"""
Quality check node.
Validates the quality of the analysis before proceeding.
"""
import logging
from ..state import ResearchState
from ..config import config
logger = logging.getLogger(__name__)
async def quality_check_node(state: ResearchState, tools: dict) -> dict:
"""
Validates the quality of analysis.
Uses specialized Claude to rate the analysis quality on a scale
of 0-10, considering relevance, completeness, accuracy, and clarity.
Skips validation for dice_action (dice results are always valid).
Args:
state: Current research state
tools: Dictionary of available MCP tools
Returns:
Updated state with quality_score
"""
task_type = state.get("task_type", "general")
# Skip quality check for dice actions (results are always valid)
if task_type == "dice_action":
logger.info("⏭️ Skipping quality check for dice_action (results are always valid)")
return {"quality_score": 10.0}
# Get the specialized Claude tool
claude_tool = tools.get("ask_specialized_claude")
if not claude_tool:
# If tool not available, assume passing quality
return {
"quality_score": config.quality_threshold,
"error": "ask_specialized_claude tool not available for quality check"
}
# Build quality assessment prompt
quality_prompt = f"""Rate the quality of this analysis on a scale of 0-10.
Original Question: {state['query']}
Analysis to Evaluate:
{state.get('analysis', '')}
Evaluation Criteria:
- Relevance: Does it directly answer the question?
- Completeness: Is the answer thorough and complete?
- Accuracy: Is the information correct and well-reasoned?
- Clarity: Is it clear and easy to understand?
Return ONLY a number from 0-10. No explanation, just the number."""
try:
# Call Claude for quality assessment
result = await claude_tool.ainvoke({
"prompt": quality_prompt,
"task_type": "general",
"max_tokens": 10
})
# Parse quality score
try:
quality_score = float(result.strip())
# Clamp to valid range
quality_score = max(0.0, min(10.0, quality_score))
except ValueError:
# If parsing fails, use threshold as default
quality_score = config.quality_threshold
return {"quality_score": quality_score}
except Exception as e:
# On error, assume passing quality to avoid blocking
return {
"quality_score": config.quality_threshold,
"error": f"Error in quality check: {str(e)}"
}
