Omniscio

OmniscioGuidesClaude Code output styles

Claude Code output styles

An output style is a standing instruction that shapes every reply in a session, so you say “be brief” once instead of in every prompt. Claude Code ships four besides its default, and you can write your own.

This page covers what each built-in style changes, the three ways to switch, how to write and load a custom style, and where a style stops being the right tool. Every claim names its source at the end.

Last checked against Anthropic's output styles documentation on November 17, 2026. By the Omniscio team. Updated .

On this page

The setting lives in the same files as everything else Claude Code reads at startup: where settings.json lives covers the levels and their precedence.

What an output style is

Anthropic describes an output style as a set of instructions that sets Claude's role, tone and response format for every response in a session. It is not a setting on the model. It is text added to what Claude is told to do.

That makes it useful for the things you would otherwise repeat: shorter answers, an explanation of each change, starting work without asking about routine decisions, or turning Claude into something other than a software engineer, such as a writing assistant.

Claude Code starts in the Default style, which adds nothing and runs on its standard software engineering instructions. Each of the four other built-in styles keeps those instructions and adds its own.

Sources: the introduction and built-in styles sections of Anthropic's output styles page, read November 17, 2026.

The four built-in styles

Proactive starts implementing as soon as you send a task. It makes reasonable assumptions about routine decisions instead of stopping to ask, and it does not switch to plan mode unless you ask for one.

Concise leads with the result. The first sentence says what happened or what the answer is, and the preamble, step-by-step narration and closing recap are left out. Anthropic says it needs Claude Code v2.1.237 or later, which claude --version tells you.

Explanatory does the task the way Default does, then adds short blocks labelled Insight that explain the choices behind the code. They appear in the conversation, never as comments in your files.

Learning adds the same Insight blocks and also asks you to write small pieces yourself. It marks the spot with a TODO(human) comment, says what to write and what to weigh, then stops and waits for you.

Proactive deliberately breaks the habit of asking first. Anthropic states that its instruction to check with you before deleting data or changing a shared or production system is an instruction Claude follows, separate from permission prompts.

Switching style does not change your permission mode. What runs without asking you is still decided by the permission mode, so the prompts look the same as before. The permission modes guide covers those.

Sources: the Proactive, Concise, Explanatory and Learning sections of Anthropic's output styles page, read November 17, 2026.

Switch to a style

The command is the shortest route. With an argument it switches; with none it lists the styles you can pick and marks the current one:

/output-style concise

The command and the terminal menu save the choice to .claude/settings.local.json at the local project level. Run /config and select Output style to pick from a menu instead.

To set a style without a menu, put outputStyle in a settings file. The value is case-sensitive here, so write the built-in names exactly as shown:

{
  "outputStyle": "Explanatory"
}

A value that does not match a style name exactly, such as explanatory, gives you the Default style. The /output-style command ignores case, so the two routes are not the same route.

To make a style the default in every project, set outputStyle in ~/.claude/settings.json. A project's own settings files take precedence over that value.

Anthropic also documents the command in non-interactive mode and Agent SDK sessions, from v2.1.269 or later, where only the built-in styles can be listed and chosen from the mobile app or web through Remote Control. The VS Code extension has its own Output styles row in its command menu, from v2.1.257.

Sources: the change-your-output-style section of Anthropic's output styles page, read November 17, 2026.

Write a custom style

A custom output style is a Markdown file: frontmatter for the metadata, then the instructions themselves. Save it at one of three levels, and the file name becomes the style name unless the frontmatter sets one.

  • ~/.claude/output-styles for every project on the machine.
  • .claude/output-styles for one project.
  • .claude/output-styles inside the managed settings directory, for a policy an organisation deploys.

So a file at ~/.claude/output-styles/diagrams-first.md with no name field is the style called diagrams-first. The .md is not part of the name.

Project styles load from every .claude/output-styles/ between the working directory and the repository root. When two of those nested folders define the same style name, the one closest to the working directory wins.

Here is Anthropic's own example, cut to its first paragraph. It leads every explanation with a diagram and keeps Claude coding the way it normally does, which is what keep-coding-instructions: true asks for. On the full system prompt, a custom style leaves Claude Code's engineering instructions out unless that field is true, so leave the field out when Claude will not be doing software engineering at all.

---
name: Diagrams first
description: Lead every explanation with a diagram
keep-coding-instructions: true
---

When explaining code, architecture, or data flow, start with a Mermaid diagram showing the structure, then explain in prose.

The other fields are name, description, and force-for-plugin. That last one belongs to styles shipped inside a plugin, and applies the style automatically whenever the plugin is enabled. All fields are optional.

Then run /output-style with no argument and pick it from the list. In the terminal, Claude Code reads style files when it starts, so a style you create or edit mid-session appears after a restart.

The rules Claude Code ships with, and the CLAUDE.md file it loads for every session, are a different mechanism. The CLAUDE.md guide explains which one to put a given instruction in.

Sources: the create-a-custom-output-style section and its frontmatter table on Anthropic's output styles page, read November 17, 2026.

Style, or a different feature

An output style applies to every reply in a session and nothing enforces it. When what you want is narrower than that, or has to happen without fail, another feature fits better. Anthropic's own table draws the line:

  • Every reply in a voice, length or format is an output style, because it covers the whole session and switches with one command.
  • What Claude should know about the project goes in CLAUDE.md, which stays loaded whichever style you pick.
  • Instructions for one kind of task, such as a release checklist, are a skill, loaded only when the task matches.
  • Something that must happen every time, such as formatting after an edit or blocking a command, is a hook, which Claude Code runs itself.
  • A helper with its own instructions, model and tools is a subagent, which runs in its own context.
  • An addition to the system prompt at launch is the --append-system-prompt flag, which appends without removing anything.

They combine. A common setup puts what Claude should know in CLAUDE.md, how it should reply in an output style, and anything that must be guaranteed in a hook.

Source: the choose-between-an-output-style-and-other-features table on Anthropic's output styles page, read November 17, 2026.

Limits and costs

A style is an instruction, not a rule. Anthropic says plainly that it does not guarantee something always happens or never happens. If a step must be guaranteed, a hook is the only mechanism that does not depend on Claude following words.

Subagents are not covered. A style applies to the main conversation and to a fork of it, which inherits the parent's whole conversation and system prompt. Every other subagent runs its own system prompt.

Styles cost tokens. The instructions are input tokens on every request, though prompt caching reduces that after the first request. Explanatory and Learning produce longer replies by design, which raises output tokens; Concise does the opposite.

A misspelled frontmatter field is ignored without an error, and YAML that does not parse still loads the style under its file name with no fields set. When a style seems to do nothing, run claude --debug to see the parse error.

Built-in styles are fixed text. If you want a built-in style with one thing changed, copy its idea into a custom file rather than expecting a setting to adjust it. For the visual theme, which is a separate setting, see dark mode and themes.

Sources: the how-output-styles-work and frontmatter sections of Anthropic's output styles page, read November 17, 2026.

Running several sessions at once

A style is saved per project, so several sessions in the same repository share it. Omniscio, the desktop app we make for running many Claude Code sessions at once, puts those sessions in one window with a sidebar instead of separate terminals. It installs on Windows, macOS and Linux.

It is free to use with up to 8 projects. Pro is $20/month or $192/year, and neither plan includes an AI allowance.

The multi-session guide compares every way to run more than one session, and the effort level guide covers how hard Claude thinks on a turn.

Where these answers come from

  • Anthropic, Output styles: the definition, the four built-in styles, every switching route, the three file locations, the frontmatter table, how styles work with subagents and tokens, and the feature comparison table. Read November 17, 2026.
  • Anthropic, Settings: settings precedence between user and project files. Read November 17, 2026.

Anthropic's pages change often, and the version numbers above are the ones its own page stated when it was read on November 17, 2026. Read them again before trusting an older copy.

Common questions

What are output styles in Claude Code?

An output style is a set of instructions that sets Claude's role, tone and response format for every reply in a session. Claude Code ships four besides its default: Proactive, Concise, Explanatory and Learning. Run /output-style to pick one, and the choice is saved to .claude/settings.local.json.

How do I switch Claude Code to a shorter output style?

Run /output-style concise. The Concise style leads with the result and drops preamble, narration and recaps. It needs Claude Code v2.1.237 or later, and it still writes at full length for error reports, failing test output and anything you ask to have explained.

Where do custom output styles go?

Save a Markdown file in ~/.claude/output-styles for every project, or .claude/output-styles in a project. Project styles load from every such folder between the working directory and the repository root. The file name becomes the style name unless the frontmatter sets one.

Why did my custom output style lose the coding instructions?

Because keep-coding-instructions is false by default. On the full system prompt, a custom style leaves out Claude Code's software engineering section unless the frontmatter sets keep-coding-instructions: true. Set it when you are changing how Claude talks, not what it does.

Does an output style apply to subagents?

It applies to the main conversation and to a fork of it. Other subagents run their own system prompt, so the style does not change how they respond. For a helper with its own instructions, define a subagent instead.

Why is my new output style not in the list?

In the terminal, Claude Code reads style files when it starts, so a style created or edited during a running session appears after you restart it. A misspelled frontmatter field is ignored without an error, and YAML that does not parse loads the style under its file name with no fields set.

Set the style once, then run several sessions in one window

Omniscio sits on top of the Claude Code you already have and runs many sessions side by side. The features page lists what it does today.