146 lines
5.0 KiB
Python
146 lines
5.0 KiB
Python
import os
|
|
import asyncio
|
|
from typing import TypedDict, Annotated, Dict
|
|
|
|
from langchain_openai import ChatOpenAI
|
|
from langchain_core.messages import HumanMessage
|
|
from langchain.tools import tool
|
|
from deepagents import create_deep_agent
|
|
from deepagents.backends import FilesystemBackend, LocalShellBackend, CompositeBackend
|
|
|
|
from langgraph.graph import StateGraph, START, END
|
|
from langgraph.graph.message import add_messages
|
|
from pydantic import BaseModel, Field
|
|
from langchain_core.output_parsers import PydanticOutputParser
|
|
|
|
# ---------- LLM ----------
|
|
llm = ChatOpenAI(
|
|
model="openai/gpt-oss-20b:free",
|
|
base_url="https://openrouter.ai/api/v1",
|
|
api_key=os.getenv("OPENAI_API_KEY"),
|
|
temperature=0.0,
|
|
)
|
|
|
|
# ---------- State ----------
|
|
class CodeReviewState(TypedDict):
|
|
code: str
|
|
draft_review: str
|
|
criteria_scores: Dict[str, int]
|
|
weakest_criterion: str
|
|
verdict: str
|
|
round: int
|
|
max_rounds: int
|
|
|
|
# ---------- Pydantic for reflect output ----------
|
|
class ReflectOutput(BaseModel):
|
|
pep8: int = Field(..., description="Score 0-10 for PEP8 compliance")
|
|
type_hints: int = Field(..., description="Score 0-10 for type hints usage")
|
|
edge_cases: int = Field(..., description="Score 0-10 for edge case handling")
|
|
naming: int = Field(..., description="Score 0-10 for naming conventions")
|
|
weakest_criterion: str = Field(..., description="Name of the weakest criterion")
|
|
verdict: str = Field(..., description="'ok' or 'needs_revision'")
|
|
|
|
reflect_parser = PydanticOutputParser(pydantic_object=ReflectOutput)
|
|
|
|
# ---------- Nodes ----------
|
|
async def draft_review(state: CodeReviewState) -> CodeReviewState:
|
|
prompt = f"""Please write a concise code review (3-6 bullet points) for the following Python function. Focus on style, type hints, edge cases, and naming.
|
|
|
|
```python
|
|
{state['code']}
|
|
```
|
|
|
|
Return only the review text."""
|
|
review = await llm.ainvoke([HumanMessage(content=prompt)])
|
|
state['draft_review'] = review.content.strip()
|
|
return state
|
|
|
|
async def reflect(state: CodeReviewState) -> CodeReviewState:
|
|
prompt = f"""You are a code review critic. Evaluate the following review text and assign scores 0-10 for each of the four criteria: pep8, type_hints, edge_cases, naming. Also identify the weakest criterion and decide if the review is "ok" or "needs_revision".
|
|
|
|
Review text:
|
|
{state['draft_review']}
|
|
|
|
Provide the output in the following JSON-like format:
|
|
{{"pep8": int, "type_hints": int, "edge_cases": int, "naming": int, "weakest_criterion": str, "verdict": str}}
|
|
"""
|
|
raw = await llm.ainvoke([HumanMessage(content=prompt)])
|
|
parsed = reflect_parser.parse(raw.content)
|
|
state['criteria_scores'] = {
|
|
"pep8": parsed.pep8,
|
|
"type_hints": parsed.type_hints,
|
|
"edge_cases": parsed.edge_cases,
|
|
"naming": parsed.naming,
|
|
}
|
|
state['weakest_criterion'] = parsed.weakest_criterion
|
|
state['verdict'] = parsed.verdict
|
|
return state
|
|
|
|
async def rewrite(state: CodeReviewState) -> CodeReviewState:
|
|
# Simple rewrite: add a sentence addressing the weakest criterion
|
|
additional = f"Additionally, the review should pay more attention to {state['weakest_criterion']}.")
|
|
state['draft_review'] = state['draft_review'] + "\n" + additional
|
|
state['round'] += 1
|
|
return state
|
|
|
|
# ---------- Graph ----------
|
|
def build_graph() -> StateGraph[CodeReviewState]:
|
|
graph = StateGraph(CodeReviewState)
|
|
graph.add_node("draft_review", draft_review)
|
|
graph.add_node("reflect", reflect)
|
|
graph.add_node("rewrite", rewrite)
|
|
|
|
graph.set_entry_point("draft_review")
|
|
graph.add_edge("draft_review", "reflect")
|
|
graph.add_conditional_edges(
|
|
"reflect",
|
|
lambda x: "END" if x['verdict'] == "ok" or x['round'] >= x['max_rounds'] else "rewrite",
|
|
)
|
|
graph.add_edge("rewrite", "reflect")
|
|
|
|
return graph.compile()
|
|
|
|
# ---------- Tool ----------
|
|
@tool
|
|
def code_review_tool(code: str) -> str:
|
|
"""Perform a structured code review with possible rewrites."""
|
|
graph = build_graph()
|
|
initial_state: CodeReviewState = {
|
|
"code": code,
|
|
"draft_review": "",
|
|
"criteria_scores": {},
|
|
"weakest_criterion": "",
|
|
"verdict": "",
|
|
"round": 0,
|
|
"max_rounds": 2,
|
|
}
|
|
final_state = graph.invoke(initial_state)
|
|
return f"Final Review:\n{final_state['draft_review']}\n\nScores: {final_state['criteria_scores']}"
|
|
|
|
# ---------- DeepAgent ----------
|
|
backend = CompositeBackend([
|
|
LocalShellBackend(workspace_dir="./workspace"),
|
|
FilesystemBackend(),
|
|
])
|
|
|
|
agent = create_deep_agent(
|
|
model=llm,
|
|
tools=[code_review_tool],
|
|
backend=backend,
|
|
system_prompt="You are a helpful code review assistant.",
|
|
)
|
|
|
|
async def main():
|
|
sample_code = """
|
|
def sort_numbers(arr):
|
|
return sorted(arr)
|
|
"""
|
|
result = await agent.ainvoke(
|
|
{"messages": [HumanMessage(content=f"Please review this function:\n{sample_code}")]},
|
|
{"configurable": {"thread_id": "session-1"}},
|
|
)
|
|
print(result["messages"][-1].content)
|
|
|
|
if __name__ == "__main__":
|
|
asyncio.run(main())
|