Skip to content

This repository implements an LLM agent that receives a user's prompt and corresponding unit tests. The agent generates the code, tests it for correct execution, and ensures it passes the unit tests. If the code fails to execute or fails any unit test, the agent regenerates the code until it succeeds.

Notifications You must be signed in to change notification settings

paulomuraroferreira/Self_correcting_coding_agent

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

11 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

README

This repository implements a LLM agent that receives a user's prompt and corresponding unit tests. The agent generates the code, tests it for correct execution, and ensures it passes the unit tests. If the code fails to execute or fails any unit test, the agent regenerates the code until it succeeds.

Alt text

This project builds upon the code from LangChain's Notebook (https://github.com/mistralai/cookbook/blob/main/third_party/langchain/langgraph_code_assistant_mistral.ipynb), where an agent generates code based on the user's prompt and checks if the execution works.

The main contribution of this project is the integration of unit tests into the code execution process.

Additionally, the original Jupyter notebook code has been refactored into a structured Python project suitable for deployment. By adhering to best coding practices, the code has been modularized, reusable functions have been created, and proper documentation and testing have been ensured. The result is a well-organized .py project that is maintainable and ready for production.

Setup Instructions

  1. Clone the repository:
git clone https://github.com/paulomuraroferreira/Self_correcting_coding_agent.git
cd Self_correcting_coding_agent
  1. Install Dependencies:
$pip install -e .
  1. Environment Configuration:

Create a .env file and fill the following environment variables:

OPENAI_API_KEY=your_openai_api_key
OPENAI_CHAT_MODEL="gpt-4o-2024-08-06"
  1. Run the Application:

Input the code's prompt in the main.py file. The prompt should be descriptive and specify the name of the classes.

For example:

QUESTION3 = '''
Python Class Description
Class Name: BankAccount

Description:
The BankAccount class represents a user's bank account, allowing for deposits, withdrawals, and viewing the transaction history. The class ensures that withdrawals cannot exceed the current balance and that deposits and withdrawals are properly recorded in the transaction history.

Attributes:

balance (float): The current balance of the account, initialized to 0.
transactions (list): A list to store the history of transactions. Each transaction is stored as a dictionary with keys type (either 'deposit' or 'withdrawal'), amount, and date.
Methods:

deposit(amount: float) -> None: Adds the specified amount to the balance and records the transaction.
withdraw(amount: float) -> bool: Attempts to subtract the specified amount from the balance. Returns True if successful, otherwise returns False. Records the transaction if successful.
get_balance() -> float: Returns the current balance.
get_transaction_history() -> list: Returns a list of all transactions.
'''

main = Main(QUESTION3)
  1. Add the unit test code in a .py file in the the src/unit_test_folder. Also, import the corresponding class from the code_solution.py file. For example, for the above prompt, the LLM will generate a class called BankAccount. Therefore, the src/unit_test_folder/test_unit_test.py should include:

    from src.code_solution import BankAccount

  2. Run the main.py file.

    python main.py

Example

For the example above, the code will display the following output:

================================ Human Message =================================


Python Class Description
Class Name: BankAccount

Description:
The BankAccount class represents a user's bank account, allowing for deposits, withdrawals, and viewing the transaction history. The class ensures that withdrawals cannot exceed the current balance and that deposits and withdrawals are properly recorded in the transaction history.

Attributes:

balance (float): The current balance of the account, initialized to 0.
transactions (list): A list to store the history of transactions. Each transaction is stored as a dictionary with keys type (either 'deposit' or 'withdrawal'), amount, and date.
Methods:

deposit(amount: float) -> None: Adds the specified amount to the balance and records the transaction.
withdraw(amount: float) -> bool: Attempts to subtract the specified amount from the balance. Returns True if successful, otherwise returns False. Records the transaction if successful.
get_balance() -> float: Returns the current balance.
get_transaction_history() -> list: Returns a list of all transactions.

2024-08-18 14:51:24,955 - INFO - agent.py - ---GENERATING CODE SOLUTION---
================================== Ai Message ==================================

Here is my attempt to solve the problem: The BankAccount class is designed to manage a user's bank account by allowing deposits, withdrawals, and tracking transaction history. It ensures that withdrawals do not exceed the available balance and records each transaction with details such as type, amount, and date. 
Imports: from datetime import datetime 
Code: class BankAccount:
    def __init__(self):
        self.balance = 0.0
        self.transactions = []

    def deposit(self, amount: float) -> None:
        self.balance += amount
        transaction = {
            'type': 'deposit',
            'amount': amount,
            'date': datetime.now()
        }
        self.transactions.append(transaction)

    def withdraw(self, amount: float) -> bool:
        if amount > self.balance:
            return False
        self.balance -= amount
        transaction = {
            'type': 'withdrawal',
            'amount': amount,
            'date': datetime.now()
        }
        self.transactions.append(transaction)
        return True

    def get_balance(self) -> float:
        return self.balance

    def get_transaction_history(self) -> list:
        return self.transactions
2024-08-18 14:51:28,979 - INFO - agent.py - ---CHECKING CODE---
2024-08-18 14:51:29,170 - INFO - agent.py - 
============================= test session starts ==============================
platform linux -- Python 3.12.3, pytest-8.3.2, pluggy-1.5.0 -- /home/paulo/Python_projects/Self_correcting_coding_agent/venv/bin/python
cachedir: .pytest_cache
rootdir: /home/paulo/Python_projects/Self_correcting_coding_agent
configfile: pyproject.toml
plugins: anyio-4.4.0
collecting ... collected 7 items

src/unit_test_folder/test_unit_test.py::test_initial_balance PASSED      [ 14%]
src/unit_test_folder/test_unit_test.py::test_deposit PASSED              [ 28%]
src/unit_test_folder/test_unit_test.py::test_withdrawal_successful PASSED [ 42%]
src/unit_test_folder/test_unit_test.py::test_withdrawal_insufficient_funds PASSED [ 57%]
src/unit_test_folder/test_unit_test.py::test_transaction_history PASSED  [ 71%]
src/unit_test_folder/test_unit_test.py::test_invalid_deposit FAILED      [ 85%]
src/unit_test_folder/test_unit_test.py::test_invalid_withdrawal FAILED   [100%]

=================================== FAILURES ===================================
_____________________________ test_invalid_deposit _____________________________

account = <src.code_solution.BankAccount object at 0x7fac0d4f0b90>

    def test_invalid_deposit(account):
        """Test that depositing a negative amount raises ValueError."""
>       with pytest.raises(ValueError):
E       Failed: DID NOT RAISE <class 'ValueError'>

src/unit_test_folder/test_unit_test.py:48: Failed
___________________________ test_invalid_withdrawal ____________________________

account = <src.code_solution.BankAccount object at 0x7fac0d4f1f40>

    def test_invalid_withdrawal(account):
        """Test that withdrawing a negative amount raises ValueError."""
>       with pytest.raises(ValueError):
E       Failed: DID NOT RAISE <class 'ValueError'>

src/unit_test_folder/test_unit_test.py:53: Failed
=========================== short test summary info ============================
FAILED src/unit_test_folder/test_unit_test.py::test_invalid_deposit - Failed:...
FAILED src/unit_test_folder/test_unit_test.py::test_invalid_withdrawal - Fail...
========================= 2 failed, 5 passed in 0.02s ==========================

2024-08-18 14:51:29,170 - INFO - agent.py - Some tests failed.
2024-08-18 14:51:29,172 - INFO - agent.py - ---DECISION: RE-TRY SOLUTION---
================================ Human Message =================================

Your solution failed the unit test: ============================= test session starts ==============================
platform linux -- Python 3.12.3, pytest-8.3.2, pluggy-1.5.0 -- /home/paulo/Python_projects/Self_correcting_coding_agent/venv/bin/python
cachedir: .pytest_cache
rootdir: /home/paulo/Python_projects/Self_correcting_coding_agent
configfile: pyproject.toml
plugins: anyio-4.4.0
collecting ... collected 7 items

src/unit_test_folder/test_unit_test.py::test_initial_balance PASSED      [ 14%]
src/unit_test_folder/test_unit_test.py::test_deposit PASSED              [ 28%]
src/unit_test_folder/test_unit_test.py::test_withdrawal_successful PASSED [ 42%]
src/unit_test_folder/test_unit_test.py::test_withdrawal_insufficient_funds PASSED [ 57%]
src/unit_test_folder/test_unit_test.py::test_transaction_history PASSED  [ 71%]
src/unit_test_folder/test_unit_test.py::test_invalid_deposit FAILED      [ 85%]
src/unit_test_folder/test_unit_test.py::test_invalid_withdrawal FAILED   [100%]

=================================== FAILURES ===================================
_____________________________ test_invalid_deposit _____________________________

account = <src.code_solution.BankAccount object at 0x7fac0d4f0b90>

    def test_invalid_deposit(account):
        """Test that depositing a negative amount raises ValueError."""
>       with pytest.raises(ValueError):
E       Failed: DID NOT RAISE <c ... (truncated)
2024-08-18 14:51:29,175 - INFO - agent.py - ---GENERATING CODE SOLUTION---
================================== Ai Message ==================================

Here is my attempt to solve the problem: The previous solution did not handle cases where negative amounts were deposited or withdrawn. The unit tests expect a ValueError to be raised in such cases, but the original implementation did not include this validation. To fix this, we need to add checks in the deposit and withdraw methods to raise a ValueError if the amount is negative. 
Imports: from datetime import datetime 
Code: class BankAccount:
    def __init__(self):
        self.balance = 0.0
        self.transactions = []

    def deposit(self, amount: float) -> None:
        if amount < 0:
            raise ValueError("Deposit amount must be positive.")
        self.balance += amount
        transaction = {
            'type': 'deposit',
            'amount': amount,
            'date': datetime.now()
        }
        self.transactions.append(transaction)

    def withdraw(self, amount: float) -> bool:
        if amount < 0:
            raise ValueError("Withdrawal amount must be positive.")
        if amount > self.balance:
            return False
        self.balance -= amount
        transaction = {
            'type': 'withdrawal',
            'amount': amount,
            'date': datetime.now()
        }
        self.transactions.append(transaction)
        return True

    def get_balance(self) -> float:
        return self.balance

    def get_transaction_history(self) -> ... (truncated)
2024-08-18 14:51:34,913 - INFO - agent.py - ---CHECKING CODE---
2024-08-18 14:51:35,131 - INFO - agent.py - 
============================= test session starts ==============================
platform linux -- Python 3.12.3, pytest-8.3.2, pluggy-1.5.0 -- /home/paulo/Python_projects/Self_correcting_coding_agent/venv/bin/python
cachedir: .pytest_cache
rootdir: /home/paulo/Python_projects/Self_correcting_coding_agent
configfile: pyproject.toml
plugins: anyio-4.4.0
collecting ... collected 7 items

src/unit_test_folder/test_unit_test.py::test_initial_balance PASSED      [ 14%]
src/unit_test_folder/test_unit_test.py::test_deposit PASSED              [ 28%]
src/unit_test_folder/test_unit_test.py::test_withdrawal_successful PASSED [ 42%]
src/unit_test_folder/test_unit_test.py::test_withdrawal_insufficient_funds PASSED [ 57%]
src/unit_test_folder/test_unit_test.py::test_transaction_history PASSED  [ 71%]
src/unit_test_folder/test_unit_test.py::test_invalid_deposit PASSED      [ 85%]
src/unit_test_folder/test_unit_test.py::test_invalid_withdrawal PASSED   [100%]

============================== 7 passed in 0.01s ===============================

2024-08-18 14:51:35,131 - INFO - agent.py - All tests passed!
2024-08-18 14:51:35,131 - INFO - agent.py - ---NO CODE TEST FAILURES---
2024-08-18 14:51:35,133 - INFO - agent.py - ---DECISION: FINISH---

About

This repository implements an LLM agent that receives a user's prompt and corresponding unit tests. The agent generates the code, tests it for correct execution, and ensures it passes the unit tests. If the code fails to execute or fails any unit test, the agent regenerates the code until it succeeds.

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages