Do I need to restart my session after creating a new style?

If you select the style via the /output-style command, it reloads instantly. However, if you manually edit the settings.json file, a session restart is required to apply changes.

Where are output style files stored?

Styles can be stored globally at ~/.claude/output-styles/ or locally within a specific project directory at .claude/output-styles/.

How do I switch output styles in an active session?

You can use the /output-style command followed by the style name for an immediate hot-reload, or use the command without arguments to see an interactive selection menu.

Can I restrict Claude from using specific tools like Bash?

Yes, by using the 'tools' field in the style's YAML frontmatter (paired with the permission-modes plugin), you can whitelist exactly which tools Claude is allowed to execute.

Claude Output Styles

Name: Claude Output Styles
Author: constellos

byconstellos

•

Productivity & Workflow

Customizes Claude's persona, behavioral tone, and tool permissions through modular markdown configuration files.

Claude Output Styles provides a robust framework for fine-tuning the AI coding experience by defining specific behavioral profiles and tool constraints. By leveraging markdown files with optional YAML frontmatter, users can create specialized 'modes'—such as an Architect Delegator who focuses on high-level planning or a specialized tester—and switch between them instantly using the /output-style command. This skill is essential for developers who need to enforce project-specific coding standards, restrict tool access for security, or optimize Claude's efficiency for different stages of the development lifecycle.

Key Features

01Metadata Support: Use YAML frontmatter to define style names, descriptions, and permission schemas.

02Instant Hot-Reloading: Switch styles mid-session using the /output-style command without losing context.

03Custom Persona Injection: Define specific tones and behavioral instructions via markdown directives.

043 GitHub stars

05Hierarchical Management: Support for both global user-level styles and project-specific configurations.

06Tool Whitelisting: Restrict Claude's access to specific tools like Bash or Task to ensure safety and focus.

Use Cases

01Enforcing consistent documentation and style guidelines across a team via project-level markdown files.

02Creating an 'Architect' mode that prioritizes high-level planning and delegation to sub-agents.

03Implementing a 'Safety-First' profile that restricts destructive tools in production or sensitive environments.

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add constellos/claude-code-plugins claude-output-styles

For use in Claude.ai and ChatGPT

Download Skill