Specialized Agents

Specialized Agents are persona-driven configurations that transform Claude Code from a general-purpose assistant into a dedicated team member with a specific role. While Skills provide the "how-to," Agents define the "who"—carrying out complex workflows, enforcing standards, and managing integrations.

The Persona-Driven Approach

Instead of providing generic instructions for every prompt, specialized agents use dedicated markdown files in .claude/agents/ to define their scope, authority, and checklists. This ensures that when Claude acts as a "Code Reviewer," it focuses on architecture and edge cases, rather than just syntax.

Key Specialized Agents

1. The Code Reviewer

Located at .claude/agents/code-reviewer.md, this agent is designed to validate changes against your specific project standards. It doesn't just look for bugs; it follows a strict checklist:

Strict Mode Compliance: Ensures TypeScript types are sound.
UX Consistency: Checks for loading states and error handling in UI components.
Pattern Matching: Validates that mutations and data fetching follow the project's GraphQL or API standards.

Usage: You can invoke the reviewer manually during a session, or let the GitHub Action trigger it automatically when a Pull Request is opened.

2. Workflow & Maintenance Agents

These agents live within your CI/CD pipeline, acting as "Scheduled Maintainers" to handle the repetitive tasks that usually fall through the cracks.

3. The Issue Integrator (`/ticket`)

By utilizing MCP (Model Context Protocol) servers, Claude Code connects directly to your project management tools like JIRA or Linear.

The /ticket agent can:

Read ticket requirements and acceptance criteria.
Implement the feature based on the codebase's existing skills.
Update the ticket status and link the resulting PR.

Intelligent Agent Activation

You don't always need to tell Claude which persona to adopt. The Skill Evaluation Engine (.claude/hooks/skill-eval.js) acts as a router. It analyzes your prompt and file context to suggest the right agent or skill automatically.

// Example of how the engine detects intent for agent activation
const triggers = {
  keywords: ['review', 'audit', 'check', 'pr'],
  keywordPatterns: [/fix.*issue/i, /update.*docs/i],
  priority: 10
};

Creating Your Own Agent

To create a new specialized agent, add a markdown file to the .claude/agents/ directory defining its mission:

Role Definition: Define who the agent is (e.g., "Security Auditor").
Context: Point to relevant .claude/skills/ it should always have active.
Checklist: Provide a set of mandatory validation steps.
Action Rules: Define what the agent is allowed to do (e.g., "Always run npm test before finishing").

Once defined, you can point Claude to this agent file to initialize a specialized session.

Specialized Agents