AgentSkillsCN

output-style-authoring

为打造能够塑造Claude行为与个性的输出风格提供详尽指南。无论您是创建、编写、设计、构建、审核，还是优化输出风格、人物角色模式，亦或是对行为进行调整与改造，本技能都能助您一臂之力。从风格文件的编写，到持续编码指令的决策，再到人物角色的设计，以及多种输出风格、智能体与技能之间的权衡取舍，本技能皆能游刃有余。同时，它更是输出风格领域中的资深专家，深谙最佳实践之道。

SKILL.md
--- frontmatter
name: output-style-authoring
description: Guide for authoring output-styles that transform Claude's behavior and personality. Use when creating, writing, designing, building, reviewing, or improving output-styles, persona modes, or behavior modifications. Helps with style files, keep-coding-instructions decisions, persona design, and choosing between output-styles, agents, and skills. Expert in style best practices.
allowed-tools:
  - Read
  - Grep
  - Glob
  - Bash
  - AskUserQuestion
# model: inherit

Reference Files

Start here:

Deep dives:

Output-Style Authoring

Guide for creating output-styles that transform Claude's behavior and personality.

Related Skills

This skill is part of the authoring skill family:

  • •agent-authoring - Guide for creating agents
  • •skill-authoring - Guide for creating skills
  • •output-style-authoring - Guide for creating output styles (this skill)

For validation, use the corresponding audit skills:

  • •audit-output-style - Validate output-style configurations
  • •audit-coordinator - Comprehensive multi-faceted audits

Quick Start

New to output-styles? Start with minimal-template.md—a 40-line baseline.

What Are Output-Styles?

System prompts that transform Claude into specialized personas. They modify behavior, personality, and approach for an entire session while keeping all tools available.

Use when:

  • •Transforming into different roles (writer, analyst, teacher)
  • •Changing response style for entire session
  • •Non-engineering use cases need different behavior
  • •Want persistent personality change

Don't use when: Need tool restrictions or separate context (use Agent or Skill instead)

Core Principles

See design-principles.md for deep dive.

  1. •Transform WHO, not HOW - Define persona and approach, not rigid processes
  2. •Start at 40-60 lines - Add complexity only when proven necessary
  3. •Choose strength wisely - Most should be Active, not Passive or Dominant (see persona-strength-spectrum.md)

File Structure

Required:

  • •YAML frontmatter with name and description
  • •Clear persona definition
  • •Concrete behavioral directives

Optional:

  • •keep-coding-instructions: true/false (default: false)
  • •Output format specification
  • •Boundaries (what you don't do)

Location:

  • •User: ~/.claude/output-styles/style-name.md (just you)
  • •Project: .claude/output-styles/style-name.md (whole team)

Patterns: See design-patterns.md for Role Transformation, Teaching Mode, Specialized Professional, and Quality/Audit templates.

Creation Process

Prefer starting with minimal-template.md and customizing.

Step 1: Define Role

  • •What persona should Claude adopt?
  • •Is this engineering-related? (determines keep-coding-instructions)
  • •Does similar style exist? (/output-style to list)
  • •Use AskUserQuestion if unclear

Step 2: Choose Scope

  • •User (~/.claude/output-styles/): Just you, all projects
  • •Project (.claude/output-styles/): Whole team, in git

Step 3: Decide Coding Instructions

  • •keep-coding-instructions: false - Non-engineering roles (writer, teacher, analyst)
  • •keep-coding-instructions: true - Engineering roles (security, DevOps, QA)
  • •Default: false

Step 4: Write Persona

Good: "You are a technical writer focused on clarity and accessibility." Bad: "You help write documentation." (too vague)

Include WHO Claude is, primary goal, and key differentiators.

Step 5: Define Behaviors

Use concrete directives, not adjectives.

Good: "1. Start with why it matters 2. Use examples within 3 paragraphs 3. Define terms on first use" Bad: "Be helpful and explanatory" (doesn't change behavior)

See anti-patterns.md #7 for more on this.

Step 6: Write Description

200-400 chars, explains what + when to use

Good: "Clear, beginner-friendly documentation with examples" Bad: "Helps with writing" (too vague)

Step 7: Create File

  • •Filename: style-name.md (lowercase-with-hyphens)
  • •Location: ~/.claude/output-styles/ or .claude/output-styles/
  • •Use minimal-template.md as starting point

Step 8: Test

Activate with /output-style style-name and verify:

  1. •Appears in menu
  2. •Behavior changes noticeably
  3. •Adds value to session
  4. •Test mid-session deactivation (shouldn't break conversation)

When to Use What

See when-to-use-what.md for complete guide.

Quick:

  • •Output-Style: Transform behavior for entire session (writer, analyst, teacher)
  • •Agent: Separate context with tool restrictions
  • •Skill: Conditional knowledge, auto-triggers

Critical Warnings

See anti-patterns.md for full diagnostic checklist.

Common smells:

  1. •Persona vs process blur - Specifying HOW to execute, not WHO Claude is
  2. •Over 100 lines - Likely over-built; audit each section
  3. •Reads like user checklist - Instructions for user, not Claude's behavior
  4. •Mid-session removal breaks conversation - Too invasive
  5. •Trying to restrict tools - Can't enforce; use Agent/Skill instead

Quick tests:

  • •Can you describe persona in one sentence?
  • •Would mid-session deactivation break conversation?
  • •Is this WHO or HOW? (should be WHO)

Examples & Resources

Quick Checklist

  • • Clear role/persona
  • • User or project scope
  • • keep-coding-instructions decision
  • • Concrete behaviors (not adjectives)
  • • 40-80 lines (justify if longer)
  • • Test activation and deactivation
  • • Refine based on usage