evaluator_agent.py

from pydantic import BaseModel, Field
from agents import Agent

INSTRUCTIONS = (
    "You evaluate whether web search summaries are sufficient to write a strong report on the user's "
    "original query. Judge coverage, relevance, and depth—not writing quality. "
    "If summaries are off-topic, empty, contradictory, or miss major aspects of the query, "
    "set proceed_to_report to false and explain what is missing or wrong in feedback_for_planner. "
    "If the material is good enough to synthesize a solid answer, set proceed_to_report to true. "
    "Be pragmatic: partial but relevant evidence can be enough; do not demand perfection."
)


class ResearchEvaluation(BaseModel):
    proceed_to_report: bool = Field(
        description="True if search summaries are adequate to write the report; false to replan searches."
    )
    rationale: str = Field(description="Brief justification for the decision.")
    feedback_for_planner: str = Field(
        description=(
            "If replanning: concrete gaps, bad angles, or suggested directions for new searches. "
            "If proceeding: may be empty or a short note."
        ),
    )


evaluator_agent = Agent(
    name="EvaluatorAgent",
    instructions=INSTRUCTIONS,
    model="gpt-4o-mini",
    output_type=ResearchEvaluation,
)