fatby/create-agent

Create Agent

Description

Create a new specialized AI agent with customized capabilities, prompts, and tool access. This skill helps you define domain-specific agents that can be reused for particular tasks.

Instructions

1. Understand the Requirements

   • Analyze the user's description of what the agent should do
   • Identify the key capabilities needed
   • Determine which tools would be most useful

2. Design the Agent Configuration

   • Create a descriptive name that reflects the agent's purpose
   • Write a clear system prompt that defines the agent's role and constraints
   • Select appropriate tools from the available tool registry

3. Create the Agent Definition File

   • Create a new directory: .duncrew/agents/{agent_name}/
   • Write config.json with:

     {
       "name": "agent-name",
       "description": "Agent description",
       "system_prompt": "Custom system prompt...",
       "tools": ["tool1", "tool2"],
       "temperature": 0.7,
       "max_tokens": 4096
     }
     

4. Add Custom Instructions (Optional)

   • Create INSTRUCTIONS.md with detailed guidance
   • Add example interactions
   • Define success criteria

5. Verify the Agent

   • Test the agent with a sample task
   • Verify tool access works correctly
   • Adjust prompts if needed

Examples

Create a Code Review Agent:

Agent Name: code-reviewer
Purpose: Review code for bugs, security issues, and best practices
Tools: search_codebase, search_symbol, read_file

Create a Documentation Agent:

Agent Name: doc-writer
Purpose: Generate and maintain project documentation
Tools: search_codebase, read_file, write_file

Notes

• Keep the system prompt focused and specific
• Don't give agents more tools than they need
• Test with edge cases before deploying
• Consider adding constraints for safety