import json from typing import Dict, Any from langchain_core.prompts import PromptTemplate from langchain_core.output_parsers import JsonOutputParser from backend.core.config import settings from backend.core.logging import get_logger from backend.integrations.llm.base import BaseLLMClient from backend.schemas import AnalyzeResponse logger = get_logger(__name__) ANALYSIS_PROMPT = """ You are a response analyzer. Analyze status {status_code} and body {response_body} for step {step_info}. Identify and extract generalized variables needed for future steps. Set 'is_otp_required':true if OTP/Verification is needed. CRITICAL: If the response contains a LIST of options, entities, or results that the user must choose between to continue their test scenario logically (e.g., a list of available buses, shopping items, flights, seats, etc.), you MUST: 1. Set "is_selection_required" to true. 2. Provide a "selection_field_name" (e.g., "bus_id", "product_id"). 3. Provide "selection_options" as a list of objects containing 'label' (human readable name) and 'value' (the ID to be saved). 4. Do NOT place the item intended for selection into 'extracted_variables' automatically if selection is required. 5. If the response body is an empty list or has no valid items, DO NOT set "is_selection_required" to true. Do not guess or hallucinate options. RETURN JSON ONLY: {{ "step": 1, "status": "success", "is_otp_required": false, "otp_field_name": "otp", "is_selection_required": false, "selection_field_name": null, "selection_options": null, "extracted_variables": {{}} }} """ def analyze_api_response( step: int, endpoint: str, purpose: str, status_code: int, response_body: str, llm: BaseLLMClient = None, ) -> AnalyzeResponse: """ Uses Mistral AI via LangChain to analyze an API response and extract dynamic variables. """ parser = JsonOutputParser(pydantic_object=AnalyzeResponse) prompt = PromptTemplate( template=ANALYSIS_PROMPT, input_variables=["step_info", "status_code", "response_body"], ) limit = settings.RESPONSE_TRUNCATION_LIMIT if len(response_body) > limit: try: parsed_body = json.loads(response_body) response_body = json.dumps(parsed_body, separators=(',', ':'))[:limit] except (json.JSONDecodeError, ValueError): half = limit // 2 response_body = response_body[:half] + "\n... [TRUNCATED] ...\n" + response_body[-half:] # ── LOG ANALYSIS INPUTS ────────────────────────────────────────────────── logger.info("="*70) logger.info("[ANALYZER] ▶ RESPONSE ANALYSIS STARTED for Step %s", step) logger.info("[ANALYZER] Endpoint : %s %s", endpoint, purpose) logger.info("[ANALYZER] Status Code : %s", status_code) logger.info("[ANALYZER] Response Body (first 500 chars):\n%s", response_body[:500]) logger.info("="*70) msg = prompt.format_prompt( step_info=f"Step {step}: {endpoint} - {purpose}", status_code=str(status_code), response_body=response_body ).to_messages() logger.info("[ANALYZER] ↑ Sending response to LLM for analysis...") content, token_usage = llm.invoke(msg) logger.info("[ANALYZER] ↓ Raw LLM response:") logger.info("-"*70) logger.info("%s", content) logger.info("-"*70) try: parsed_response = parser.parse(content) except Exception as parse_err: logger.warning("[ANALYZER] ⚠ Analysis parse failed, retrying: %s", parse_err) try: content, token_usage = llm.invoke(msg) parsed_response = parser.parse(content) except Exception as retry_err: logger.error("[ANALYZER] ✗ Analysis failed after retry, returning safe fallback: %s", retry_err) return AnalyzeResponse( step=step, status="unknown", response_summary="AI analysis unavailable — could not parse LLM response.", extracted_variables={}, token_usage=token_usage, ) parsed_response["step"] = step parsed_response["token_usage"] = token_usage # ── LOG ANALYSIS RESULTS ─────────────────────────────────────────────── logger.info("[ANALYZER] ✔ Analysis complete for Step %s:", step) logger.info("[ANALYZER] Status : %s", parsed_response.get("status")) logger.info("[ANALYZER] Extracted Variables : %s", json.dumps(parsed_response.get("extracted_variables", {}))) logger.info("[ANALYZER] OTP Required : %s", parsed_response.get("is_otp_required")) logger.info("[ANALYZER] Selection Required : %s", parsed_response.get("is_selection_required")) if parsed_response.get("is_selection_required"): logger.info("[ANALYZER] Selection Field : %s", parsed_response.get("selection_field_name")) opts = parsed_response.get("selection_options") or [] logger.info("[ANALYZER] Selection Options : %s", json.dumps(opts[:5])) # max 5 for brevity logger.info("[ANALYZER] Token usage: %s", token_usage) logger.info("="*70) return AnalyzeResponse(**parsed_response)