A Claude Code output style that makes Claude's prose shorter, more direct, and easier to scan, aimed at making Claude Opus 5 outpus easier to parse. It's inspired by various technical writing style guides, turned into rules Claude applies to every reply.
The style covers chat replies, documentation, commit messages, PR text, and code comments. It does not cover code, identifiers, command syntax, or quoted material.
An output style occupies one slot in the Claude Code system prompt. The
environment blocks around it (tool schemas, working directory, model data,
memory rules) stay in place whatever style is active. This style sets
keep-coding-instructions: true, so Claude Code keeps its coding behavior and
applies the writing standard on top.
| File | Contents |
|---|---|
output-styles/technical-writing.md |
The Technical Writing style. |
-
Copy the style into your Claude Code styles directory:
cp output-styles/technical-writing.md ~/.claude/output-styles/ -
In Claude Code, run
/output-styleand select Technical Writing.
To scope the style to one project, copy the file to .claude/output-styles/
inside that project instead.
- Succinctness — answer first, say each thing once.
- Sentences — one idea per sentence, named actors, active voice, condition before action.
- Words and terminology — the plainest precise word, one name per concept, vocabulary taken from the layer being described.
- Claims and numbers — no unsupported certainty, no time anchors, defensible precision.
- Structure — headings, lists, tables, warnings, and numbered procedures.
- Code comments — no comment by default; a comment must carry what the code cannot.
- Google developer documentation style guide
- Google technical writing courses
- Gernot Heiser's style guide
- nbj-write-clearly
- developer-documentation-skill
MIT. The rules are inspired by the sources above, not copied from them.