agent/tools/credibility_tool.py

import json
from crewai.tools import BaseTool
from pydantic import BaseModel, Field
from credibility.scorer import score_source
import config


class CredibilityInput(BaseModel):
    sources_json: str = Field(
        description="JSON string of list of source dicts with keys: url, title, content, published_date"
    )
    fast_moving_topic: bool = Field(
        default=False,
        description="True if topic is time-sensitive (news, recent tech), False for evergreen topics"
    )


class CredibilityScorerTool(BaseTool):
    name: str = "Source Credibility Scorer"
    description: str = (
        "Score a list of sources for credibility (0.0–1.0). "
        "Filters out unreliable sources. Returns only accepted sources with their scores. "
        "Sources below 0.35 are dropped. Sources 0.35–0.55 are flagged low-confidence."
    )
    args_schema: type[BaseModel] = CredibilityInput

    def _run(self, sources_json: str, fast_moving_topic: bool = False) -> str:
        try:
            sources = json.loads(sources_json)
        except json.JSONDecodeError as e:
            return json.dumps({"error": f"Invalid JSON: {e}", "accepted": [], "dropped": []})

        accepted = []
        dropped = []

        for src in sources:
            result = score_source(
                url=src.get("url", ""),
                title=src.get("title", ""),
                content=src.get("content", ""),
                pub_date=src.get("published_date"),
                fast_moving=fast_moving_topic,
            )

            if result.get("blocked") or result["score"] < config.CREDIBILITY_CUTOFF:
                dropped.append({"url": src.get("url"), "score": result["score"], "reason": "below threshold"})
                continue

            confidence = (
                "high" if result["score"] >= config.LOW_CONFIDENCE_CUTOFF else "low"
            )

            accepted.append({
                **src,
                "credibility_score": result["score"],
                "confidence": confidence,
                "score_breakdown": result.get("breakdown", {}),
            })

        accepted.sort(key=lambda x: x["credibility_score"], reverse=True)
        accepted = accepted[:config.MAX_ACCEPTED_SOURCES]

        return json.dumps({
            "accepted": accepted,
            "dropped_count": len(dropped),
            "accepted_count": len(accepted),
        })