agents

Claude Code Agents

A collection of specialized AI agents for Claude Code CLI. Each agent is an expert in a specific framework, language, or development task.

Getting Started

1. Install the agents (see Setup below) so they live at ~/.claude/agents.

2. Configure your project once. From your project root, run the team-configurator to detect the stack and write an agent assignment table into your project's CLAUDE.md:

   claude "Use team-configurator to set up my AI development team"

   If you use the bundled .zshrc, the claudeInit function runs this exact command.

3. Work normally. Claude auto-delegates to the right specialists based on the assignment table. Use the /scaffold or /grind skills when you want the tech-lead-orchestrator to drive a feature end-to-end.

Setup

If ~/.claude/agents already exists as a real directory, move it aside first (mv ~/.claude/agents ~/.claude/agents.bak) - otherwise the symlink nests inside it instead of replacing it.

From the repo root:

# Copy
cp -r claude/agents ~/.claude/agents

# Or symlink to keep it linked to the repo
ln -sfn "$(pwd)/claude/agents" ~/.claude/agents

Usage

Ask for an agent in plain English and Claude routes the work to it:

claude "use the tech-lead-orchestrator to plan a user auth system"
claude "use the code-reviewer to review my changes"
claude "have the react-nextjs-expert add a server component for product listing"

Tip: in an interactive session you can also @-mention an agent directly (e.g. @agent-code-reviewer). A mention that doesn't resolve is dropped silently with no error, so plain English is the safer habit and works everywhere.

Naming Conventions

Agent names follow a consistent suffix pattern:

Suffix	Role	Examples
expert	Domain specialist - implements and advises within a specific technology	typescript-expert, react-nextjs-expert, aws-expert
architect	Design-only - produces specs and plans, never implements	api-architect, scraper-architect
orchestrator / analyst / configurator	Coordination - plans tasks, routes to specialists, never implements directly	tech-lead-orchestrator, project-analyst, team-configurator
Unique names	Earned exceptions where the name better describes the role	code-reviewer, code-archaeologist, performance-optimizer

Coding Philosophy

Code produced by these agents in your projects should follow these principles. The team-configurator writes them into each project's CLAUDE.md when it sets up the AI team, so every session and specialist working in that project inherits them:

• DRY - Don't repeat yourself. Extract reusable utilities instead of copy-pasting. But don't abstract prematurely - wait until a pattern is clear before extracting.
• Functional over imperative - Prefer pure functions over side effects. Use map/filter/reduce over imperative loops. Compose small functions over writing monolithic ones.
• Immutable by default - Use const, readonly, and immutable data structures. Mutate only when there's a clear performance reason.
• Composable and reusable - Design functions and components for reuse from the start. Small, focused units that compose together.
• Scalable patterns - Write code that works for 10 items and 10,000 items. Consider data growth, not just current state.

Agents Included

Core Specialists

Meta-development activities that apply across any project:

• code-reviewer - Code review and security auditing
• performance-optimizer - Performance profiling and optimization
• playwright-expert - E2E testing and browser automation
• scraper-architect - Web scraping architecture and data pipelines
• documentation-expert - READMEs, API docs, architecture guides
• code-archaeologist - Codebase exploration and documentation
• skill-expert - Create, review, and update Claude Code skills
• agent-expert - Create, review, and update Claude Code agent definitions

Orchestrators (3)

High-level planning and coordination:

• project-analyst - Detect tech stacks and route to specialists
• team-configurator - Auto-configure AI team for your project
• tech-lead-orchestrator - Plan and coordinate multi-step tasks

Framework Specialists

• Python: python-expert, testing-expert, performance-expert, python-security-expert, web-scraping-expert, devops-cicd-expert
• React: react-component-expert, react-nextjs-expert, material-ui-expert, jest-react-testing-expert
• React Native: react-native-expert, expo-expert
• Node.js: express-tsoa-expert, fastify-expert, vitest-expert
• Database: prisma-database-expert, mysql-prisma-expert, postgresql-prisma-expert
• Firebase: firebase-expert
• Godot: gdscript-expert, godot-scene-architect, godot-performance-expert

Universal Specialists

Cross-framework tools:

• api-architect - REST/GraphQL design
• openapi-contract-expert - Contract-first API development and client generation
• auth-integration-expert - Authentication/authorization (NextAuth, Auth.js, OAuth)
• backend-expert - Framework-agnostic backend
• frontend-expert - Framework-agnostic frontend
• tailwind-css-expert - Tailwind CSS styling
• typescript-expert - TypeScript patterns and best practices
• aws-expert - AWS services integration
• node-security-expert - Node.js security (JWT/jose, bcryptjs, OWASP, rate limiting)

Adding a New Agent

Use the tech-lead-orchestrator to add agents that follow the established patterns:

claude "use the tech-lead-orchestrator to add a new 'svelte-expert' agent to ~/.claude/agents/

The agent should be:
- Category: Framework specialist (specialized/svelte/)
- Purpose: Svelte and SvelteKit development expert

Tasks needed:
1. Analyze existing framework specialist patterns (react, python)
2. Create svelte-expert.md following the standard format
3. Update orchestrators to know about the new agent
4. Review the integration"