order. It prefers a few meaningful edits over many mechanical swaps. Over-editing counts as a failure, so a sentence that is already clear is left alone.
When you give it your own finished draft, the default scope narrows to typos, broken sentences and clear AI tells. In that case the skill is meant to return a per-sentence list of every deleted or rewritten sentence so you can veto each change. For pasted text, the output is the edited prose.
The skill also has specialized modes for long-form articles, public GitHub replies, release notes, before-and-after comparisons and product localization. It loads a mode file only when your request matches it.
- Long article with several sections: structural review first
- GitHub issue or PR reply: short reply with the factual status and next step
- Before/after versions: compare drafts and judge the newer one
- Product copy across locales: localization review
What does the skill need, and what does it run?
The skill needs Python 3 to run its punctuation checker script, scripts/check_punctuation.py. Bash is needed only for the optional shell wrapper, scripts/check-punctuation.sh. The source does not give exact commands for running these, so check the files in the skill folder before running them.
The skill runs scripts, so review them before you use it on sensitive material. It does not use the network. Its document review mode asks the agent to flag sensitive information that should not be disclosed to the intended audience.
- Requires: Python 3
- Optional: Bash for the shell wrapper
- Runs scripts: yes (punctuation checker)
- Uses network: no
- Licence: MIT
What are the limits and tips for good results?
The skill preserves factual claims but does not verify them. Check facts yourself before publishing. It also keeps your colloquial words, judgments and emotional stance, so it is not a heavy style rewriter. If you want a bold restyle, the skill will not deliver it by design.
For best results, give the full text with context, state the audience and the language, and ask for the specific change you want. Review the output against your original, especially on technical terms and numbers.
- Give the audience and surface (blog, release note, tweet, product page)
- Ask for light editing if you want your voice kept
- Verify facts and figures after the edit
- Use the Chinese/English punctuation check for bilingual copy
What people use it for
- Editing a blog draft so it reads in the author's own voice
- Writing or polishing release notes and changelog entries
- Drafting a short maintainer reply on a GitHub issue or PR
- Checking Chinese and English product copy for drift and punctuation errors
- Clarifying installation steps or error text in technical documentation
Frequently asked questions
Is the Write: Cut the AI Taste skill free?
The skill is published under the MIT licence, as stated on this page. The source does not mention prices, so check the project's repository for current terms before using it.
What do I need to run the Write skill?
You need Python 3 for the punctuation checker script. Bash is optional and only needed for the shell wrapper. The skill does not use the network.
How do I install the Write skill in Claude?
Unzip the downloaded skill folder into ~/.claude/skills/ or a project's .claude/skills/ folder. In the Claude apps, you can instead upload it under Settings, then Capabilities, then Skills.
Does the skill work on Chinese text?
Yes. It handles Chinese and English prose, and it loads Chinese style references and a punctuation checker for Chinese/English mixing and punctuation errors.
Can it write code comments or commit messages?
No. The skill explicitly excludes code comments and commit messages.
Does this skill check whether my facts are correct?
No. It preserves the factual claims in your text rather than verifying them, so you need to check facts yourself.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
Use this skill if you want a draft to sound like you rather than a chatbot, if you write release notes or public GitHub replies, or if you localize and check bilingual Chinese and English copy. Pick something else if you want heavy stylistic rewriting, fact-checking, or help with code comments or commit messages.
Download the skill