172 lines
6.4 KiB
Python
172 lines
6.4 KiB
Python
"""
|
||
# main.py
|
||
# LangGraph code review agent with reflection and rewrite loop.
|
||
# Implements the task specification without any deepagents dependency.
|
||
# Uses OpenRouter via langchain-openai.
|
||
|
||
import os
|
||
import asyncio
|
||
from typing import TypedDict, Annotated, Dict
|
||
from langchain_openai import ChatOpenAI
|
||
from langchain_core.messages import HumanMessage, AIMessage
|
||
from langchain_core.output_parsers import PydanticOutputParser
|
||
from langchain_core.pydantic_v1 import BaseModel, Field
|
||
from langgraph.graph import StateGraph, START, END
|
||
from langgraph.graph.message import add_messages
|
||
from dotenv import load_dotenv
|
||
|
||
load_dotenv()
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# State definition
|
||
# ---------------------------------------------------------------------------
|
||
class CodeReviewState(TypedDict):
|
||
code: str
|
||
draft_review: str
|
||
criteria_scores: Dict[str, int]
|
||
weakest_criterion: str
|
||
verdict: str # "ok" | "needs_revision"
|
||
round: int
|
||
max_rounds: int
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# LLM setup
|
||
# ---------------------------------------------------------------------------
|
||
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,
|
||
)
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# Structured output models for reflect node
|
||
# ---------------------------------------------------------------------------
|
||
class ReflectOutput(BaseModel):
|
||
pep8: int = Field(..., ge=0, le=10)
|
||
type_hints: int = Field(..., ge=0, le=10)
|
||
edge_cases: int = Field(..., ge=0, le=10)
|
||
naming: int = Field(..., ge=0, le=10)
|
||
weakest_criterion: str = Field(...)
|
||
verdict: str = Field(..., regex="^(ok|needs_revision)$")
|
||
|
||
reflect_parser = PydanticOutputParser(pydantic_object=ReflectOutput)
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# Node implementations
|
||
# ---------------------------------------------------------------------------
|
||
async def draft_review_node(state: CodeReviewState) -> CodeReviewState:
|
||
"""Generate an initial code review with 3–6 bullet points."""
|
||
prompt = f"""
|
||
You are a senior Python developer. You will write a concise code review for the following function. Provide 3 to 6 bullet points, each starting with a dash.
|
||
|
||
Function code:
|
||
{state['code']}
|
||
|
||
Review:"""
|
||
response = await llm.ainvoke([HumanMessage(content=prompt)])
|
||
state['draft_review'] = response.content.strip()
|
||
return state
|
||
|
||
async def reflect_node(state: CodeReviewState) -> CodeReviewState:
|
||
"""Critic evaluates the draft review on 4 criteria and returns structured scores."""
|
||
prompt = f"""
|
||
You are a code review critic. Evaluate the following draft review on the four criteria below, assigning a score from 0 (worst) to 10 (excellent). Return the scores and the weakest criterion in a JSON format matching the schema:
|
||
|
||
{reflect_parser.get_format_instructions()}
|
||
|
||
Draft review:
|
||
{state['draft_review']}
|
||
|
||
Scores:"""
|
||
response = await llm.ainvoke([HumanMessage(content=prompt)])
|
||
try:
|
||
parsed = reflect_parser.parse(response.content)
|
||
except Exception as e:
|
||
# Fallback: treat as all zeros
|
||
parsed = ReflectOutput(pep8=0, type_hints=0, edge_cases=0, naming=0, weakest_criterion="pep8", verdict="needs_revision")
|
||
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_node(state: CodeReviewState) -> CodeReviewState:
|
||
"""Rewrite the part of the review that addresses the weakest criterion."""
|
||
prompt = f"""
|
||
You are a senior Python developer. The following code review has been identified as weak in the criterion: {state['weakest_criterion']}. Rewrite only the section of the review that addresses this criterion, improving clarity and depth. Keep the rest of the review unchanged.
|
||
|
||
Original review:
|
||
{state['draft_review']}
|
||
|
||
Rewritten review:"""
|
||
response = await llm.ainvoke([HumanMessage(content=prompt)])
|
||
# Replace only the weak section. For simplicity, we replace the whole review.
|
||
state['draft_review'] = response.content.strip()
|
||
state['round'] += 1
|
||
return state
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# Graph construction
|
||
# ---------------------------------------------------------------------------
|
||
def create_graph() -> StateGraph:
|
||
graph = StateGraph(CodeReviewState)
|
||
graph.add_node("draft_review", draft_review_node)
|
||
graph.add_node("reflect", reflect_node)
|
||
graph.add_node("rewrite", rewrite_node)
|
||
|
||
# Entry point
|
||
graph.set_entry_point("draft_review")
|
||
|
||
# Transitions
|
||
graph.add_edge("draft_review", "reflect")
|
||
graph.add_conditional_edges(
|
||
"reflect",
|
||
lambda state: "rewrite" if state["verdict"] == "needs_revision" and state["round"] < state["max_rounds"] else "END",
|
||
)
|
||
graph.add_edge("rewrite", "reflect")
|
||
|
||
return graph
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# CLI helper
|
||
# ---------------------------------------------------------------------------
|
||
async def run_review(code: str, max_rounds: int = 2) -> CodeReviewState:
|
||
initial_state: CodeReviewState = {
|
||
"code": code,
|
||
"draft_review": "",
|
||
"criteria_scores": {},
|
||
"weakest_criterion": "",
|
||
"verdict": "",
|
||
"round": 0,
|
||
"max_rounds": max_rounds,
|
||
}
|
||
graph = create_graph()
|
||
final_state = await graph.astate(initial_state)
|
||
return final_state
|
||
|
||
# ---------------------------------------------------------------------------
|
||
# Demo main
|
||
# ---------------------------------------------------------------------------
|
||
if __name__ == "__main__":
|
||
sample_code = """
|
||
def sort_numbers(arr):
|
||
return sorted(arr)
|
||
"""
|
||
result = asyncio.run(run_review(sample_code))
|
||
print("\n=== Initial Draft Review ===")
|
||
print(result["draft_review"])
|
||
print("\n=== Scores ===")
|
||
print(result["criteria_scores"])
|
||
print("\n=== Verdict ===")
|
||
print(result["verdict"])
|
||
if result["verdict"] == "needs_revision":
|
||
print("\n=== Rewritten Review ===")
|
||
print(result["draft_review"]) # after last rewrite
|
||
print("\n=== Updated Scores ===")
|
||
print(result["criteria_scores"])
|
||
"" |