Skills: Intelligent Standards Integration

As of Agent OS 2.1, seamless integration with Claude Code Skills is enabled by default, offering a more context-efficient and intelligent way to apply your coding standards during development.

When enabled, your standards are converted into Claude Code Skills that Claude automatically applies when relevant, rather than being explicitly injected into every command and agent instruction.

This page covers:

• What are Claude Code Skills
• When Skills are used
• How Skills integration works
• Skills vs. injected standards
• Improving your Skills

What are Claude Code Skills

Claude Code Skills are specialized knowledge modules that Claude can automatically recognize and apply when working on tasks. Unlike explicitly referenced files or triggered subagents, Skills rely on Claude's intelligence to determine when they're relevant.

Each Skill contains:

• Content - The actual coding standards, patterns, or best practices
• Description - Metadata that helps Claude understand when this Skill is relevant
• Trigger conditions - Implicit cues that prompt Claude to activate the Skill

When Skills are used

Skills integration is enabled by default in Agent OS 2.1+ and applies when both of these configuration options are set to true:

• claude_code_commands: true - You're using Claude Code commands
• standards_as_claude_code_skills: true - Skills integration is enabled (default)

You can verify or modify these settings in your ~/agent-os/config.yml file:

defaults:
  profile: default
  claude_code_commands: true
  use_claude_code_subagents: true
  agent_os_commands: false
  standards_as_claude_code_skills: true  # Skills enabled

How Skills integration works

When standards_as_claude_code_skills is enabled, Agent OS changes how your standards are applied:

Standard behavior

Without Skills (traditional injection)

Normally, standards are explicitly referenced in commands and agent instructions using file references:

@agent-os/standards/global/naming.md
@agent-os/standards/backend/api-design.md
@agent-os/standards/frontend/component-patterns.md

This approach ensures standards are always available, but it can consume significant context tokens, especially when multiple standards are referenced at once.

Skills mode

With Skills enabled

When Skills are enabled:

1. Standards are not injected - References to standards files are removed from commands and agent instructions
2. Skills are compiled - Each standard file becomes a separate Claude Code Skill
3. Skills are installed - Compiled Skills are placed in your project's .claude/skills/ folder
4. Claude applies them intelligently - Claude automatically recognizes when to use each Skill based on the current task

Skills vs. injected standards

Understanding the tradeoffs helps you choose the right approach for your workflow:

How Skills are triggered

Unlike Claude Code subagents (which you explicitly trigger with instructions like "delegate to agent-name") or injected standards (which are specifically referenced), Claude Code Skills cannot be directly invoked.

Skills rely on Claude's natural intelligence to recognize when a particular Skill is relevant to the current task. This happens based on:

• The Skill's description and metadata
• The nature of the code being written
• The context of the current task
• Patterns Claude recognizes in your request

While this may seem like a limitation (and sometimes it can be), the benefit is significant: Skills are far more context and token efficient compared to reading lots of referenced standards files. Plus, only the most relevant Skill for the current task gets used, making them more targeted and effective.

Improving your Skills

By default, each Skill has a basic structure that's usable but not yet optimized. To maximize the effectiveness of your Claude Code Skills and help ensure they get triggered whenever and wherever they should, Agent OS provides the /improve-skills command.

What /improve-skills does

This command performs a deep analysis of each Skill and:

• Analyzes the Skill's content to understand its purpose
• Identifies key patterns, technologies, and use cases
• Rewrites and expands the Skill's description
• Improves "triggerability" by adding relevant metadata
• Optimizes each Skill for automatic recognition by Claude

When to run it

Run the improve-skills command:

• After initial installation - Transform basic Skills into optimized ones
• When you add new standards - Ensure new Skills are well-described
• When Skills aren't being used - Improve their descriptions to make them more recognizable
• After updating standards content - Align descriptions with updated content

/improve-skills

Learn more about Claude Code Skills

For comprehensive documentation about how Claude Code Skills work under the hood, including how to create custom Skills and advanced usage patterns:

Read the official Claude Code Skills documentation

Next steps

Learn about Standards

See configuration options

Learn about the 3-Layer Context system