zen-marketing/tools/factcheck.py

"""Fact Check Tool

Quick technical fact verification via web search.
Verifies claims against authoritative sources and provides confidence levels.
"""

from typing import Optional

from pydantic import Field

from config import TEMPERATURE_PRECISION
from systemprompts import FACTCHECK_PROMPT
from tools.models import ToolModelCategory
from tools.shared.base_models import ToolRequest
from tools.simple.base import SimpleTool


class FactCheckRequest(ToolRequest):
    """Request model for Fact Check"""

    content: str = Field(
        ...,
        description="Content containing factual claims to verify. Can be a full article, social post, or specific claims.",
    )
    technical_domain: Optional[str] = Field(
        default=None,
        description="Domain for specialized fact-checking: 'hvac', 'software', 'saas', 'general'. Helps identify authoritative sources.",
    )
    claim_type: Optional[str] = Field(
        default=None,
        description="Type of claims to focus on: 'product_specs', 'technical_process', 'statistics', 'general', 'all'. Default is 'all'.",
    )
    confidence_threshold: str = Field(
        default="balanced",
        description="Verification depth: 'high_confidence_only' (strict, flag anything uncertain), 'balanced' (reasonable verification), 'comprehensive' (check everything thoroughly).",
    )
    focus_claims: Optional[list[str]] = Field(
        default=None,
        description="Specific claims to verify if you don't want to check all content. Provide as list of quoted text.",
    )


class FactCheckTool(SimpleTool):
    """Verify factual claims using web search"""

    def get_name(self) -> str:
        return "factcheck"

    def get_description(self) -> str:
        return (
            "Verify factual claims in content using web search. Checks product specs, statistics, "
            "technical processes, and general facts against authoritative sources. Returns verification "
            "status (✅ Verified / ⚠️ Partial / ❌ Unsupported / 🔍 Needs Context) with sources and "
            "confidence levels. Essential for technical content accuracy before publishing."
        )

    def get_system_prompt(self) -> str:
        return FACTCHECK_PROMPT

    def get_default_temperature(self) -> float:
        return TEMPERATURE_PRECISION

    def get_model_category(self) -> ToolModelCategory:
        return ToolModelCategory.FAST_RESPONSE

    def get_request_model(self):
        return FactCheckRequest

    def get_tool_fields(self) -> dict:
        """Tool-specific field definitions for FactCheck"""
        return {
            "content": {
                "type": "string",
                "description": "Content with claims to verify",
            }
        }

    async def prepare_prompt(self, request: FactCheckRequest) -> str:
        """Prepare the fact check prompt"""
        prompt_parts = ["Verify the factual claims in this content:"]

        if request.focus_claims:
            prompt_parts.append("\n**Specific Claims to Verify:**")
            for claim in request.focus_claims:
                prompt_parts.append(f"- {claim}")
            prompt_parts.append(f"\n**Full Content for Context:**\n{request.content}")
        else:
            prompt_parts.append(f"\n**Content to Fact-Check:**\n{request.content}")

        if request.technical_domain:
            prompt_parts.append(
                f"\n**Technical Domain:** {request.technical_domain} (use domain-specific authoritative sources)"
            )

        if request.claim_type and request.claim_type != "all":
            prompt_parts.append(
                f"\n**Focus on Claim Type:** {request.claim_type}"
            )

        prompt_parts.append(
            f"\n**Verification Approach:** {request.confidence_threshold}"
        )

        if request.confidence_threshold == "high_confidence_only":
            prompt_parts.append(
                "→ Flag any claims that cannot be verified with high confidence from multiple reliable sources"
            )
        elif request.confidence_threshold == "comprehensive":
            prompt_parts.append(
                "→ Thoroughly check every factual statement, including common knowledge claims"
            )
        else:  # balanced
            prompt_parts.append(
                "→ Focus on key claims and statistics; flag obvious issues; provide context for nuanced claims"
            )

        prompt_parts.append(
            "\n\nUse web search to verify claims. For each claim, provide verification status, "
            "evidence summary, confidence level, sources, and recommendations if needed."
        )

        # Build prompt text without mutating request object
        prompt_text = "\n".join(prompt_parts)

        # Create a copy for chat-style preparation
        from copy import copy

        request_copy = copy(request)
        request_copy.prompt = prompt_text
        return self.prepare_chat_style_prompt(request_copy)

    def get_input_schema(self) -> dict:
        """Return the JSON schema for this tool's input"""
        return {
            "type": "object",
            "properties": {
                "content": {
                    "type": "string",
                    "description": "Content containing factual claims to verify",
                },
                "technical_domain": {
                    "type": "string",
                    "description": "Domain: 'hvac', 'software', 'saas', 'general'",
                },
                "claim_type": {
                    "type": "string",
                    "description": "Claim type: 'product_specs', 'technical_process', 'statistics', 'general', 'all'",
                },
                "confidence_threshold": {
                    "type": "string",
                    "description": "Verification depth: 'high_confidence_only', 'balanced', 'comprehensive'",
                    "default": "balanced",
                },
                "focus_claims": {
                    "type": "array",
                    "items": {"type": "string"},
                    "description": "Specific claims to verify (optional, checks all if not provided)",
                },
                "files": {
                    "type": "array",
                    "items": {"type": "string"},
                    "description": "Optional reference documents or source materials",
                },
                "images": {
                    "type": "array",
                    "items": {"type": "string"},
                    "description": "Optional screenshots or spec sheets to verify",
                },
                "continuation_id": {
                    "type": "string",
                    "description": "Thread ID to continue previous conversation",
                },
                "model": {
                    "type": "string",
                    "description": "AI model to use (leave empty for default analytical model)",
                },
                "temperature": {
                    "type": "number",
                    "description": "Creativity level 0.0-1.0 (default 0.2 for precision)",
                    "minimum": 0.0,
                    "maximum": 1.0,
                },
                "thinking_mode": {
                    "type": "string",
                    "description": "Thinking depth: minimal, low, medium, high, max",
                    "enum": ["minimal", "low", "medium", "high", "max"],
                },
                "use_websearch": {
                    "type": "boolean",
                    "description": "Enable web search (REQUIRED for fact-checking, defaults to true)",
                    "default": True,
                },
            },
            "required": ["content"],
        }