Adds `Coder.StepLength`, a Vale `script` rule for [Keep steps
short](https://github.com/coder/coder/blob/main/docs/.style/style-guide/procedural-writing.md#keep-steps-short).
It flags an ordered-list step whose instruction, the step's first
paragraph, has more than 20 words, counted as [How to count
words](https://github.com/coder/coder/blob/main/docs/.style/style-guide/voice-and-tone.md#how-to-count-words)
describes.
Paragraphs below the instruction explain the step and aren't counted.
- Severity: `warning`.
- Scope: `docs/**` except `docs/.style/style-guide/**`.
- Baseline: 124 findings in 60 hand-written files.
The cleanup shortens each long instruction, moves its explanation into
one paragraph below it, or splits a step that holds several actions.
After cleanup `vale docs/` reports 0 `Coder.StepLength` findings, and
the other rules report no new findings on the changed files.
- The guide section now says the limit covers the instruction, and that
one action or one explanation must not be broken across paragraphs to
meet it.
- Generated pages (`docs/reference/{api,cli,agent-api}`,
`configuration-reference.md`, `prometheus.md`): no findings, so no
`.vale.ini` exclusion is needed.
Linear: DOCS-1272
<details><summary>Decision log</summary>
**Review change (Ian's review)**
- The first version limited every paragraph of a step to 20 words.
The cleanup for it chopped explanations into several short paragraphs,
which made steps harder to read and sometimes changed their meaning.
- The rule now counts only the first paragraph segment of each step.
The content files were reset to `main` and the cleanup was redone
against the narrower rule.
- Cleanup preferences, in order: shorten the instruction; keep a short
instruction and put the explanation, with its original wording, in one
paragraph below it; split a step into one step for each action; for
descriptive lists (such as "How it works"), keep each item one
statement.
- `docs/admin/users/oidc-auth/microsoft.md` keeps the split into
separate steps, with Ian's wording suggestions applied.
"click" became "select" in step 16 because `Coder.SelectClick` flags it.
**Rule shape**
- A `script` rule (Tengo) with `scope: raw`, because the count depends
on Markdown structure that per-paragraph scopes hide.
- A step is an ordered-list item with a `1.` or `1)` marker.
A nested ordered item is its own step.
Bullets, even under a step, are not steps.
- The instruction ends at a blank line, a fence, a nested list, a table
row, a blockquote, block-level HTML, or an explicit hard break (`<br>`,
a trailing backslash, or two trailing spaces).
- Fenced code blocks, HTML comments, tables, images, and nested bullets
add no words.
- CRLF: `\r` counts as whitespace, and offsets come from the raw text.
**Counting approximations (also in the rule header)**
- A code span and a `**bold**` or `__bold__` span are one word each.
- Words split on whitespace, so a hyphenated compound and
`30 seconds` are one word.
- Link text counts word by word, destinations are skipped, images add no
words, and `<https://...>` autolinks are one word.
- A token with no letter or digit, such as the `>` in `**A** > **B**`,
is not a word.
- Tengo indexes strings by rune but measures them in bytes, so the
script rewrites each non-ASCII rune to ASCII of the same byte length
first.
**Tests** (fixtures under `/tmp/sltest/`, not committed; run with the
repo `.vale.ini`, LF and CRLF)
- Flagged: a long instruction, a short action and a long result in one
paragraph, a long nested ordered step.
- Not flagged: a long paragraph below a short instruction, text after a
`<br>` hard break, a long paragraph after a code block, a long nested
bullet.
- LF and CRLF output is identical.
- Pre-commit hooks (`fmt/markdown`, `lint/markdown`, `lint/emdash`,
`lint/style-claims`, `lint/typos`) pass.
- Not run on Windows or macOS.
The rule is plain text processing, and the CRLF cases simulate Windows
line endings on Linux.
</details>
*Generated by [Coder Agents](https://coder.com/docs/ai-coder/agents) on
behalf of @nickvigilante.*