Agent wyjaśniający kod

Cel projektu

Dlaczego to przydatne?

Architektura

Step 1: Read the File

import sys

with open('example.py', 'w') as f:
    f.write('print("hello")\n')

path = sys.argv[1] if len(sys.argv) > 1 else 'example.py'
with open(path) as f:
    code = f.read()
print(f'Read {len(code)} characters from {path}')

Step 2: Prompt for Documentation

from openai import OpenAI
oai = OpenAI()

prompt = f'''
You are a senior engineer writing developer-friendly docs.

Given this source file, produce a Markdown document with:
# {path}

## Purpose
(One paragraph)

## Public API
(Each function/class with one-line description)

## Usage Example
(One short, runnable snippet)

Source:
```
{code}
```
'''

response = oai.chat.completions.create(
    model='gpt-4o-mini',
    messages=[{'role': 'user', 'content': prompt}],
    temperature=0.2,
)
print(response.choices[0].message.content)

Obsługa długich plików

import ast

tree = ast.parse(code)
functions = [node for node in ast.walk(tree) if isinstance(node, ast.FunctionDef)]

for func in functions:
    snippet = ast.unparse(func)
    explain(snippet)

Krok 3: Łączenie wyników

docs = []
for func_name, snippet in functions:
    explanation = explain(snippet)
    docs.append(f'### {func_name}\n\n{explanation}\n')

full_doc = '\n'.join(docs)
open('docs.md', 'w').write(full_doc)

Dodawanie podsumowania na poziomie projektu

summary_prompt = 'Summarise the purpose of this package in 3 sentences, given these function docs:\n\n' + full_doc
summary = ask(summary_prompt)

Wiele języków

prompt = f'You are documenting {language} code. ...'

Dokumentacja na podstawie różnic

import subprocess
changed = subprocess.check_output(['git', 'diff', '--name-only', 'HEAD~1']).decode().splitlines()
for path in changed:
    if path.endswith('.py'):
        regenerate_doc(path)

Użycie narzędzia do uruchomienia przykładu

def run_python(code):
    try:
        exec(code, {})
        return {'stdout': 'ok', 'stderr': ''}
    except Exception as e:
        return {'stdout': '', 'stderr': str(e)}

tools = [{'name': 'run_python', 'description': 'Execute a Python snippet and return stdout/stderr', 'parameters': {'code': 'str'}}]

broken_example = 'print(1/0)'
result = run_python(broken_example)
if result['stderr']:
    print('Example failed:', result['stderr'])
    fixed_example = 'print(1)'
    result = run_python(fixed_example)
    print('Self-corrected result:', result)
else:
    print('Example ran fine:', result)

Gdy agent halucynuje

Przekształcenie w produkt

# pip install -e .
# docgen src/myproject/agent.py
# Outputs docs.md

Od narzędzia do CI

Dlaczego czytać cały plik?

Podsumowanie