Custom executors for built-in tools

# Proposal: Customizable Built-in Tools Architecture

## Problem

Claude Code's built-in tools (read, write, etc.) are tightly coupled to the local VM, creating limitations:
- Unnecessary VM requirement for read-only operations
- Security risk: API tokens and proprietary code must be on the same VM as third-party code
- Complexity: There are many services like E2B, codesandbox with a simple SDK. Instead of running a server on the VM

## Proposed Solution

Allow passing custom execution function to the built-in tools.

## Use Cases

1. **GitHub Read**: Read open source projects directly from GitHub without VM provisioning or repository cloning
2. **Rork Project Read**: Access Rork public projects without VM infrastructure
3. **Enhanced Security**: Keep agent code, proprietary tools, and sensitive configurations separate from VMs running third-party code