Output styles

An output style changes how Claude Code answers by modifying its system prompt: role, tone, format, or teaching approach. Keep project rules, coding conventions, and durable context in Memory and rules, not an output style.

#Built-in styles

Proactive is not a permission mode. Tools still follow permission rules.

#Switch styles

Use /config:

/config

Select Output style. The choice is normally stored in .claude/settings.local.json:

{
  "outputStyle": "Explanatory"
}

Because the style changes the system prompt, changing it mid-session does not immediately replace the loaded prompt. Use /clear or start a new session.

#Custom output styles

A custom style is a Markdown file. Its filename supplies the default name unless frontmatter overrides it.

Example:

---
name: Diagrams first
description: Start code explanations with a Mermaid diagram
keep-coding-instructions: true
---

When explaining code, architecture, or data flow, show a Mermaid diagram first,
then explain it in short paragraphs. Keep the diagram under 15 nodes.

Set keep-coding-instructions: true when changing presentation while retaining software-engineering behavior. You can omit it for writing, data analysis, or another non-coding role.

#Frontmatter

Plugins can ship styles in output-styles/. Users can select them through /config after enabling the plugin.

#Boundaries with other mechanisms

#Cost and caching

Choose the style at the start of a task. Repeatedly changing tone mid-task is usually inefficient.

#Passion8 recommendations

Passion8 does not change the output-style mechanism. Model availability, billing, and caching still depend on the actual Claude Code request and gateway behavior.

#Official references

• Output styles
• Settings
• Prompt caching
• Plugins