tool_permission_callback

Tool Permission Callback Example

This example demonstrates how to use tool permission callbacks to control which tools Claude can use and modify their inputs for safety.

Overview

The CanUseTool callback allows you to:

• Allow/deny tools based on tool type and input parameters
• Modify tool inputs to redirect operations to safe locations
• Log tool usage for auditing and debugging
• Block dangerous operations like system directory modifications

Usage

go run main.go

The example will:

1. Connect to Claude with a custom permission callback
2. Send a query asking Claude to list files, create a program, and read it
3. Intercept each tool use request and apply permission rules
4. Display all tool usage with decisions made

Permission Rules Implemented

Read Operations (Always Allow)

• Read, Glob, Grep - Automatically allowed as read-only operations

Write Operations (Conditional)

• System directories (/etc/, /usr/) - Denied
• Unsafe paths - Redirected to ./safe_output/ directory
• Safe paths (/tmp/, ./) - Allowed

Bash Commands (Filtered)

• Dangerous patterns - Denied:
  • rm -rf
  • sudo
  • chmod 777
  • dd if=
  • mkfs
• Safe commands - Allowed and logged

Unknown Tools

• Denied by default (in production, you might prompt the user)

Example Output

🔧 Tool Permission Request: Glob
   Input: {
     "pattern": "*"
   }
   ✅ Automatically allowing Glob (read-only operation)

🔧 Tool Permission Request: Write
   Input: {
     "file_path": "/etc/config.txt",
     "content": "..."
   }
   ❌ Denying write to system directory: /etc/config.txt

🔧 Tool Permission Request: Write
   Input: {
     "file_path": "hello.go",
     "content": "..."
   }
   ⚠️  Redirecting write from hello.go to ./safe_output/hello.go

🔧 Tool Permission Request: Bash
   Input: {
     "command": "rm -rf /"
   }
   ❌ Denying dangerous command: rm -rf /

Customization

Modify the myPermissionCallback function to implement your own permission logic:

func myPermissionCallback(
    ctx context.Context,
    toolName string,
    inputData map[string]any,
    permCtx *claudesdk.ToolPermissionContext,
) (claudesdk.PermissionResult, error) {
    // Your custom logic here

    // Allow with modifications
    return &claudesdk.PermissionResultAllow{
        Behavior:     "allow",
        UpdatedInput: modifiedInput,
    }, nil

    // Or deny
    return &claudesdk.PermissionResultDeny{
        Behavior: "deny",
        Message:  "Reason for denial",
    }, nil
}

Security Considerations

• Always validate file paths to prevent directory traversal attacks
• Use allowlists rather than denylists when possible
• Log all permission decisions for audit trails
• Consider the full input parameters, not just simple pattern matching
• Test your permission rules thoroughly

Related Examples

• examples/hooks/ - Shows PreToolUse hooks for logging
• examples/streaming_client/ - Demonstrates basic client usage