# Specialized Agent Profiles

Created: February 22, 2026

## Overview

Three specialized agent profiles have been configured for different types of tasks. Each agent has its own skills, system prompt, and best practices built-in.

## Available Agents

| Agent ID | Name | Emoji | Purpose |
|----------|------|-------|---------|
| `ios-dev` | iOS Developer | 📱 | iOS/Swift/SwiftUI development |
| `web-dev` | Web Developer | 🌐 | Next.js/React/TypeScript development |
| `research` | Research Assistant | 🔍 | Research, analysis, compilation |

---

## iOS Developer (`ios-dev`)

**Best for:** iOS app development, SwiftUI, Swift, Xcode projects

**Auto-loaded Skills:**
- ios-26-role
- swiftui-expert-skill
- swift-clean-architecture
- swift-model-design
- swiftui-mvvm
- ios-project-structure

**Default Patterns:**
- Modern Swift concurrency (async/await)
- SwiftUI preferred over UIKit
- MVVM with @Observable
- iOS 26+ target
- Proper error handling & accessibility
- Unit tests for business logic

**Usage:**
```bash
# Switch to iOS agent
openclaw agents switch ios-dev

# Or spawn subagent for iOS task
# (In code: sessions_spawn with agentId: "ios-dev")
```

---

## Web Developer (`web-dev`)

**Best for:** Next.js, React, TypeScript, web apps

**Auto-loaded Skills:**
- nextjs-expert
- frontend-design
- shadcn-ui
- ui-ux-pro-max
- firebase-basics
- firebase-auth-basics
- firebase-firestore-basics

**Default Patterns:**
- Next.js App Router
- TypeScript with strict typing
- Tailwind CSS for styling
- Responsive design
- Component composition
- Error boundaries & loading states
- Semantic HTML & accessibility
- **Supabase for backend/database (preferred)**
- **Shared Supabase pattern for all Vercel apps** (same project, prefixed tables)
- Supabase SSR auth patterns
- Server components for data fetching
- Proper RLS policies
- Service role key only in API routes

**Shared Supabase Pattern (IMPORTANT):**
All Vercel-deployed web apps share the same Supabase project (`qnatchrjlpehiijwtreh`):
- Copy `.env.local` from existing project (blog-backup or gantt-board)
- **Table naming:** MUST use `<projectName>_tableName` format
- Examples: `blog_messages`, `mission-control_tasks`, `heartbeat-status`
- This prevents table name collisions across apps
- Auth is shared (same users table across all apps)

**Usage:**
```bash
# Switch to web agent
openclaw agents switch web-dev

# Or spawn subagent for web task
# (In code: sessions_spawn with agentId: "web-dev")
```

---

## Research Assistant (`research`)

**Best for:** Research tasks, information gathering, documentation

**Auto-loaded Skills:**
- browser-automation
- session-logs

**Default Patterns:**
- Web search for current info
- Multiple source verification
- Proper citation with URLs
- Well-structured markdown output
- Attribution sections
- Effective summarization

**Usage:**
```bash
# Spawn subagent for research
# (In code: sessions_spawn with agentId: "research")
```

---

## How to Use

### 1. Interactive Mode

Switch which agent you're talking to:

```bash
openclaw agents switch ios-dev
openclaw agents switch web-dev
openclaw agents switch research
openclaw agents switch main    # Back to default
```

### 2. Subagent Tasks

When spawning tasks, specify the right agent:

```json
{
  "agentId": "ios-dev",
  "task": "Build a SwiftUI todo app with..."
}
```

```json
{
  "agentId": "web-dev", 
  "task": "Create a Next.js dashboard with..."
}
```

```json
{
  "agentId": "research",
  "task": "Research the latest React patterns..."
}
```

### 3. From Main Agent

When I'm (Max) handling your requests, I can spawn the right specialist:

**iOS Task:**
```
User: "Build me a habit tracker app"
→ I spawn ios-dev subagent
→ It follows iOS skills automatically
→ Returns properly architected SwiftUI code
```

**Web Task:**
```
User: "Create a landing page"
→ I spawn web-dev subagent  
→ It follows Next.js patterns from skills
→ Returns properly structured React components
```

**Research Task:**
```
User: "Find OpenClaw best practices"
→ I spawn research subagent
→ It searches, compiles, cites sources
→ Returns markdown document with attributions
```

---

## Agent Workspace Structure

Each agent has its own workspace folder with domain-specific guidance:

```
~/.openclaw/agents/<agentId>/
├── agent/
│   └── agent.json          # Agent configuration (model, skills, rules)
└── workspace/
    └── AGENTS.md           # Domain-specific workflow and patterns
```

### Key Files

| File | Purpose |
|------|---------|
| `agent/agent.json` | Configuration: model, system prompt, auto-loaded skills, rules |
| `workspace/AGENTS.md` | **Domain workflow**: patterns, file structure, common tasks, examples |

### Why AGENTS.md Matters

The `AGENTS.md` in each agent's workspace provides:
- **Standard workflow** for that domain
- **File organization** patterns
- **Code examples** and templates
- **Common tasks** with step-by-step guides
- **Tool-specific patterns** (Supabase, Xcode, etc.)

When a subagent spawns, it reads its `AGENTS.md` to understand:
- How to structure the work
- What patterns to follow
- What skills to check
- How to deliver results

---

## Benefits

1. **Specialized Knowledge** - Each agent loads relevant skills automatically
2. **Consistent Patterns** - iOS agent doesn't use web patterns, and vice versa
3. **Better Results** - Domain-specific expertise in each area
4. **Parallel Work** - Can spawn multiple specialized agents simultaneously

---

## Configuration Files

Agent configs stored at:
- `~/.openclaw/agents/ios-dev/agent/agent.json`
- `~/.openclaw/agents/web-dev/agent/agent.json`
- `~/.openclaw/agents/research/agent/agent.json`

Main config: `~/.openclaw/openclaw.json`

---

## Future Agents

Potential additional agents to create:
- `devops` - Infrastructure, Docker, CI/CD
- `data` - Python, data analysis, ML
- `design` - UI/UX, Figma, design systems
- `docs` - Technical writing, documentation

Just create a new `~/.openclaw/agents/<name>/agent/agent.json` and register in `openclaw.json`.