Skip to content

agents.md

Latest commit

 

History

History
211 lines (158 loc) · 5.36 KB

agents.md

File metadata and controls

211 lines (158 loc) · 5.36 KB

Agent Management System

Overview

The Agent Management System allows you to assign specialized AI agents to specific tasks, leveraging different models and configurations for optimal results.

What are Agents?

Agents are specialized AI configurations designed to handle specific types of tasks. Each agent can have:

  • Unique Identity: Name, role, and description
  • Specialized Instructions: Task-specific prompts and guidelines
  • Visual Identification: Color coding and avatars
  • Model Preferences: Optimized for different AI models

Key Features

Agent Assignment

  • Assign agents to individual tasks
  • Bulk assignment for multiple tasks
  • Automatic agent selection based on task type

Agent Templates

  • Pre-configured agent templates for common roles
  • Custom agent creation
  • Template sharing and export

Agent Specializations

Agent Type Best For Example Tasks
Frontend Developer UI/UX implementation React components, CSS styling
Backend Engineer Server-side logic API endpoints, database queries
DevOps Specialist Infrastructure CI/CD, deployment, monitoring
Data Scientist Data analysis ML models, data processing
Security Expert Security reviews Vulnerability assessment, auth
QA Engineer Testing Test cases, bug verification
Technical Writer Documentation API docs, user guides

Using Agents

Basic Usage

  1. Assign an agent to a task:

    "assign frontend developer to task 1"

  2. Create custom agent:

    "create agent: performance optimizer with focus on speed"

  3. View assigned agents:

    "show agents for current tasks"

Advanced Features

Agent Instructions

Each agent can have specialized instructions that guide their approach:

{
  "name": "React Specialist",
  "instructions": "Focus on React best practices, hooks, and performance optimization",
  "preferredModel": "claude-3-opus",
  "color": "#61DAFB"
}

Agent Collaboration

Multiple agents can work on related tasks:

  1. Frontend Developer - Creates UI components
  2. Backend Engineer - Implements API
  3. QA Engineer - Writes tests

Agent Management in Task Viewer

The Task Viewer provides a visual interface for agent management:

  1. Agent List View: See all available agents
  2. Drag & Drop Assignment: Assign agents to tasks visually
  3. Agent Editor: Create and modify agents
  4. Bulk Operations: Assign agents to multiple tasks

Agent List View

Best Practices

When to Use Agents

✅ Good Use Cases:

  • Complex projects requiring specialized expertise
  • Tasks that benefit from different AI models
  • Maintaining consistent coding styles
  • Role-based task distribution

❌ When Not Needed:

  • Simple, straightforward tasks
  • Quick fixes or minor changes
  • Exploratory research

Agent Selection Guidelines

  1. Match expertise to task:

    • UI tasks → Frontend Developer
    • API tasks → Backend Engineer
    • Testing → QA Engineer

  2. Consider model strengths:

    • Complex reasoning → Advanced models
    • Code generation → Specialized coding models
    • Documentation → Language-focused models

  3. Use templates:

    • Start with pre-configured templates
    • Customize based on project needs
    • Save successful configurations

Configuration

Creating Custom Agents

{
  "agents": [
    {
      "id": "custom-001",
      "name": "Performance Optimizer",
      "role": "Performance Engineer",
      "description": "Specializes in optimization and performance tuning",
      "instructions": "Focus on performance metrics, optimization techniques, and benchmarking",
      "color": "#FF6B6B",
      "avatar": "⚡",
      "preferredModel": "claude-3-opus"
    }
  ]
}

Agent Templates

Located in templates/agents/:

  • frontend.json - Frontend development
  • backend.json - Backend development
  • devops.json - DevOps and infrastructure
  • qa.json - Quality assurance
  • security.json - Security analysis

Integration with Claude

When using Claude Desktop with agents:

  1. Agents provide context-specific instructions
  2. Claude adapts approach based on agent role
  3. Maintains consistency across tasks
  4. Optimizes for assigned agent's expertise

API Reference

Agent Structure

interface Agent {
  id: string;
  name: string;
  role: string;
  description?: string;
  instructions?: string;
  color?: string;
  avatar?: string;
  preferredModel?: string;
  metadata?: Record<string, any>;
}

Agent Assignment

interface TaskAgent {
  taskId: string;
  agentId: string;
  assignedAt: Date;
  assignedBy?: string;
}

Troubleshooting

Common Issues

Issue Solution
Agent not applying Verify agent ID and task ID
Instructions ignored Check instruction format and length
Model mismatch Ensure model is available in your client
Visual not showing Update Task Viewer to latest version

Tips

  1. Keep agent instructions concise and focused
  2. Use consistent naming conventions
  3. Document custom agents for team use
  4. Regularly review and update agent configurations
  5. Test agents on sample tasks before production use