api_documentation_creator.py

from argparse import ArgumentParser
from openai import OpenAI
import os

COMMAND = 'COMMAND'
NO_OUTPUT = 'NO OUTPUT'


def main():
    arg_parser = ArgumentParser(prog='API Documentation Creator')
    arg_parser.add_argument('-m', '--method', required=True)
    arg_parser.add_argument('-p', '--path', required=True, nargs='+')

    args = arg_parser.parse_args()
    method = args.method
    paths = ','.join(args.path)

    client = OpenAI(api_key=os.environ['OPENAI_KEY'])

    system_message = f"""You are a software engineer assistant for writing REST API documentation. You will be provided
with java class name and method that is entrypoint for an API resource. Spring Boot MVC framework is used to create
application.

You need to create documentation in a confluence format. It should include: single code block with http method and
API path; request parameters table; request body table; code block with json body example; response 
fields table; code block with json response example. Table fields: name, type, required (TRUE or FALSE), description.

Use linux commands to get all needed classes content. Use find command to search for java class files. Get file content 
using cat command. Issue commands use '{COMMAND}' word as the first line. Use only 1 command per message. Do not mix 
multiple commands or command and documentation in one message. Do not include description or explanation for commands. 
Use response data as context for the following command. Output full result as a final message only without intermediate 
results.

It is important to take into account generic java types. For example for Mono<Response<Model>> you should get Response 
class content also and include in documentation also as it contains fields not only Model class. Mono class should be 
ignored as it doesn't contain API fields. 

If command output returns {NO_OUTPUT} just use class name for documentation skipping internal details.
"""

    conversation = [
        {
            "role": "system",
            "content": system_message
        },
        {
            "role": "user",
            "content": f"Class and method: {method}. Search paths: {paths}"
        }
    ]

    while True:
        response = client.chat.completions.create(
            model="gpt-4",
            messages=conversation,
            temperature=0.7,
            max_tokens=500,
            top_p=1
        )

        response_message = response.choices[0].message
        conversation.append(response_message)

        content = response_message.content
        if content.startswith(f"{COMMAND}\n"):
            command = content.replace(f"{COMMAND}\n", "")
            print(command)

            cmd_output = os.popen(command).read()
            if cmd_output == '':
                cmd_output = NO_OUTPUT

            message = {
                "role": "user",
                "content": cmd_output
            }
            conversation.append(message)
        else:
            print(content)
            break


if __name__ == '__main__':
    main()