ailiteracynepal 🇳🇵
Text size

Chapter 03 · Section IV · 10 min read

3.4 Output styles: prompts, creation, switching

Output styles are named system-prompt overlays that shape how Claude replies for a whole session. Learn to author one and to switch between them.

An output style is a named system-prompt override that changes how Claude replies for the entire session. Same underlying tools, same model, but tighter or looser writing, different formatting, a different persona if you must. Good for teams that want one shape of output across everyone’s terminal.

Where they live

Output styles live at <repo>/.claude/output-styles/<name>.md or ~/.claude/output-styles/<name>.md. Same precedence rules as everything else.

Anatomy

---
name: brief
description: Short, terse replies. No preambles. No sign-offs.
---

You reply briefly. Follow these rules for every response:

- No greeting, no sign-off.
- No "I'll help you with that" / "Let me look into that" preambles.
- Answer questions with the smallest complete response. If a code
  change is needed, produce the diff, not an explanation of the diff.
- If you have a question, ask it as one sentence.
- If a task is complete, say so and stop. Do not summarise what you
  just did unless asked.

The file body is spliced into the system prompt when the style is active. It stacks on top of the CLI’s built-in system prompt rather than replacing it — so you are shaping behaviour, not rebuilding it.

Creation

You can hand-write the file, or bootstrap one from a session:

/output-style create brief

Claude Code opens an editor with a scaffold you can fill in. Save; the style is available immediately.

Switching

/output-style          # list available styles
/output-style brief    # switch to `brief`
/output-style default  # revert

Switching is scoped to the current session. The next session opens with the default style unless you set one in settings:

{
  "outputStyle": "brief"
}

When output styles earn their keep

  • Team consistency. Every teammate gets the same style of replies for the same repo — same terseness, same formatting.
  • Personal preference. You just prefer terser output; codify it once, forget about it.
  • Task-shaped modes. A pair-review style that always asks clarifying questions before edits; a code-only style that produces diffs and nothing else.

When not to reach for output styles

  • For a one-off task. Just say it in your prompt.
  • To grant tools. Output styles do not touch tool allowlists — those live in permissions.
  • To change the underlying model. That is model in settings, not output style.

Check your understanding

Quick check

You want every Claude Code reply in your repo to skip preambles and sign-offs and produce diffs rather than long explanations. Which primitive is right?