
Stop Re-Prompting for the Same Voice Every Session: Claude Code's Five Built-In Output Styles
Chris Harper
2 min read
Aug 20, 2026 · 12:11 UTC
Claude Code ships five built-in output styles that change how it responds — tone, detail, execution posture — without changing what it's allowed to do. Concise was added today in v2.1.237. If you keep re-prompting "be brief" or "explain your choices", you want a style instead.
The five styles and when each fits:
| Style | What it changes | Best session type |
|---|---|---|
| Default | Standard Claude Code | Routine engineering |
| Concise (new today) | Leads with results, skips narration | Experienced users, fast iteration |
| Proactive | Executes immediately, assumes reasonable choices | Hands-off sessions, long background runs |
| Explanatory | Adds "Insights" explaining choices | Learning a new codebase, audit trail |
| Learning | Insights + TODO(human) markers | Active pair-programming, onboarding |
Proactive vs. auto mode: Proactive adds stronger autonomous-execution guidance than auto mode does — Claude decides faster and narrates less — but it doesn't change your permission mode. Your permission settings still gate what runs without a prompt. If you want auto mode AND less narration, Proactive gives you that without opening new permissions.
Set it:
# Interactive: run in any session
/config # → Output style → pick from menu
# Or set in .claude/settings.json for a project default
{ "outputStyle": "concise" }
Takes effect after /clear or a new session.
Token impact: Explanatory and Learning produce noticeably longer responses — the trade-off is worth it when you need the reasoning, wasteful when you don't. Styles inject a few hundred tokens into the system prompt, but prompt caching handles that overhead after the first request in a session.
Limits: One style per session — you can't blend them. Custom styles aren't in the built-in menu (you'd need a hook or SDK system-prompt edit). Independent practitioners note Proactive can miss edge cases that Default would flag; it's best for sessions where you've already validated the approach.
Sources: Output styles — code.claude.com · Pair programming with Claude Code output styles — Shipyard · Output styles practical reference — ClaudeKit