Claude Code Output Styles: Give Claude a Personality

Claude Code output styles change how Claude responds — its role, tone and default answer format — by editing the system prompt itself, the standing job description Claude reads before every session. You switch between them in /config under Output style, and you can write your own as a small Markdown file.

The analogy I use: same employee, different job description. Output styles do not change what Claude knows or which model runs underneath — they change how it behaves every single turn. If you keep re-prompting «explain it simply» or «answer in bullet points» over and over, that instruction belongs in an output style, not in your fortieth message of the day.

1What is a custom output style in Claude Code?

A custom output style is one Markdown file: a short frontmatter header for metadata, then the instructions you want appended to the system prompt. Save it at ~/.claude/output-styles/ to use everywhere, or in a project's .claude/output-styles/ for that project only. The file name becomes the style name unless you set a name field. The one field that really matters is keep-coding-instructions. Set it to true when you only change how Claude communicates but still want its software-engineering brain intact. Leave it out (it defaults to false) when Claude is not doing engineering at all — say, a pure writing assistant or data analyst.

2Which built-in styles does Claude Code have?

Five. Default is the standard system prompt, tuned for getting work done efficiently. Then four alternatives. Proactive makes Claude execute immediately and make reasonable assumptions instead of pausing for routine decisions — stronger autonomy without touching your permission settings, so approval prompts still appear. Concise is the get-out-of-my-way style: Claude leads with the result, skips the preamble and narration, and keeps answers short by default, while doing the engineering work just as thoroughly as Default. Ask for more detail and it answers in full; error reports and destructive-action confirmations always stay complete. Requires v2.1.237 or later. Explanatory adds educational Insights along the way, explaining why it chose each approach. Learning is the interactive one: Claude shares insights and asks you to write small strategic pieces yourself, leaving TODO(human) markers in the code for you to fill in. If you came here looking for «обучение Claude Code» or a free course on how to work with Claude — the Learning style is the closest thing to a built-in course.

3How do you switch output styles, and is /output-style back?

Run /config, select Output style, pick from the menu — or run /output-style <style> directly, for example /output-style concise (needs v2.1.269 or later). Your choice lands in .claude/settings.local.json as the outputStyle field, which you can also edit directly. Since v2.1.251, switching mid-session applies starting with your very next message — you no longer need /clear or a new session for it to take effect.

{ "outputStyle": "Explanatory" }

4How do you make Claude write in your style?

5Do output styles change the model, or force structured output?

Neither, and both questions come up constantly. Styles do not pick the model — that is a separate setting — and they are not a hard guarantee of structure like an API schema. They are standing instructions: strong defaults Claude follows every turn, reinforced by reminders during the conversation.

Output styles vs CLAUDE.md, skills and agents

Know the boundaries with neighboring features, because this is where people get lost. CLAUDE.md is for facts about your project that Claude should always know. Skills are reusable workflows loaded on demand. Agents are separate helpers with their own prompt and tools. Output styles are for one thing: the default voice and format of every response. Claude Code best practices in one sentence: facts in CLAUDE.md, procedures in skills, personality in an output style.

The kicker: you do not have to write the style yourself. Tell Claude — «Draft an output style based on how I have been correcting you this session» — then save it, select it in /config, and stop repeating yourself. That is the whole point of the feature.