google-style
Skill4 starsChanged 3 months agoApache-2.0View on GitHub
Use when asked to apply, check, or rewrite something against the Google developer documentation style guide, or when /google-style is typed. Applies to READMEs, AGENTS.md, runbooks, API docs, procedures, project docs, specs, and PR descriptions. Never apply this to a cover letter, email, or personal post unless asked for it by name.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- marcus/marcus-skillstheir own version, first seen Aug 8, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- marcus/marcus-skills★ 4same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 4 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 35
No coinages or figurative language.
- Distinctive: line 50
Before editing anything in voice mode, read that file: If no voice guide is configured, say so and stay in doc mode rather than guessing at …
- Distinctive: line 73
Re-run the checker: Every remaining failure needs a stated reason to keep, not silence.
- Distinctive: line 160
Treat the checker as the floor it is.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Google developer documentation style
Edit a document to the Google developer documentation style guide (developers.google.com/style). Optimizes for a reader who is skimming under time pressure, may not be a native English speaker, and may be reading a machine translation.
Target: $ARGUMENTS. If no file was named, ask which document, or apply the
rules to the reply being written in this turn.
First: classify the document, then pick the mode
This guide is a technical-documentation house style. It is right for docs and wrong for anything carrying a personal or brand voice. Applying it everywhere produces exactly the bloodless register that makes writing sound machine-made.
| The document is | Mode | What applies |
| README, AGENTS.md, runbook, API doc, procedure, how-to, project doc, PR description, code comment, spec | doc | Everything below |
| Cover letter, personal essay, marketing copy, email, social post, positioning doc, resume | voice | Clarity rules only. The configured voice guide wins every conflict, no exceptions. |
| Meeting notes, journal, scratch file | neither | Say so and stop. Don't polish a scratch file. |
In voice mode, these Google rules are suspended. They are house rules for reference documentation, and enforcing them on voiced writing is a failure, not a compromise:
- No exclamation points.
- Second person only, never "we" or "I".
- No parenthetical asides.
- No coinages or figurative language.
- No "currently" or "as of this writing".
Which of these matter, and what replaces them, is the voice guide's job — not this skill's.
Configure your voice guide
Voice mode needs one file describing how the writing should sound. Copy
VOICE-GUIDE.example.md, fill it in, and point at it:
VOICE_GUIDE=path/to/your-voice-guide.mdBefore editing anything in voice mode, read that file. If no voice guide is configured, say so and stay in doc mode rather than guessing at a voice.
The strongest version of this pairs the guide with a folder of samples the person actually wrote. A style guide describes a voice abstractly, and text can satisfy every rule in it and still read as machine-written. Samples are the thing itself.
Workflow
- Classify the document and state the mode out loud before editing.
- Run the checker on the current file to get a baseline:
python3 ${CLAUDE_SKILL_DIR}/scripts/gstyle-check.py <file> [--mode voice] - Read the document. The checker finds words. It cannot tell whether the procedure works, whether the sections are in the right order, or whether a reader can find what they need. Those are the edits that matter most, and in practice they are where every worthwhile finding has come from.
- Check any fenced block that holds prose separately. The checker strips code fences to avoid false positives, so a document that is mostly fenced is a document that mostly went unchecked. Extract and run it as its own file. This has repeatedly been where the real errors were.
- Edit, applying the rules below. Show the diff and the reasoning before overwriting anything you did not write.
- Re-run the checker. Every remaining failure needs a stated reason to keep, not silence.
The rules that matter most
Ranked by how often they actually improve a document.
- Put the condition before the instruction. "To save your changes, click Save", not "Click Save to save your changes." The reader who already knows they want to save stops reading at the comma; the reader who doesn't needs the condition first to know whether the sentence is for them.
- Second person, active voice, present tense. "You configure the token", not "The token is configured by the user" or "The token will be configured." Name who does the thing.
- Sentence case for every heading and title. "Set up the build pipeline", not "Set Up The Build Pipeline".
- Descriptive link text. The link text must say where it goes. Never "click here", "this link", "read more", or a bare URL.
<!-- gstyle-ignore-start -->- Cut the words that minimize the reader's difficulty: simply, easy, easily, just, obviously, clearly, of course. If it were obvious they would not be reading this line.
- Spell out the Latin and the vagueness: "for example" not e.g., "that is" not i.e., a real list not etc. or "and so on", "and" or "or" not and/or, "use" not utilize or leverage.
- Short sentences and simple words. Under about 25 words. "Start" not "commence". At most two nouns stacked as modifiers.
- Serial commas. "tokens, secrets, and keys".
- Unambiguous dates. 2026-08-20 or 20 August 2026. Never 8/20/26.
- No directional references. "See Configure the token", not "see above". Layout changes; section names don't, and a pointer that says "above" is wrong the moment anything moves.
- Define every abbreviation on first use, then use it consistently. Use one word for one thing throughout, even where a synonym would read better.
- Formatting carries meaning: code font for filenames, commands, class and method names, placeholders, and console output. Bold for UI elements the reader clicks. Italic for a term being introduced. Underline for links only.
- Numbered lists for sequences, bulleted lists for sets, description lists for term-definition pairs. Don't number a set.
- Inclusive terms: allowlist/blocklist, primary/replica, placeholder (not dummy), person-hours, "stops responding" (not hangs). Drop crazy, insane, sanity check, cripple, guys.
<!-- gstyle-ignore-end -->- Don't pre-announce. No "coming soon", "in a future release", "we plan to". Document what exists.
Fuller detail is in reference.md in this skill's directory. Load it when a
specific question comes up, not by default.
Where this departs from Google
State these rather than silently following either side:
- Some inclusive substitutions lose meaning. "Hangs" and "doesn't respond" are different failures, and a runbook needs the distinction. Prefer the substitution where it is accurate; keep the precise term where changing it would mislead an on-call engineer, and say why.
- The word list carries Google product branding. Skip that section entirely. It is house naming, not style.
- "Just" and "via" are warnings, not failures. "Just" meaning "only" is often the precise word, and it is the most common false positive here.
- Quoted material is never edited. Error messages, UI labels, log output, and other people's words stay verbatim, including when they violate every rule above. A guide's own bad examples are the point of the example.
Common mistakes
| Mistake | Fix |
| Running the checker and stopping there | It checks words, not structure. Read the document; fix the ordering and the gaps too. |
| Trusting a clean result on a fence-heavy file | The checker skipped the fences. Extract prose blocks and check them separately. |
| Applying doc mode to voiced writing | Classify first. In voice mode the voice guide wins. |
| Rewriting quoted material | Examples, error text, and quotations stay exactly as they are. |
| Editing imperatives in a file that governs behavior | Rewording an instruction the system depends on, to satisfy a style rule, is a bad trade. Leave it. |
| Chasing every warning to zero | Warnings need judgment. A stated reason to keep one is a valid outcome. |
Real-world note
Across the first several documents this was run on, the checker's word-level rules found little: the vocabulary was already fine. Every finding that mattered came from step 3 or step 4 — a section organized as a changelog instead of by what the reader needs first, a cross-reference pointing at a section 108 lines the wrong way, an ASCII diagram carrying information no screen reader could reach, and a heading rename that would have silently broken every link to it.
Treat the checker as the floor it is.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

