[FEAT-TOOLS][Added proper computer-use tools, fully working and tested] (#1692)

* Added computer-use tools

* Add PyTorch model creation and debugging example

This script provides an example of using an agent to create and debug a PyTorch model. It outlines the workflow for developing the model, including planning, creating, testing, fixing, and reporting.

* Add create_computer_use_tools to __all__ exports

* Refactor file handling functions to include workspace_root

* Update print statement from 'Hello' to 'Goodbye'

Author: IlumCI

examples/tools/computer-use/computer-use-example.py

@@ -0,0 +1,78 @@
+"""
+Agent Computer Use Example: Create and Debug a PyTorch Model
+
+This example demonstrates an agent using the computer_use toolkit to:
+1. Create a PyTorch model file
+2. Run tests to verify it works
+3. Debug and fix any issues over multiple loops
+
+The agent has access to all computer_use tools:
+- read_file, list_directory, grep_files (read-only)
+- write_file, edit_file, patch_file, delete_file (write)
+- run_command (shell)
+"""
+
+import os
+from swarms import Agent
+from swarms.tools.computer_use import create_computer_use_tools
+
+tools = create_computer_use_tools()
+
+
+agent = Agent(
+    agent_name="PyTorch-Debugger",
+    system_prompt="""
+You are an expert Python/PyTorch developer agent. Your task is to create and debug
+a PyTorch model file called `/$HOME/pytorch_model.py`.
+
+Follow this workflow in each loop:
+
+1. PLAN: Decide what needs to be done based on current state
+2. CREATE/MODIFY: Use write_file/edit_file to create or fix code
+3. TEST: Use run_command to run `python -c "from examples.models.pytorch_model import *; print('Import OK')"`
+4. FIX: If there are errors, use edit_file to fix them
+5. REPORT: Summarize what was done
+
+The model should:
+- Define a NeuralNetwork class with at least 3 layers
+- Include forward() method
+- Include a training step function
+- Include an evaluation function
+- Be well-documented with docstrings
+- Handle both CPU and GPU (cuda) devices
+- Have proper error handling
+
+When you see import errors or runtime errors:
+- Read the model file to understand the code
+- Identify the bug
+- Fix it with edit_file
+- Re-run the test
+
+Continue until the model imports and runs without errors.
+When fully working, create a simple test that demonstrates the model training.
+
+Return a final report with:
+- The final working code
+- Test results showing successful training
+- Any bugs that were fixed
+""",
+    model_name="gpt-4o",
+    max_loops=10,
+    tools=list(tools.values()),
+)
+
+result = agent.run("""
+Create and debug a PyTorch model at `/$HOME/pytorch_model.py`.
+
+Start by creating the initial model file, then test it.
+If there are errors, fix them. Keep iterating until the model works correctly.
+
+The model should demonstrate:
+1. A feedforward neural network
+2. Forward pass
+3. Training loop with loss computation
+4. Evaluation mode
+5. GPU/CPU device handling
+""")
+
+print(result)

swarms/tools/__init__.py

@@ -31,12 +31,14 @@
     multi_base_model_to_openai_function,
 )
 from swarms.tools.tool_registry import ToolStorage, tool_registry
+from swarms.tools.computer_use import create_computer_use_tools
 from swarms.tools.tool_utils import (
     scrape_tool_func_docs,
     tool_find_by_name,
 )
 
 __all__ = [
+    "create_computer_use_tools",
     "scrape_tool_func_docs",
     "tool_find_by_name",
     "_remove_a_key",