task-6a22c713fd30e81cf315ea04/main.py

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,
)

# ---------- Backend ----------
backend = CompositeBackend([
    LocalShellBackend(workspace_dir="./workspace"),
    FilesystemBackend(),
])

# ---------- 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 model for reflect output ----------
class ReflectOutput(BaseModel):
    pep8: int = Field(description="Score for PEP8 compliance (0-10)")
    type_hints: int = Field(description="Score for type hints (0-10)")
    edge_cases: int = Field(description="Score for edge case handling (0-10)")
    naming: int = Field(description="Score for naming conventions (0-10)")
    weakest_criterion: str = Field(description="The criterion with the lowest score")
    verdict: str = Field(description="'ok' or 'needs_revision'")

parser = PydanticOutputParser(pydantic_object=ReflectOutput)

# ---------- Nodes ----------
async def draft_review(state: CodeReviewState) -> CodeReviewState:
    prompt = f"""
    Write a concise code review for the following Python function. Provide 3-6 bullet points.

    Function:
    ```python
    {state['code']}
    ```
    """
    response = await llm.ainvoke(HumanMessage(content=prompt))
    state['draft_review'] = response.content.strip()
    return state

async def reflect(state: CodeReviewState) -> CodeReviewState:
    prompt = f"""
    Evaluate the following code and draft review. Score each of the four criteria on a scale 0-10:
    - PEP8 compliance
    - Type hints
    - Edge case handling
    - Naming conventions

    Provide the scores, identify the weakest criterion, and give a verdict ('ok' or 'needs_revision').

    Code:
    ```python
    {state['code']}
    ```

    Draft Review:
    ```text
    {state['draft_review']}
    ```

    Return a JSON object with keys: pep8, type_hints, edge_cases, naming, weakest_criterion, verdict.
    """
    response = await llm.ainvoke(HumanMessage(content=prompt))
    try:
        parsed = parser.parse(response.content)
    except Exception as e:
        # Fallback: set all scores to 0 and verdict to needs_revision
        parsed = ReflectOutput(pep8=0, type_hints=0, edge_cases=0, naming=0, weakest_criterion="unknown", 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(state: CodeReviewState) -> CodeReviewState:
    prompt = f"""
    Rewrite the section of the draft review that addresses the weakest criterion: {state['weakest_criterion']}.
    Keep all other parts of the review unchanged.

    Current Draft Review:
    ```text
    {state['draft_review']}
    ```
    """
    response = await llm.ainvoke(HumanMessage(content=prompt))
    state['draft_review'] = response.content.strip()
    state['round'] += 1
    return state

# ---------- Graph ----------
review_graph = StateGraph(CodeReviewState)
review_graph.add_node("draft_review", draft_review)
review_graph.add_node("reflect", reflect)
review_graph.add_node("rewrite", rewrite)

review_graph.set_entry_point("draft_review")
review_graph.add_edge("draft_review", "reflect")

# Conditional edges after reflect
review_graph.add_conditional_edges(
    "reflect",
    lambda state: "END" if state["verdict"] == "ok" else "rewrite" if state["round"] < state["max_rounds"] else "END",
)
review_graph.add_edge("rewrite", "reflect")

compiled_graph = review_graph.compile()

# ---------- Tool ----------
@tool
async def run_code_review(code: str) -> Dict:
    """Run a code review on the provided Python function."""
    initial_state: CodeReviewState = {
        "code": code,
        "draft_review": "",
        "criteria_scores": {},
        "weakest_criterion": "",
        "verdict": "",
        "round": 0,
        "max_rounds": 2,
    }
    final_state = compiled_graph.invoke(initial_state)
    return {
        "final_review": final_state["draft_review"],
        "scores": final_state["criteria_scores"],
        "weakest_criterion": final_state["weakest_criterion"],
        "verdict": final_state["verdict"],
        "rounds": final_state["round"],
    }

# ---------- Agent ----------
agent = create_deep_agent(
    model=llm,
    tools=[run_code_review],
    backend=backend,
    system_prompt="You are a helpful code review agent.",
)

# ---------- Demo ----------
async def main():
    # Sample function to review
    code = """
    def sort_numbers(arr):
        return sorted(arr)
    """
    result = await run_code_review(code)
    print("\n=== Final Review ===\n")
    print(result["final_review"])
    print("\n=== Scores ===\n")
    for k, v in result["scores"].items():
        print(f"{k}: {v}")
    print(f"\nVerdict: {result['verdict']} (Rounds: {result['rounds']})")

if __name__ == "__main__":
    asyncio.run(main())