Merge branch 'main' into juleshenry/main

.env.sample

@@ -0,0 +1,2 @@
+OPENAI_API_KEY=your_api_key
+#DEFAULT_MODEL=gpt-3.5-turbo

.gitignore

@@ -1,2 +1,5 @@
 venv
-openai_key.txt
+.venv
+.env
+env/
+.vscode/

README.md

@@ -13,8 +13,11 @@ For a quick demonstration see my [demo video on twitter](https://twitter.com/bio
     python3 -m venv venv
     source venv/bin/activate
     pip install -r requirements.txt
+    cp .env.sample .env
 
-Add your openAI api key to `openai_key.txt` - _warning!_ by default this uses GPT-4 and may make many repeated calls to the api.
+Add your openAI api key to `.env`
+
+_warning!_ By default wolverine uses GPT-4 and may make many repeated calls to the api.
 
 ## Example Usage
 
@@ -26,13 +29,25 @@ You can also run with other models, but be warned they may not adhere to the edi
 
     python wolverine.py --model=gpt-3.5-turbo buggy_script.py "subtract" 20 3
 
+If you want to use GPT-3.5 by default instead of GPT-4 uncomment the default model line in `.env`:
+
+    DEFAULT_MODEL=gpt-3.5-turbo
+
+You can also use flag `--confirm=True` which will ask you `yes or no` before making changes to the file. If flag is not used then it will apply the changes to the file
+
+    python wolverine.py buggy_script.py "subtract" 20 3 --confirm=True
+
 ## Future Plans
 
 This is just a quick prototype I threw together in a few hours. There are many possible extensions and contributions are welcome:
 
 - add flags to customize usage, such as asking for user confirmation before running changed code
 - further iterations on the edit format that GPT responds in. Currently it struggles a bit with indentation, but I'm sure that can be improved
-- a suite of example buggy files that we can test prompts on to ensure reliablity and measure improvement
+- a suite of example buggy files that we can test prompts on to ensure reliability and measure improvement
 - multiple files / codebases: send GPT everything that appears in the stacktrace
 - graceful handling of large files - should we just send GPT relevant classes / functions?
 - extension to languages other than python
+
+## Star History
+
+[![Star History Chart](https://api.star-history.com/svg?repos=biobootloader/wolverine&type=Date)](https://star-history.com/#biobootloader/wolverine)

buggy_script.js

@@ -0,0 +1,31 @@
+const subtractNumbers = (a, b) => {
+  return a - b;
+};
+
+const multiplyNumbers = (a, b) => {
+  return a * b;
+};
+
+const divideNumbers = (a, b) => {
+  return a / b;
+};
+
+function calculate(operation, num1, num2) {
+  let result = '';
+  if (operation == 'add') {
+    result = addNumbers(num1, num2);
+  } else if (operation == 'subtract') {
+    result = subtractNumbers(num1, num2);
+  } else if (operation == 'multiply') {
+    result = multiplyNumbers(num1, num2);
+  } else if (operation == 'divide') {
+    result = divideNumbers(num1, num2);
+  } else {
+    console.log('Invalid operation');
+  }
+
+  return res;
+}
+
+const [, , operation, num1, num2] = process.argv;
+calculate(operation, num1, num2);

prompt.txt

@@ -4,10 +4,13 @@ Because you are part of an automated system, the format you respond in is very s
 
 In addition to the changes, please also provide short explanations of the what went wrong. A single explanation is required, but if you think it's helpful, feel free to provide more explanations for groups of more complicated changes. Be careful to use proper indentation and spacing in your changes. An example response could be:
 
+Be ABSOLUTELY SURE to include the CORRECT INDENTATION when making replacements.
+
+example response:
 [
   {"explanation": "this is just an example, this would usually be a brief explanation of what went wrong"},
   {"operation": "InsertAfter", "line": 10, "content": "x = 1\ny = 2\nz = x * y"},
   {"operation": "Delete", "line": 15, "content": ""},
-  {"operation": "Replace", "line": 18, "content": "x += 1"},
+  {"operation": "Replace", "line": 18, "content": "        x += 1"},
   {"operation": "Delete", "line": 20, "content": ""}
 ]

requirements.txt

@@ -13,6 +13,7 @@ multidict==6.0.4
 openai==0.27.2
 pycodestyle==2.10.0
 pyflakes==3.0.1
+python-dotenv==1.0.0
 requests==2.28.2
 six==1.16.0
 termcolor==2.2.0

wolverine.py

@@ -5,27 +5,84 @@ import os
 import shutil
 import subprocess
 import sys
-
 import openai
 from termcolor import cprint
+from dotenv import load_dotenv
+
 
 # Set up the OpenAI API
-with open("openai_key.txt") as f:
-    openai.api_key = f.read().strip()
+load_dotenv()
+openai.api_key = os.getenv("OPENAI_API_KEY")
+
+DEFAULT_MODEL = os.environ.get("DEFAULT_MODEL", "gpt-4")
+
+
+with open("prompt.txt") as f:
+    SYSTEM_PROMPT = f.read()
 
 
 def run_script(script_name, script_args):
     script_args = [str(arg) for arg in script_args]
+    """
+    If script_name.endswith(".py") then run with python
+    else run with node
+    """
+    subprocess_args = (
+        [sys.executable, script_name, *script_args]
+        if script_name.endswith(".py")
+        else ["node", script_name, *script_args]
+    )
+
     try:
-        result = subprocess.check_output(
-            [sys.executable, script_name, *script_args], stderr=subprocess.STDOUT
-        )
+        result = subprocess.check_output(subprocess_args, stderr=subprocess.STDOUT)
     except subprocess.CalledProcessError as e:
         return e.output.decode("utf-8"), e.returncode
     return result.decode("utf-8"), 0
 
 
-def send_error_to_gpt(file_path, args, error_message, model):
+def json_validated_response(model, messages):
+    """
+    This function is needed because the API can return a non-json response.
+    This will run recursively until a valid json response is returned.
+    todo: might want to stop after a certain number of retries
+    """
+    response = openai.ChatCompletion.create(
+        model=model,
+        messages=messages,
+        temperature=0.5,
+    )
+    messages.append(response.choices[0].message)
+    content = response.choices[0].message.content
+    # see if json can be parsed
+    try:
+        json_start_index = content.index(
+            "["
+        )  # find the starting position of the JSON data
+        json_data = content[
+            json_start_index:
+        ]  # extract the JSON data from the response string
+        json_response = json.loads(json_data)
+    except (json.decoder.JSONDecodeError, ValueError) as e:
+        cprint(f"{e}. Re-running the query.", "red")
+        # debug
+        cprint(f"\nGPT RESPONSE:\n\n{content}\n\n", "yellow")
+        # append a user message that says the json is invalid
+        messages.append(
+            {
+                "role": "user",
+                "content": "Your response could not be parsed by json.loads. Please restate your last message as pure JSON.",
+            }
+        )
+        # rerun the api call
+        return json_validated_response(model, messages)
+    except Exception as e:
+        cprint(f"Unknown error: {e}", "red")
+        cprint(f"\nGPT RESPONSE:\n\n{content}\n\n", "yellow")
+        raise e
+    return json_response
+
+
+def send_error_to_gpt(file_path, args, error_message, model=DEFAULT_MODEL):
     with open(file_path, "r") as f:
         file_lines = f.readlines()
 
@@ -34,12 +91,7 @@ def send_error_to_gpt(file_path, args, error_message, model):
         file_with_lines.append(str(i + 1) + ": " + line)
     file_with_lines = "".join(file_with_lines)
 
-    with open("prompt.txt") as f:
-        initial_prompt_text = f.read()
-
     prompt = (
-        initial_prompt_text +
-        "\n\n"
         "Here is the script that needs fixing:\n\n"
         f"{file_with_lines}\n\n"
         "Here are the arguments it was provided:\n\n"
@@ -51,27 +103,27 @@ def send_error_to_gpt(file_path, args, error_message, model):
     )
 
     # print(prompt)
+    messages = [
+        {
+            "role": "system",
+            "content": SYSTEM_PROMPT,
+        },
+        {
+            "role": "user",
+            "content": prompt,
+        },
+    ]
 
-    response = openai.ChatCompletion.create(
-        model=model,
-        messages=[
-            {
-                "role": "user",
-                "content": prompt,
-            }
-        ],
-        temperature=1.0,
-    )
-
-    return response.choices[0].message.content.strip()
+    return json_validated_response(model, messages)
 
 
-def apply_changes(file_path, changes_json):
+def apply_changes(file_path, changes: list, confirm=False):
+    """
+    Pass changes as loaded json (list of dicts)
+    """
     with open(file_path, "r") as f:
         original_file_lines = f.readlines()
 
-    changes = json.loads(changes_json)
-
     # Filter out explanation elements
     operation_changes = [change for change in changes if "operation" in change]
     explanations = [
@@ -94,6 +146,25 @@ def apply_changes(file_path, changes_json):
         elif operation == "InsertAfter":
             file_lines.insert(line, content + "\n")
 
+    # Ask for user confirmation before writing changes
+    print("\nChanges to be made:")
+
+    diff = difflib.unified_diff(original_file_lines, file_lines, lineterm="")
+    for line in diff:
+        if line.startswith("+"):
+            cprint(line, "green", end="")
+        elif line.startswith("-"):
+            cprint(line, "red", end="")
+        else:
+            print(line, end="")
+
+    # Checking if user used confirm flag
+    if confirm:
+        confirmation = input("Do you want to apply these changes? (y/n): ")
+        if confirmation.lower() != "y":
+            print("Changes not applied")
+            sys.exit(0)
+
     with open(file_path, "w") as f:
         f.writelines(file_lines)
 
@@ -104,7 +175,8 @@ def apply_changes(file_path, changes_json):
 
     # Show the diff
     print("\nChanges:")
-    diff = difflib.unified_diff(original_file_lines, file_lines, lineterm="")
+    diff = difflib.unified_diff(
+        original_file_lines, file_lines, lineterm="")
     for line in diff:
         if line.startswith("+"):
             cprint(line, "green", end="")
@@ -113,8 +185,10 @@ def apply_changes(file_path, changes_json):
         else:
             print(line, end="")
 
+    print("Changes applied.")
 
-def main(script_name, *script_args, revert=False, model="gpt-4"):
+
+def main(script_name, *script_args, revert=False, model=DEFAULT_MODEL, confirm=False):
     if revert:
         backup_file = script_name + ".bak"
         if os.path.exists(backup_file):
@@ -140,12 +214,13 @@ def main(script_name, *script_args, revert=False, model="gpt-4"):
             print("Output:", output)
 
             json_response = send_error_to_gpt(
-                    file_path=script_name,
-                    args=script_args,
-                    error_message=output,
-                    model=model,
+                file_path=script_name,
+                args=script_args,
+                error_message=output,
+                model=model,
             )
-            apply_changes(script_name, json_response)
+
+            apply_changes(script_name, json_response, confirm=confirm)
             cprint("Changes applied. Rerunning...", "blue")