Skip to content

AI Agents - Specialized Experts

Sylphx Flow includes production-ready AI agents, each an expert in their domain. Instead of a single generic AI that tries to do everything, Flow provides specialized agents optimized for specific workflows.

Available Agents

🔨 Coder

Primary Purpose: Feature implementation and bug fixes

Best For:

  • Building new features from scratch
  • Implementing user stories
  • Fixing bugs and issues
  • Refactoring existing code
  • Writing production code

Expertise:

  • Full-stack development
  • Multiple languages and frameworks
  • Design patterns and best practices
  • Performance optimization
  • Clean code principles

Example Usage:

bash
# Default agent - no flag needed
sylphx-flow "implement user authentication"

# Explicit coder
sylphx-flow "add password reset feature" --agent coder

# Bug fixes
sylphx-flow "fix memory leak in data processing" --agent coder

🔍 Reviewer

Primary Purpose: Code review and security analysis

Best For:

  • Pre-merge code reviews
  • Security audits
  • Performance analysis
  • Code quality checks
  • Architecture reviews

Expertise:

  • Security vulnerabilities (OWASP Top 10)
  • Code smells and anti-patterns
  • Performance bottlenecks
  • Best practice violations
  • Accessibility issues

Example Usage:

bash
# Review current changes
sylphx-flow "review my latest changes" --agent reviewer

# Security audit
sylphx-flow "check for security vulnerabilities" --agent reviewer

# Performance review
sylphx-flow "analyze performance bottlenecks" --agent reviewer

What Reviewer Checks:

  • ✅ Security vulnerabilities
  • ✅ Code quality issues
  • ✅ Performance problems
  • ✅ Best practice violations
  • ✅ Test coverage gaps
  • ✅ Documentation completeness

✍️ Writer

Primary Purpose: Documentation and technical writing

Best For:

  • API documentation
  • README files
  • Code comments
  • Technical guides
  • Architecture documentation

Expertise:

  • Clear, concise technical writing
  • API reference documentation
  • Tutorial creation
  • README optimization
  • Inline code documentation

Example Usage:

bash
# Generate API docs
sylphx-flow "document all API endpoints" --agent writer

# Update README
sylphx-flow "improve README with usage examples" --agent writer

# Add code comments
sylphx-flow "add JSDoc comments to utils" --agent writer

Documentation Standards:

  • Clear and concise language
  • Code examples for all features
  • Proper formatting (Markdown, JSDoc, etc.)
  • Beginner-friendly explanations
  • Complete API references

🎯 Orchestrator

Primary Purpose: Complex multi-step tasks and coordination

Best For:

  • Large-scale refactoring
  • Architecture changes
  • Multi-component features
  • System design
  • Complex migrations

Expertise:

  • Breaking down complex tasks
  • Coordinating multiple changes
  • System architecture
  • Dependency management
  • Risk assessment

Example Usage:

bash
# Architecture changes
sylphx-flow "migrate from REST to GraphQL" --agent orchestrator

# Large refactoring
sylphx-flow "refactor authentication system" --agent orchestrator

# Complex features
sylphx-flow "implement real-time notifications across app" --agent orchestrator

Orchestrator's Approach:

  1. Analyze - Understand the full scope
  2. Plan - Break down into steps
  3. Execute - Implement systematically
  4. Verify - Test each component
  5. Integrate - Ensure everything works together

Choosing the Right Agent

Task TypeRecommended AgentWhy
New featureCoderOptimized for implementation
Bug fixCoderFast, focused fixes
Code reviewReviewerSecurity and quality expertise
Pre-merge checkReviewerCatches issues before merge
DocumentationWriterClear technical writing
README updateWriterUser-focused content
Architecture changeOrchestratorHandles complexity
System refactorOrchestratorCoordinates multiple changes

Agent Configuration

Default Agent

The default agent is Coder - optimized for most common tasks:

bash
# These are equivalent
sylphx-flow "implement login"
sylphx-flow "implement login" --agent coder

Setting Default Agent

Flow remembers your agent preference:

bash
# Use reviewer once
sylphx-flow "check security" --agent reviewer

# Next time, Flow remembers (if you want)
# Or specify agent each time for clarity

Configuration File

Your agent preferences are stored in .sylphx-flow/settings.json:

json
{
  "defaultAgent": "coder",
  "lastUsedAgent": "reviewer"
}

How Agents Work

1. Specialized Prompting

Each agent has a unique system prompt optimized for its role:

  • Coder: "You are an expert software engineer..."
  • Reviewer: "You are a senior code reviewer..."
  • Writer: "You are a technical documentation expert..."
  • Orchestrator: "You are a software architect..."

2. Domain-Specific Rules

Each agent follows different rules:

typescript
// Coder rules
- Focus on clean, working code
- Write comprehensive tests
- Optimize for performance

// Reviewer rules
- Be thorough and critical
- Check security first
- Suggest improvements

// Writer rules
- Be clear and concise
- Provide examples
- Assume beginner level

// Orchestrator rules
- Think systemically
- Plan before acting
- Consider all dependencies

3. Tool Access

All agents have access to the same tools (MCP servers, codebase search, etc.), but use them differently:

  • Coder: Uses codebase search to find implementation patterns
  • Reviewer: Uses codebase search to find similar code to review
  • Writer: Uses codebase search to understand what to document
  • Orchestrator: Uses codebase search to understand system architecture

Advanced Usage

Chaining Agents

Use different agents in sequence for complex workflows:

bash
# 1. Implement feature
sylphx-flow "implement OAuth login" --agent coder

# 2. Review implementation
sylphx-flow "review OAuth implementation" --agent reviewer

# 3. Document it
sylphx-flow "document OAuth flow" --agent writer

Loop Mode with Agents

Combine agents with loop mode:

bash
# Continuous code review
sylphx-flow "review all PRs" --agent reviewer --loop --target claude-code

# Continuous documentation updates
sylphx-flow "keep docs in sync with code" --agent writer --loop --target claude-code

Best Practices

✅ Do

  • Use Coder for 80% of tasks (it's the default for a reason)
  • Use Reviewer before merging important changes
  • Use Writer for all documentation tasks
  • Use Orchestrator for complex, multi-step work

❌ Don't

  • Don't use Orchestrator for simple tasks (overkill)
  • Don't use Coder for reviews (Reviewer is specialized)
  • Don't use Writer for code (Coder is better)
  • Don't mix agents in the same command (use sequential commands instead)

Customizing Agents

Adding Custom Agents

You can define custom agents in .sylphx-flow/agents/:

typescript
// .sylphx-flow/agents/custom-agent.ts
export default {
  name: "Database Expert",
  description: "Specialized in database optimization",
  systemPrompt: "You are a database expert...",
  rules: ["Always use indexes", "Optimize queries"]
}

Then use it:

bash
sylphx-flow "optimize database queries" --agent database-expert

Syncing Agent Updates

Keep your agents up-to-date with Flow's latest improvements:

bash
# Sync all agents
sylphx-flow --sync

# Sync specific platform
sylphx-flow --sync --target claude-code

Troubleshooting

Agent Not Found

bash
Error: Agent 'xyz' not found

Solution:

  1. Check spelling: --agent coder (not --agent code)
  2. List available agents: sylphx-flow --list-agents
  3. Use default agent (omit --agent flag)

Wrong Agent Behavior

If an agent isn't performing as expected:

  1. Verify agent: sylphx-flow --list-agents
  2. Check task match: Make sure task fits agent's specialty
  3. Try different agent: Sometimes Orchestrator works better than Coder
  4. Check rules: Ensure agent rules haven't been modified

Learn More

Released under the MIT License.