Output Styles Reference

Output Styles Reference

Claude Kit ships 5 Claude Code output styles — system-prompt overlays that change how Claude communicates and reasons for the entire session. Output styles are auto-discovered when the plugin is installed; no /claudekit:init step required.

All 5 styles use keep-coding-instructions: true, so Claude’s default coding/testing/verification discipline still applies underneath. The style adds posture and format on top.

Switching styles

Via /config (recommended)

/config

Pick Output style from the menu, then choose one of the 5 styles. The choice persists across sessions.

Via settings file

Edit .claude/settings.local.json (project) or ~/.claude/settings.json (personal):

{
  "outputStyle": "Brainstorm"
}

Built-in vs claudekit styles

Claude Code has built-in styles (Default, Explanatory, Learning). Claudekit adds 5 more: Brainstorm, Deep Research, Implementation, Review, Token Efficient. They appear together in the /config picker.

---

The 5 styles

Brainstorm

Creative exploration mode — divergent thinking, multiple alternatives, structured trade-offs before any code.

• Posture: Diverge first, converge second. Surface 2-3 distinct approaches before recommending one.
• Output format: Lettered approaches with pros / cons / effort, then a one-line recommendation.
• Best for: Feature design, architecture decisions, exploring alternatives.

APPROACH A: <name>
  Summary: <1 sentence>
  Pros: ...
  Cons: ...
  Effort: <S/M/L/XL>

APPROACH B: <name>
  ...

RECOMMENDATION: <which one and why>

Deep Research

Thorough investigation mode — completeness over speed, evidence-cited findings, confidence levels named.

• Posture: Cite, don’t recall. Every claim has a source — file:line, doc URL, or command output.
• Output format: Structured reports with Question / Method / Findings (with confidence) / Conclusions / Gaps.
• Best for: Technology evaluation, incident investigation, security audits, due diligence.

Implementation

Code-focused execution mode — minimal prose, action-oriented updates, follow established patterns.

• Posture: Execute, don’t deliberate. The decisions were made upstream.
• Output format: Per-file edits with code blocks, then test-run output, then commit.
• Best for: Executing approved plans, repetitive tasks, when design is already decided.

Creating `src/services/user-service.ts`
[code]

Running tests... ✓ 5 passing
Committing: feat(user): add user service

Review

Critical analysis mode — find issues first, severity-tagged findings, actionable suggestions.

• Posture: Find first, fix second. A reviewer’s job is to surface issues with concrete file:line locations.
• Output format: Findings tagged Critical / Important / Minor / Nitpick with file citations.
• Best for: Pre-merge code review, security audits, architecture review.

### Critical (must fix before merge)
1. **<issue>** — `file:line`
   - Problem: ...
   - Fix: ...

Token Efficient

Compressed output mode — minimal prose, code-first, no preambles.

• Posture: Skip ceremony. No “Sure, I can help” / “Let me explain first” — just do.
• Output format: Code blocks with one-line captions; reference docs instead of re-explaining mechanism.
• Best for: High-volume sessions, repeated similar tasks, cost-conscious work.
• Saving: 40-60% on average vs default verbosity.

---

Style comparison

Style	Verbosity	Focus	Output shape
Brainstorm	High	Exploration	Approach tables + trade-offs
Deep Research	High	Analysis	Structured reports with citations
Implementation	Low	Execution	Code-first per-file blocks
Review	Medium	Quality	Severity-tagged issue lists
Token Efficient	Minimal	Density	Code with one-line captions

Customizing

Output styles are markdown files at the plugin root in output-styles/. To customize, copy the file you want to modify into .claude/output-styles/<name>.md (project) or ~/.claude/output-styles/<name>.md (personal). Project styles override personal styles, which override plugin-shipped styles.

Format:

---
name: My Custom Style
description: A short description shown in the /config picker
keep-coding-instructions: true
---

# My Custom Style

[behavioral instructions...]

Set keep-coding-instructions: false if you want to fully replace Claude’s default coding discipline (rare; usually leave it true).