Skip to content

reference.md

Latest commit

 

History

History
347 lines (258 loc) · 31.8 KB

reference.md

File metadata and controls

347 lines (258 loc) · 31.8 KB
Raw

Reference

简体中文

Draft format

---
template: sheet        # sheet = grid of panels (default), doc = one column with a table of contents
theme: auto            # auto (default) picks paper for long text, blueprint for diagrams; or blueprint, shadcn, paper, your own theme
title: Page title
subtitle: One line
cols: 3                # columns for sheet
source: RFC 9293       # any other field shows under the title
---
A sentence or two with the main point.

## A Panel title {span=2 meta="small text, top right"}
Plain Markdown: paragraphs, lists, tables.

```flow LR
A -> B: label
```
  • Every ## heading is a panel. The letters A, B, C are optional and added for you.
  • span=2 makes a panel two columns wide, rows=2 makes it two rows tall, and bare removes its title bar. span is a hint: the page sizes each panel to its content, so wide tables and diagrams need no span. Write span only for a panel that must stand out. rows applies only to the plain grid (without JavaScript, in print and on narrow screens); the justified layout in a browser ignores it.
  • When no component fits, use a ```html or ```svg block to embed raw markup.
  • A placeholder such as <host> is shown as text. Inside a sentence only text-level tags stay (b, kbd, sup, a, span, br, img ...), and tags that break the page (script, style, iframe ...) are shown as text too. Put raw markup in an html or svg block and code in backticks. The render lists every tag it changed.

Full syntax for a component: am help <component>.

Answer on the page

Every page has a Reply button. The reader comments on any panel (the speech-bubble button in its title bar) and copies one Markdown reply back to the agent. Answers and comments survive a reload. Once the reader answers, comments or remarks, a bar at the bottom of the page shows the counts and copies the reply in one click; the Reply sheet shows the reply drawn, with the copied Markdown under a fold.

The Remark button turns on block marking. While it is on, the reader clicks a paragraph, list item, quote, code block, table or $$ formula block, picks a kind (suggestion, keep, question or concern) and may add a note. The mark shows on the block, survives a reload, and comes back in the reply under Remarks with the start of the block quoted. A page title or panel title takes no mark; use the panel comment for it.

For a decision the reader must make, write an ask block in the panel it changes:

```ask
Which cache do we use?
* Redis | keeps data after a restart
- Memcached | simpler, no disk
```
  • The first line is the question. Each option starts with * (the suggestion, selected at the start) or -. Write 2 to 6 options.
  • One choice needs exactly one *. ask multi lets the reader pick several; * marks the options that start picked.
  • The reply says for each decision whether the reader changed it, confirmed the suggestion, or did not answer it. Comment text and remark notes come back quoted with >.

Full syntax: am help ask.

Code blocks

Quote code that exists in the project instead of typing it:

```ts src=server/routes.ts lines=18-30 hl=22
```
  • Any fence whose language is not a component is a code block, with a header and a Copy button.
  • src= reads the file and lines=18-30 picks the lines, numbered as in the file. Leave the block empty. The path is read from the current folder, and only files inside it are quoted.
  • hl=22 or hl=20-22,25 highlights lines. For code you type, title="limits.ts · sketch" names it and start=38 numbers it from line 38.
  • A block longer than 40 lines gets a warning; more than 200 lines is an error. Files that hold keys by convention (.env, *.pem, id_rsa, .ssh/, .git/ …) and files with anything that looks like a key or a token are refused. The rules follow html-plan's.
  • The render lists every file it embedded. The page keeps the path; am patch reads the file again, or keeps the page's copy when the file has moved.

To show a change, paste a unified diff in a diff fence and the CLI draws it:

```diff file=src/code.js
@@ -60,3 +60,3 @@
 export function parseCodeArgs(args) {
-  const attrs = parseAttrs(args);
+  const attrs = parseAttrs(args, { diff: true });
   const opts = {};
```
  • Lines starting with + are added (green), - removed (red), a space is context. @@ -60,3 +60,3 @@ starts a hunk: it shows as a thin row and sets the line numbers, in two gutters (old and new). Without @@, start=N numbers from N; with neither, the block has no numbers.
  • file= names the file in the header, and the language label comes from its extension. A +++ b/path line names it when file= is missing. The header also shows a stat such as +1 −1. Copy copies the diff as written.
  • hl= uses new-side numbers. diff --git, ---, +++ and \ No newline at end of file lines are accepted and not drawn.
  • A line that does not start with +, -, a space or @@ is an error, and so is a cut (...): split the diff into two hunks. A hunk whose counts do not match its @@ header is a warning. src= is not supported with diff: paste the diff. The limits are the same as for other code blocks.

Full syntax: am help code.

Images

For something a diagram cannot show, such as a real screen, use an image that already exists as a file:

![The three viewing modes of the game](/absolute/path/to/screenshot.png)
  • Put the image alone on its line. The alt text becomes its caption, so write what the picture shows. The writing check reads it.
  • Use the absolute path. A relative path is read from the draft file's folder, or from the current folder when the draft comes from stdin. A space in the path is fine.
  • PNG, JPG, GIF, WebP, AVIF and SVG files up to 5 MB. The file is embedded in the page, so the page stays one file that opens offline. A wide image scales down to its panel.
  • http(s) URLs and data: URIs are left as they are. A URL needs the network when the page is opened.
  • The page keeps the path of each image. am patch embeds the image again from the file, or from the page when the file has moved.
  • The tool does not generate images.

Full syntax: am help image.

Math

For a formula in the text, write LaTeX between dollar signs:

The weight of a vertex is $w(v)=\prod_{u\prec v}N_u^{-1}$.

$$
\widehat g=\sum_{i=1}^{n} w_i\widehat A_i\nabla_\theta\log\pi_\theta(a_i\mid s_i)
$$
  • $…$ is an inline formula. A block is a line with only $$, the TeX on the lines after it, and a line with only $$; one line $$ E = mc^2 $$ also works. A block has no blank line inside it.
  • Temml typeset the TeX when the page is made and writes MathML, so the page needs no script and no network. A page with math carries a small math font (about 9 KB) and its styles; a page without math is unchanged. The formula takes the colour of the text, so every theme and mode works.
  • A formula stays left to right on a right-to-left page.
  • A display formula that is wider than its panel scrolls inside the panel. An inline formula wraps after a relation or an operator (=, +) in Chrome and Firefox; Safari does not wrap it. Put a long formula in a $$ block.
  • A dollar sign that is not a formula stays text. The opening $ has a non-space after it. The closing $ has a non-space before it and no digit after it, so $5 and $10 and $5-$10 stay text. A closing $ with a letter (A-Z, a-z) or _ after it does not close either, so $PATH:$HOME stays text; write $n$-th, not $n$th. Write \$ for a dollar sign. An inline formula is on one line.
  • Math works in the text of paragraphs, lists, tables, quotes, headings, callouts, kv, ask, timeline and tree. Code spans, code blocks, diagrams (flow, er, sequence and the others) and am video take no math. In a table cell write \vert x \vert, not |x|.
  • If Temml cannot read the TeX, the page is still written. The formula shows as raw TeX in red and monospace (the message is in its tooltip), and the render prints a warning with the draft line, like L12 [math] $\foo$ is shown as raw TeX: .... A page with a math warning does not open; fix the TeX and render again with --replace.
  • \ref, \eqref and \label need a script that the page does not have. They give a warning and show nothing. \tag{1} works in a $$ block.
  • A Remark or a Reply quotes a formula as the TeX you wrote, with its $ or $$: Energy $E = mc^2$ holds.
  • The writing check skips the TeX.

Syntax in the CLI: am help format.

Use the CLI directly, without an agent

The CLI is scripts/am.mjs inside the skill folder.

AM=skills/answer-me-with-html/scripts/am.mjs

node $AM render examples/tcp.en.md                # render and open in the browser
node $AM render notes.md -o out.html --no-open    # choose the output file, don't open
node $AM render notes.md --theme shadcn           # pick a theme for this run
node $AM patch page.html --panel "Why three messages" < panel.md   # replace one ## panel, overwrite the same file
node $AM lint notes.md                            # writing check only
node $AM list                                     # list components
node $AM config                                   # view settings

# Read from stdin. This is how agents call it.
node $AM render - <<'AM_EOF'
## A One panel
```flow
A -> B: hello
```
AM_EOF

Pages go to ~/.answer-me-with-html/pages/ by default. Set AM_HOME to move them. A page with STE or code warnings does not open in the browser (add --open to open it anyway). To render the fixed draft again, run am render … --replace <page>: the earlier page is deleted once the new one is written, so pages/ keeps one page per answer. The default folder is private to your user (mode 0700), so other users on a shared host cannot read your pages; an existing AM_HOME folder keeps its mode.

Languages

The language of a draft sets the page's lang attribute, the language of the buttons and theme names, the fonts, and the video player labels. A draft can declare it with lang: in the header. Without it, the tool detects the language from the script of the text.

lang: <html lang> Labels and fonts
zh, zh-CN, zh-Hans as written (a bare zh becomes zh-CN) Simplified Chinese
zh-Hant, zh-TW, zh-HK, zh-MO as written Traditional Chinese, with Traditional fonts first
en, en-US and other English tags as written English
ja, ja-JP as written Japanese, with Japanese fonts first
he, he-IL as written Hebrew, with Hebrew fonts first; the page is right to left
any other tag, such as fr or ko as written English labels

A tag is written as you declare it: zh-tw and zh_TW both become zh-TW. An empty, und or invalid value is ignored and the text decides.

A right-to-left language (he, ar, fa, ur, yi) also gets <html dir="rtl">. The page mirrors: text, panels, tables, the table of contents and the toolbar start from the right; flow and sequence diagrams are drawn as their mirror image, so the first node or participant is on the right and arrows point left; timelines, trees and limits bars start from the right. Code blocks, diffs and inline code stay left to right, and so does any run of text with no right-to-left letter (a path such as src/, a key such as lint: at the start of a label, a domain, a signed number), so its punctuation stays where it was written. Hebrew has its own labels, Hebrew names for common frontmatter keys (author, date, source, version ...) and a day.month.year render date; the other right-to-left languages use English labels. A video page (am video) follows the same rule: the title block, the scenes, the captions, and the chapter strip start from the right, while the controls row (play button, clock, progress bar, speed and export) stays left to right, as in common video players. Exported MP4 and WebM files carry the mirrored picture.

Script in the text Detected as
Han zh-CN (Simplified), or zh-Hant when more characters are written only in the Traditional form than only in the Simplified form
Han with kana ja
Hangul ko
Cyrillic, Arabic, Hebrew, Thai, Greek ru, ar, he, th, el (the most common language of the script)
Latin en

Declare the language when you write a Latin-script language other than English, when a Chinese text is too short or too plain to show Traditional characters (characters both forms share say nothing, and the text stays Simplified), or when the exact language matters: Arabic script also writes Persian, and Cyrillic also Ukrainian.

A patched page keeps the language it had, unless the draft declares one.

Diagram text wraps by the unit the script uses: one Han, kana or fullwidth character, one word where the script writes spaces, and one dictionary word for the scripts that write none (Thai, Lao, Khmer and Myanmar), so a long word does not push a node wider than its budget. A word that does not fit the line on its own falls back to graphemes, so a combining mark stays with its base character.

The writing check follows the language. Chinese, English and Japanese keep their rules. Any other language gets only the language-neutral ones: sentence length (in words, or in characters for CJK text) and paragraph length. A script that writes no space between words is measured in words found by the word segmenter, the same four scripts the diagrams wrap with.

Your own theme

Put one JSON file per theme in ~/.answer-me-with-html/themes/. The file name is the theme name, so themes/notes.json is the theme notes. Pick it like a built-in theme: theme: notes in the draft, --theme notes, or am config set theme notes. The agent writes the same draft, so a theme adds nothing to each answer.

{
  "label": "Notes",
  "tokens": {
    "common": { "--font-sans": "\"IBM Plex Sans\", \"Noto Sans CJK SC\"" },
    "light": { "--bg": "#f7f5ef", "--paper": "#fffdf8", "--ink": "#1f1d1a" },
    "dark": { "--bg": "#14130f", "--paper": "#1c1b17", "--ink": "#eeeae0" }
  },
  "css": "& .am-panel-head { letter-spacing: 0.01em; }"
}
  • Light and dark must each set every color variable. am help theme lists them and shows the whole format.
  • Name installed fonts only. The default fonts are added as the fallback, so a reader without your fonts still gets readable text.
  • In css, start every selector with &. It stands for the theme's root, so the rules apply only under this theme.
  • font (optional) names a font this tool ships. A page rendered with the theme embeds it, so it needs no network and the other pages do not grow. The only value for now is bangers (Bangers, SIL OFL 1.1); use the family in css, for example & .am-head h1 { font-family: "Bangers", Impact, sans-serif; }. A version that does not know the field ignores it and shows the fallback font. Video pages do not embed the font.
  • requires (optional) is the oldest version of this tool the theme supports, as x.y.z. On an older version the theme is skipped with needs Answer me with HTML x.y.z or newer; a version that does not know the field ignores it.
  • Keep the file self-contained: no url() or image-set() to http:, https: or //, no @import, no @font-face and no backslash escapes (they can hide a link from the check). A link would send a request every time someone opens a page, and pages work offline. am theme check warns about them; your existing theme keeps working. The theme catalog and am theme add reject them.
  • Each page carries the built-in themes plus its own theme, so it still opens on any machine.
  • am theme check notes reports missing variables, invalid colors and low contrast in light and dark, warns about external resources, and renders two specimen pages with every component.

The classes your css may target stay stable: a later version keeps every one of them, and never adds a required color variable.

Part Classes
Page head .am-head, .am-head h1, .am-head-meta, .am-intro
Panel .am-panel, .am-panel--bare, .am-panel-head, .am-panel-head h2, .am-panel-id, .am-panel-meta, .am-panel-body
Content .am-md, .am-callout, .am-callout--ok, .am-callout--warn, .am-callout--err, .am-callout-title, .am-kv
Drawing frame .am-frame, .am-ruler
Comment button .am-comment-btn, .am-comment-btn--on

Diagram tokens

A theme restyles what is inside a diagram (flow, sequence, er, and the boxes of a tree) with tokens, not with selectors: set them in tokens and every diagram follows. They are all optional. A theme that sets none draws the diagrams it always did, and an older version ignores them.

Part Tokens Default
Node (flow, er, tree box) --node-fill, --node-stroke, --node-width, --node-radius, --node-shadow, --node-text --paper, --line, --bw, 3px (the tree box: --radius), none, --ink
Highlighted node (*) --hi-fill, --hi-stroke, --hi-text --accent-bg, --accent, --accent
Edge and message --edge-color, --edge-width, --edge-label --ink-2, 1.3, --accent
Sequence actor (and the box of a == divider ==) --actor-fill, --actor-stroke, --actor-text --node-fill, --node-stroke, --node-text
Sequence note --note-fill, --note-stroke, --note-width, --note-radius --warn-bg, --warn, 1, 2px
Group box (flow) --cluster-fill, --cluster-stroke --fill, --line-2
Lifeline --lifeline-color --ink-3
  • --node-width, --edge-width, --node-radius, --note-width and --note-radius take a number or a px or em value (a bare number means px). --node-radius also rounds actors and does not change the pill and diamond shapes. --node-shadow is none or one or more drop-shadow(...), for example drop-shadow(3px 3px 0 #111111), and also casts the shadow of actors. Any other value is an error, and the theme is skipped.
  • Colors are any CSS color. In light and dark mode am theme check reads these pairs and needs 4.5:1, only when the theme sets one side: --node-text on --node-fill, --hi-text on --hi-fill, --actor-text on --actor-fill, --ink on --note-fill, and --edge-label on --paper. Dark mode keeps the light value of a token it does not set, so set both.
  • A token value goes through the same external-resource check as css: no url() or image-set() to http:, https: or //, and no backslash. am theme check warns; the catalog and am theme add reject.
  • There is no token for the font family, size, weight or letter spacing in a diagram. The CLI measures diagram text for the sans font, so a different font would overflow its box.
  • The corner radius needs Safari 17.4 or newer; an older Safari draws the default radius.

Theme catalog

answer-me-with-html.com/themes/ shows ready-made themes, each drawn live in light and dark, with the commands that install it: comic, manga and pop, which use the Bangers font. A catalog theme is one file in your themes folder, so a page that does not use it does not carry it. The catalog list is at /themes/index.json (name, version, size and sha256 of every file), and the files are in marketplace/themes/. To add a theme to the catalog, see CONTRIBUTING.md.

Install a theme: am theme add

am theme add comic --use       # download, check and save the theme, and make it the default
am theme list [--remote]       # built-in themes, then yours with version and source
am theme update [name]         # update themes that came from the catalog
am theme remove <name>         # delete a theme file
  • am theme add <name> reads the catalog index (https://answer-me-with-html.com/themes/index.json, or the URL in the AM_CATALOG_URL environment variable: https only, or http for localhost, 127.0.0.1 and [::1], because over plain http the index and the file could both be swapped and the sha256 would prove nothing; a redirect to an address that breaks this rule is refused), downloads themes/<name>.json from the same address, and checks its size and sha256 against the index. It then runs the checks of the catalog CI: a valid theme file, no contrast error, no external resource. It saves the file to ~/.answer-me-with-html/themes/<name>.json through a temporary file, so an interrupted run leaves no half file. Only catalog names work, not URLs or paths. A theme that requires a newer version is refused with the version it needs. Network calls time out after 8 seconds.
  • --use also sets the default theme, the same as am config set theme <name>. It asks nothing else.
  • A file with that name that is not the catalog's is never overwritten, because it may be your own theme: the command stops and suggests --force. The identical file is reported as already installed. A file that is still exactly what the catalog gave you is replaced by a newer version without --force.
  • Where each theme came from is recorded in state.json in the data folder (catalogThemes: name, version and sha256 of the installed file). The themes folder holds theme files only. Without a record, a file counts as your own (local).
  • am theme list needs no network. A file you changed after installing shows as catalog, edited. --remote also reads the catalog and marks themes with an update and themes you do not have.
  • am theme update replaces each theme that came from the catalog and is unchanged since the install; a file you edited is skipped with a note (am theme add <name> --force replaces it). am theme update <name> does one theme.
  • am theme remove <name> deletes the file and its record (never a built-in theme). If it was the default theme, the default goes back to auto. Pages already made keep their look, because a page carries its own theme.
  • The output prints theme names and versions only. Labels and descriptions from the catalog are for the website and are never printed, so the text of a catalog cannot reach an agent through the command.
  • Versions before 0.5.3 have no am theme add: save the file by hand, as the catalog page shows (curl or PowerShell).

Opening pages from a remote machine

If your agent runs on a remote or headless machine, am serve lets you open its pages in your own browser over http://. Run it in a terminal on that machine and leave it running:

am serve              # http://127.0.0.1:8765, Ctrl-C to stop; --port N changes the port (0 picks a free one)
ssh -L 8765:127.0.0.1:8765 user@host   # on your own computer, then open the links in your browser
am serve --lan        # also listen on the local network (0.0.0.0), only on a network you trust
am serve --public-url https://pages.example.com   # links use this address, for a reverse proxy or tunnel that adds TLS
  • While it runs, am render, am patch and am video print a link: line for each page they write into ~/.answer-me-with-html/pages/ or videos/, and the agent gives you that link instead of a file:// one.
  • By default the server listens on 127.0.0.1 only, and the SSH tunnel is the recommended way to reach it. There is no password and no TLS.
  • --lan makes the server listen on 0.0.0.0, every IPv4 interface and not only the address in the links, so another computer on the same network can open the links. It prints every LAN address with a warning, and the links use the first one: 192.168.x.x before 10.x.x.x before 172.16-31.x.x. The connection is plain HTTP and the token in a link is a bearer secret, so anyone on that network who sees the traffic or the link can open the page. Use it only on a trusted network, behind a firewall.
  • --public-url https://host is for a reverse proxy or tunnel that you run in front of am serve and that adds TLS. It sets the address that link: lines start with and the Host name the server accepts. It does not create TLS or open any port itself. Give it an origin only, with no path. Without --lan the server still listens on 127.0.0.1, so the proxy must run on the same machine, and it must pass on the original Host header.
  • Without those flags the server answers only to localhost, 127.0.0.1 and [::1]; a request for any other host name or IP is a 404. Do not expose the port to the internet.
  • Links are per page: each one carries a token for that one page, and it does not open any other page. They stop working when the server restarts. The server shows no folder listing and serves only .html files.
  • Treat a link as the password of its page: anyone who can reach the port and has the link can open the page, including other users of the same machine. Share it only with yourself.
  • Give the tunnel a local port of its own. Do not reuse a port where you opened pages from another server or another host: a page from that other server could have left a service worker in your browser for that address, and it would see the pages served there later.
  • Each page opens in a browser sandbox of its own, so a script in one page cannot read another. Because of that, Reply answers, comments and remarks on an http:// link last only until you close the page: copy the reply before you close it.

Updating and cleaning up

Updating. Updates are manual, and the tool tells you when one is out. Once a week, a background process downloads this project's package.json and highlights.json from GitHub to read the latest version number and its highlights. Nothing about you or your pages is sent, and the page you asked for never waits on it. When there is a newer version, the next render adds a one-line notice and the agent asks whether you want to update. Each page made while you are out of date also has a "New version" button at the start of the toolbar; click it for up to three highlights of the new version in the page language, the update command and a copy button. am serve takes the button out of a page it sends to another device (--lan or --public-url), and the page itself never goes online. Turn this off with /answer-me-with-html:config update_check off.

Installed with Update with
npx skills add npx skills update answer-me-with-html -y, or tell your agent "update answer-me-with-html"
git clone + npm link git pull && npm install in the repository
DeepSeek Harness plugin dsh plugin --profile <profile> up <package>; the update notice fills in the profile and the package
Claude Code plugin In a terminal run claude plugin update answer-me-with-html@answer-me-with-html (or open /plugin → Installed → Update now), then /reload-plugins. To update automatically, turn on auto-update for this marketplace in /plugin → Marketplaces. Claude Code leaves it off for third-party marketplaces

Cleaning up. Pages, videos and the narration cache build up in ~/.answer-me-with-html/. If that folder grows past 200 MB, or passes 20 MB with no cleanup for 30 days, the agent asks once a week whether to clean it. Nothing is deleted without your OK.

  • /answer-me-with-html:clean (plugin), or say "clean up the pages", previews first and then asks.
  • am clean deletes pages and videos older than 30 days and empties the narration cache. --days N changes the cutoff, --all removes every page and video, and --dry-run only shows what would go. Your settings and themes are always kept. If pages, videos or cache is itself a symlink, it is skipped: am clean never touches the files it points to, and the size count and cleanup hint ignore it. A folder that cannot be read is skipped too: its files are left out of the size count and the cleanup hint, so the size you see can be lower than the real use, and am clean does not delete them.

Old always-on plugin. It is gone from this repository, but your copy keeps adding the reminder until you remove it. Run /plugin uninstall answer-me-with-html-always@answer-me-with-html, then add the always-on rule from the README.

Settings

Key Default What it does
open on Open each page in the browser after it is made. Turn it off if pop-ups interrupt you
theme auto Default theme: auto (paper for long text, blueprint for diagrams), blueprint, shadcn, paper, or your own theme
mode auto Default color mode: auto, light or dark
style 80 Writing check: off, 80 (warn only) or strict (refuse to render)
update_check on Check GitHub for a new version once a week and mention it. Never updates by itself
close_after_reply off Close the page tab 3 seconds after you copy the reply, with a countdown and a Keep open button. See the note below
voice auto Video narration: auto (ElevenLabs if ELEVENLABS_API_KEY is set, else system voice), elevenlabs, local, system or off

Settings live in ~/.answer-me-with-html/config.json. A theme written in a draft beats the default. --open and --no-open affect one run only.

close_after_reply applies to pages made while it is on. Pages that am serve sends keep your answers only until the tab closes, so with this setting on the answers go with the tab; the clipboard still holds the reply. A browser refuses to close a tab after a link in the page has moved it (for example a contents link), and then the page says that you can close the tab yourself.

Less proactive

By default the agent makes a page whenever one would help. If that is too much, pick one of two ways. Do not combine either with always-on mode.

Only when you ask in words. Add this rule to your rules file, such as ~/.claude/CLAUDE.md or AGENTS.md:

Do not use the answer-me-with-html skill unless I ask for a page, a diagram or a visual explanation, or say I don't get it.

The rule survives updates. The agent still sees the skill, so it follows the rule by judgment.

Only with the slash command (Claude Code). Add disable-model-invocation: true to the frontmatter of the installed SKILL.md, such as ~/.claude/skills/answer-me-with-html/SKILL.md. The agent then never sees the skill, and a page appears only when you type /answer-me-with-html. An update replaces the file, so add the line again afterwards. See the Claude Code skills docs.

Writing check

Answer me with HTML turns the parts a machine can check into an English and Chinese rule set, and runs it on every render:

  • Length: Steps stay under 20 English words or 35 Chinese characters. Descriptions stay under 25 words or 45 characters. Paragraphs have at most 6 sentences.
  • Words: Prefer common words: "use", not "utilize"; "before", not "prior to". In Chinese, drop empty verbs: write 优化, not 进行优化.
  • Chinese vocabulary: Common typos (登陆 → 登录), vague quantities (尽快, 若干, 大概, 多次), 以上 / 以下 / 以内 after a number (the endpoint is ambiguous), and one-meaning-one-word choices (单击 → 点击, 键入 → 输入, 入参 → 参数). Only the entries that are almost never wrong, taken from Simplified Technical Chinese, a controlled Chinese modelled on the STE method.
  • Style: Flags English passive voice, three or more 的 in one sentence, and stock phrases such as 赋能 and 闭环.
  • Japanese: A draft with kana is treated as Japanese: the page buttons are in Japanese and the page gets lang="ja". Only the length rules apply, with the Chinese character limits. Write lang: ja in the draft to force it.
  • Other languages: Any other language gets only the length rules: sentences in words (characters for Chinese and Japanese text), paragraphs in sentences. The English and Chinese word lists and the passive-voice rule do not run on it.

Set the strictness with /answer-me-with-html:config style strict, or per page with style: in the draft.