Python code generated by Agent results in an error during the first pass

[msmmpts]

Hi team,

I was making use of langchain agents along with PythonREPLTool to execute python code.

Here are the logs:

To generate a plot for the function ( y = x^2 ) over the range from -1 to 1, I will create a set of x values within this range, compute the corresponding y values, and then use matplotlib to visualize the function. After plotting, I will analyze the graph for insights.

Action: Code Interpreter with Plot Display
Action Input:

import numpy as np
import matplotlib.pyplot as plt

# Define the range of x values
x = np.linspace(-1, 1, 100)
# Calculate y values
y = x**2

# Create the plot
plt.figure(figsize=(8, 6))
plt.plot(x, y, label='y = x^2', color='blue')
plt.title('Plot of y = x^2')
plt.xlabel('x')
plt.ylabel('y')
plt.axhline(0, color='black',linewidth=0.5, ls='--')
plt.axvline(0, color='black',linewidth=0.5, ls='--')
plt.grid(color = 'gray', linestyle = '--', linewidth = 0.5)
plt.legend()
plt.xlim(-1.5, 1.5)
plt.ylim(-0.5, 1.5)
plt.show()

Observation: Error: invalid syntax (, line 1)

During the second pass, agent corrects the code and then runs it properly.

i.e for every code generated (starts with """```python""") generates code with syntax error and might take several iterations for it to execute and run code resulting in additional latency and tokens.

Can anyone suggest alternatives for fixing this?

Note: I have already included in the description of the agent not to start with keywords like ```python etc but still it generates python code with syntax issues.

Model tried : gpt-4o, gpt-4o-mini

Library versions installed

langchain 0.3.14
langchain-aws 0.2.10
langchain-community 0.3.14
langchain-core 0.3.29
langchain-experimental 0.3.4
langchain-google-genai 2.0.8
langchain-openai 0.3.0
langchain-text-splitters 0.3.5

Code that was used:

`
import matplotlib.pyplot as plt
from langchain.agents import initialize_agent, Tool
from langchain_experimental.tools.python.tool import PythonREPLTool
from langchain_openai import AzureChatOpenAI

class LocalPythonREPLTool(PythonREPLTool):
def run(self, code: str) -> str:
try:
# Create a secure execution environment
exec_globals = {"plt": plt}
exec_locals = {}
exec(code, exec_globals, exec_locals)

        # Check if a plot is generated
        if "plt" in exec_globals:
            plt.show()  # Display the plot locally
            return "Plot displayed successfully."
        return "Code executed successfully, but no plot was generated."
    except Exception as e:
        return f"Error: {str(e)}"

Initialize the Tool

code_tool = LocalPythonREPLTool()

tools = [
Tool(
name="Code Interpreter with Plot Display",
func=code_tool.run,
description=(
"A Python code interpreter capable of executing Python code locally. "
"It can generate and display plots using matplotlib or other libraries. "
"Input should be valid Python code. The python code should not start with keywords like ```python"
),
)
]

llm = AzureChatOpenAI()

Initialize the Agent

agent = initialize_agent(
tools=tools,
llm=llm,
agent="zero-shot-react-description",
verbose=True
)

Test the Agent

if name == "main":
query = (
"Generate a plot for y = x^2 for x in the range -1 to 1 using matplotlib and report insights from the plot"
)
response = agent.run(query)
print(response)

`