Agents

Agents are customized AI assistants with specific roles, permissions, and behaviors. Create specialized agents for different tasks like code review, documentation, or testing.

Understanding Agents

An agent in Claude Forge consists of:

Property	Description
Name	A friendly identifier for the agent
Role	System prompt that defines the agent's behavior
Permissions	What actions the agent can perform

Default Agent

Claude Forge comes with a Default agent that:

• Has standard Claude Code CLI capabilities
• Can read and write files
• Can execute shell commands
• Has no specialized behavior

Creating an Agent

1. In the sidebar, scroll to the Agents section
2. Click + New Agent
3. Fill in the agent details:

Name

Choose a descriptive name like:

• "Code Reviewer"
• "Documentation Writer"
• "Test Generator"
• "Bug Hunter"

Role (System Prompt)

The role defines how the agent behaves. Write it as instructions:

You are a code review specialist. Your job is to:
- Find potential bugs and security issues
- Suggest performance improvements
- Ensure code follows best practices
- Be constructive and educational in feedback

Always explain WHY something should be changed, not just what.

Permissions

Set what the agent can do:

Permission	Description
Read	Can read files in the project
Write	Can create and modify files
Execute	Can run shell commands

WARNING

Be cautious with Execute permissions. Only enable for trusted agents.

Example Agents

Code Reviewer

Name: Code Reviewer
Role: You are a senior developer performing code reviews. Focus on:
- Code quality and readability
- Potential bugs and edge cases
- Security vulnerabilities
- Performance considerations
Always be constructive and explain your reasoning.
Permissions: Read only

Documentation Writer

Name: Doc Writer
Role: You are a technical documentation specialist. Your tasks:
- Write clear, concise documentation
- Add JSDoc/TSDoc comments to functions
- Create README files
- Document APIs and interfaces
Use consistent formatting and include examples.
Permissions: Read, Write

Test Generator

Name: Test Generator
Role: You are a QA engineer specializing in automated testing. You:
- Write unit tests for functions
- Create integration tests
- Ensure good test coverage
- Use appropriate testing patterns
Follow the existing test structure in the project.
Permissions: Read, Write

Bug Hunter

Name: Bug Hunter
Role: You are a security researcher and bug hunter. Focus on:
- Finding security vulnerabilities
- Identifying race conditions
- Detecting memory leaks
- Spotting logic errors
Report findings with severity levels and remediation steps.
Permissions: Read only

Using Agents

Selecting an Agent

1. Click on an agent in the Agents sidebar section
2. The selected agent becomes active
3. All subsequent prompts use this agent

Switching Agents

Simply click a different agent to switch. Each agent:

• Maintains its own behavior
• Shares the same project context
• Has independent conversation history

Editing Agents

1. Click on the agent you want to edit
2. Modify the name, role, or permissions
3. Changes are saved automatically

Deleting Agents

1. Select the agent
2. Click the Delete button (or right-click and select Delete)
3. Confirm the deletion

INFO

The Default agent cannot be deleted.

Agent Best Practices

Be Specific in Roles

Instead of:

You help with code.

Write:

You are a TypeScript expert who helps refactor JavaScript code to TypeScript.
Focus on:
- Adding proper type annotations
- Using interfaces for complex objects
- Leveraging TypeScript features like generics
- Maintaining backward compatibility

Match Permissions to Purpose

Agent Type	Recommended Permissions
Reviewer/Analyzer	Read only
Writer/Generator	Read, Write
Automation scripts	Read, Write, Execute

Create Task-Specific Agents

Rather than one generic agent, create several focused ones:

• Frontend Agent - Specializes in React, CSS, accessibility
• Backend Agent - Focuses on APIs, databases, security
• DevOps Agent - Handles CI/CD, Docker, deployment

Troubleshooting

Agent not following instructions

Solutions:

1. Make the role more explicit
2. Add examples of desired output
3. Use bullet points for clarity
4. Specify what NOT to do

Unexpected behavior

Solutions:

1. Check permissions are correctly set
2. Verify the role doesn't conflict with base capabilities
3. Try rephrasing the role instructions