write Skill from Waza for Claude: Edit Chinese and English Prose to Sound Human

write makes Claude edit Chinese or English prose so it sounds natural, by cutting AI patterns instead of swapping in fancier words.

Installs: 15.7K; on skills.sh
Stars: 7.5K; on GitHub
Files: 13; besides SKILL.md
Updated: Nov 2026; last repo commit

write is the /write skill in tw93's Waza collection. Claude edits your text in place, keeps your meaning and voice, and removes AI tells such as stiff phrasing, restated conclusions and em-dashes. It picks Chinese or English rules from the text you give it, not from the language of your command.

It has modes for release notes, public issue and PR replies, technical instructions, long drafts, before-and-after review, bilingual copy, product localization, document review, paragraph flow and tweets. Its core rule is that over-editing counts as failure, so a natural sentence stays as it is.

Before it hands text back, the skill runs a bundled punctuation checker with python3. Some modes also read git history or GitHub issues with the gh CLI. The skill tells Claude to start its first line with the 🥷 emoji.

Good for

Writers and bloggers who want AI-sounding drafts edited in place
Maintainers writing release notes, launch posts and public issue replies
Teams reviewing docs, install steps and error text for clarity
Anyone checking Chinese and English copy or localized product text for drift

Install write

Claude Code

Copy this prompt.

Install the write skill in Claude Code. Run this command:

/plugin marketplace add tw93/Waza
/plugin install waza@waza

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Pick one route. The plugin, the README skills CLI command and the ZIP each install all eight Waza skills (check, health, hunt, learn, read, think, ui, write). The write-only command installs just this one; its punctuation gate script sits inside the write folder. The README also describes optional rules and a statusline that this page does not cover.

From the project README. Source.

What's in the skill

When it loads

Rewrites and polishes Chinese or English prose, technical instructions, and product copy. Use when drafting, editing, localizing, or cutting AI tone. Not for code comments or commit messages.

Sections in SKILL.md

Outcome Contract
Durable Context Preflight
Core Stance
Pre-flight
Mode Picker
Document Review Mode
Paragraph Coherence Mode
Tweet / Social Post Mode
Hard Rules
Punctuation Gate
Gotchas
Output

Key facts

License: MIT
Author: tw93 (repo owner)
Repository: tw93/waza, folder skills/write
Installs on skills.sh: 15,656 as of 2026-11-02
Repo stars: 7,488 as of 2026-11-02 (whole repo)
Last repo commit: 2026-11-01
Files in the skill folder: 14 files including SKILL.md
skills.sh audits: Gen Agent Trust Hub: pass; Socket: pass; Snyk: pass

What it includes

Edit supplied text in place
Keeps meaning, claims and structure. Pasted text comes back without a change list, except the author's own finished draft.
Files · Default: line-level rewrite
Changelog and update-feed copy
Loads references/mode-release-notes.md, which reads the last release (with gh release view where available) and git log between the last release and the candidate.
Runs code · Release notes mode
Maintainer replies on issues and PRs
Loads references/mode-public-reply.md. It tells Claude to re-read the live issue or PR with gh before replying and again after posting.
Runs code · Public reply mode
Install steps, error text, tool descriptions
Loads references/write-technical.md for wording-only edits to instructions.
Technical writing mode
Drafts with sections, tables or images
Loads references/mode-long-form.md for structural work.
Long-form mode
Is the new version better?
Loads references/mode-revision-review.md to compare two drafts or another editor's polish.
Revision review mode
EN/CN pairs and multi-locale product copy
Loads write-zh-bilingual.md and write-product-localization.md to check drift between versions.
Bilingual and localization modes
PDFs and white papers
Checks privacy, tone consistency, bilingual accuracy, placeholder text and stale dated claims.
Document review mode
Flow checks and social posts
Coherence gives a numbered list of minimal fixes. The tweet mode applies announcement rules from references/write-zh-release-notes.md.
Paragraph coherence and tweet modes
Runs check-punctuation.sh before returning text
Runs bash check-punctuation.sh, which calls python3 check_punctuation.py with a language flag. The checker can print corrected text with --fix.
Runs code · Punctuation Gate

From the official docs

Permissions

Low risk

The skill runs a punctuation script and, in some modes, reads git history and GitHub issues. Keep approving commits, pushes and public replies yourself.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

It runs a bundled script before returning text
The Punctuation Gate runs check-punctuation.sh and python3 check_punctuation.py. With --fix it prints corrected text to stdout.
Some modes read git history and GitHub items
Release notes use git log and gh release view. Public replies use gh issue view or gh pr view. Approve any post or push yourself.
It can read memory summaries
The durable-context file says to read memory only when you name it or the project exposes a MEMORY.md, and that memory never authorizes writes or posts.

Technical details

Talks to	GitHub, through the gh CLI, in the release-notes and public-reply modes
Scopes	Text and files you point it at Git history and GitHub issues, PRs and releases, in some modes Memory summaries such as MEMORY.md, if you name them or the project exposes them

Questions

What does the write skill do?

It makes Claude edit prose so it sounds natural, in Chinese or English. It cuts AI patterns, keeps your meaning and voice, and avoids rewriting sentences that already read well.

How do I use it?

Install it with the install box above and type /write with your text or a file name. If you use the Waza plugin route in Claude Code, the namespaced command is /waza:write. In Codex, invoke the installed skill by name, as the Waza README says.

Which language does it use?

The language of the text being edited, not the language of your command. Chinese text loads the Chinese rules, bilingual work loads the bilingual rules, and everything else loads the English rules.

Does it need an API key?

The SKILL.md names no key. It needs python3 for the punctuation check, and gh for the modes that read GitHub issues, PRs and releases.

Is it safe?

It edits text you point at, and in repository work it can edit files and read git history and GitHub items. The skill says it grants no extra write or publishing authority, so keep approving commits, pushes and public replies yourself.

Why does it ban em-dashes?

The SKILL.md calls em-dashes and en-dashes the strongest AI-tone fingerprint, so it replaces every one in an edited draft with commas, periods, colons or parentheses.

Do I need the other Waza skills?

No. Waza also has think, ui, check, hunt, learn, read and health. The README lists them as separate skills, and the write SKILL.md does not require them. The plugin, ZIP and README skills CLI routes install all eight; the install box also has a write-only command.

Information

updated Nov 2, 2026

Repository: tw93/waza
Folder: skills/write

tw93/waza skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 9

write SKILL.md · github.com
Waza README · github.com
skills CLI commands · github.com
write SKILL.md · github.com
references/mode-public-reply.md · github.com
references/mode-release-notes.md · github.com
references/durable-context.md · github.com
scripts/check_punctuation.py · github.com
Waza README · github.com