Spaces:
Running
Running
File size: 19,468 Bytes
a23082c 114747f a23082c 68bd1d5 a23082c 114747f a23082c 68bd1d5 a23082c 114747f a23082c 68bd1d5 a23082c 114747f a23082c 68bd1d5 a23082c |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 |
import os
import logging
import json
from typing import List, Dict, Optional, Union
from llama_index.core.agent.workflow import ReActAgent
from llama_index.core.tools import FunctionTool
from llama_index.llms.google_genai import GoogleGenAI
# Assuming research_agent might be needed for handoff, but not directly imported
# Setup logging
logger = logging.getLogger(__name__)
# Helper function to load prompt from file
def load_prompt_from_file(filename: str, default_prompt: str) -> str:
"""Loads a prompt from a text file."""
try:
script_dir = os.path.dirname(__file__)
prompt_path = os.path.join(script_dir, filename)
with open(prompt_path, "r") as f:
prompt = f.read()
logger.info(f"Successfully loaded prompt from {prompt_path}")
return prompt
except FileNotFoundError:
logger.warning(f"Prompt file {filename} not found at {prompt_path}. Using default.")
return default_prompt
except Exception as e:
logger.error(f"Error loading prompt file {filename}: {e}", exc_info=True)
return default_prompt
# --- Tool Functions ---
# Note: cross_reference_check might require fetching content.
# This version assumes content is provided or delegates fetching via handoff.
def cross_reference_check(claim: str, sources_content: List[Dict[str, str]]) -> Dict[str, Union[str, List[str]]]:
"""Verifies a claim against provided source content.
Args:
claim (str): The statement or piece of information to verify.
sources_content (List[Dict[str, str]]): A list of dictionaries, each with "url" (optional) and "content" keys.
Returns:
Dict: A dictionary summarizing findings (supporting, contradicting, inconclusive) per source.
"""
logger.info(f"Cross-referencing claim: {claim[:100]}... against {len(sources_content)} sources.")
if not sources_content:
return {"error": "No source content provided for cross-referencing."}
# LLM configuration
llm_model = os.getenv("VALIDATION_LLM_MODEL", "gemini-2.5-pro-preview-03-25") # Use a capable model
gemini_api_key = os.getenv("GEMINI_API_KEY")
if not gemini_api_key:
logger.error("GEMINI_API_KEY not found for cross-referencing LLM.")
return {"error": "GEMINI_API_KEY not set."}
results = []
try:
llm = GoogleGenAI(api_key=gemini_api_key, model="gemini-2.5-pro-preview-03-25", temperature=0.05)
logger.info(f"Using cross-referencing LLM: {llm_model}")
for i, source in enumerate(sources_content):
source_url = source.get("url", f"Source {i+1}")
content = source.get("content", "")
if not content:
logger.warning(f"Source {source_url} has no content.")
results.append({"source": source_url, "finding": "inconclusive", "reason": "No content provided"})
continue
# Truncate long content
max_content_len = 15000
if len(content) > max_content_len:
logger.warning(f"Truncating content from {source_url} to {max_content_len} chars.")
content = content[:max_content_len]
prompt = (
f"Review the following source content and determine if it supports, "
f"contradicts, or is inconclusive regarding the claim.\n\n"
f"CLAIM: {claim}\n\n"
f"SOURCE CONTENT from {source_url}:\n{content}\n\n"
f"ANALYSIS: Does the source content directly support the claim, directly contradict it, "
f"or provide no relevant information (inconclusive)? "
f"Provide a brief reason for your conclusion. Respond in JSON format: "
f'{{"finding": "support/contradict/inconclusive", "reason": "Your brief explanation"}}'
)
response = llm.complete(prompt)
try:
# Attempt to parse JSON, handle potential markdown fences
json_str = response.text.strip()
if json_str.startswith("```json"):
json_str = json_str[7:]
if json_str.endswith("```"):
json_str = json_str[:-3]
finding_data = json.loads(json_str.strip())
results.append({
"source": source_url,
"finding": finding_data.get("finding", "error"),
"reason": finding_data.get("reason", "LLM response parsing failed")
})
except json.JSONDecodeError:
logger.error(f"Failed to parse JSON response for source {source_url}: {response.text}")
results.append({"source": source_url, "finding": "error", "reason": "LLM response not valid JSON"})
except Exception as parse_err:
logger.error(f"Error processing LLM response for source {source_url}: {parse_err}")
results.append({"source": source_url, "finding": "error", "reason": f"Processing error: {parse_err}"})
logger.info("Cross-referencing check completed.")
return {"claim": claim, "results": results}
except Exception as e:
logger.error(f"LLM call failed during cross-referencing: {e}", exc_info=True)
return {"error": f"Error during cross-referencing: {e}"}
def logical_consistency_check(text: str) -> Dict[str, Union[bool, str, List[str]]]:
"""Analyzes text for internal logical contradictions or fallacies using an LLM."""
logger.info(f"Checking logical consistency for text (length: {len(text)} chars).")
# LLM configuration
llm_model = os.getenv("VALIDATION_LLM_MODEL", "gemini-2.5-pro-preview-03-25")
gemini_api_key = os.getenv("GEMINI_API_KEY")
if not gemini_api_key:
logger.error("GEMINI_API_KEY not found for consistency check LLM.")
return {"error": "GEMINI_API_KEY not set."}
# Truncate long text
max_input_chars = 30000
if len(text) > max_input_chars:
logger.warning(f"Input text truncated to {max_input_chars} chars for consistency check.")
text = text[:max_input_chars]
prompt = (
f"Analyze the following text for logical consistency. Identify any internal contradictions, "
f"logical fallacies, or significant inconsistencies in reasoning. "
f"If the text is logically consistent, state that clearly. If inconsistencies are found, "
f"list them with brief explanations.\n\n"
f"TEXT:\n{text}\n\n"
f"ANALYSIS: Respond in JSON format: "
f'{{"consistent": true/false, "findings": ["Description of inconsistency 1", "Description of inconsistency 2", ...]}}'
f"(If consistent is true, findings should be an empty list)."
)
try:
llm = GoogleGenAI(api_key=gemini_api_key, model="gemini-2.5-pro-preview-03-25", temperature=0.05)
logger.info(f"Using consistency check LLM: {llm_model}")
response = llm.complete(prompt)
# Attempt to parse JSON
json_str = response.text.strip()
if json_str.startswith("```json"):
json_str = json_str[7:]
if json_str.endswith("```"):
json_str = json_str[:-3]
result_data = json.loads(json_str.strip())
# Basic validation
if "consistent" not in result_data or "findings" not in result_data:
raise ValueError("LLM response missing required keys: consistent, findings")
if not isinstance(result_data["findings"], list):
raise ValueError("LLM response findings key is not a list")
logger.info(f"Logical consistency check completed. Consistent: {result_data.get('consistent')}")
return result_data
except json.JSONDecodeError as json_err:
logger.error(f"Failed to parse JSON response from LLM: {json_err}. Response text: {response.text}")
return {"error": f"Failed to parse LLM JSON response: {json_err}"}
except ValueError as val_err:
logger.error(f"Invalid JSON structure from LLM: {val_err}. Response text: {response.text}")
return {"error": f"Invalid JSON structure from LLM: {val_err}"}
except Exception as e:
logger.error(f"LLM call failed during consistency check: {e}", exc_info=True)
return {"error": f"Error during consistency check: {e}"}
def bias_detection(text: str, source_context: Optional[str] = None) -> Dict[str, Union[bool, List[Dict[str, str]]]]:
"""Examines text for potential biases using an LLM, considering source context if provided."""
logger.info(f"Detecting bias in text (length: {len(text)} chars). Context provided: {source_context is not None}")
# LLM configuration
llm_model = os.getenv("VALIDATION_LLM_MODEL", "gemini-2.5-pro-preview-03-25")
gemini_api_key = os.getenv("GEMINI_API_KEY")
if not gemini_api_key:
logger.error("GEMINI_API_KEY not found for bias detection LLM.")
return {"error": "GEMINI_API_KEY not set."}
# Truncate long text/context
max_input_chars = 25000
if len(text) > max_input_chars:
logger.warning(f"Input text truncated to {max_input_chars} chars for bias detection.")
text = text[:max_input_chars]
if source_context and len(source_context) > 5000:
logger.warning(f"Source context truncated to 5000 chars for bias detection.")
source_context = source_context[:5000]
context_prompt = f"\nSOURCE CONTEXT (optional background about the source):\n{source_context}" if source_context else ""
prompt = (
f"Analyze the following text for potential cognitive and presentation biases (e.g., confirmation bias, framing, selection bias, loaded language, appeal to emotion). "
f"Consider the language, tone, and selection of information. Also consider the source context if provided. "
f"If no significant biases are detected, state that clearly. If biases are found, list them, identify the type of bias, and provide a brief explanation with evidence from the text.\n\n"
f"TEXT:\n{text}"
f"{context_prompt}\n\n"
f"ANALYSIS: Respond in JSON format: "
f'{{"bias_detected": true/false, "findings": [{{"bias_type": "Type of Bias", "explanation": "Explanation with evidence"}}, ...]}}'
f"(If bias_detected is false, findings should be an empty list)."
)
try:
llm = GoogleGenAI(api_key=gemini_api_key, model="gemini-2.5-pro-preview-03-25", temperature=0.05)
logger.info(f"Using bias detection LLM: {llm_model}")
response = llm.complete(prompt)
# Attempt to parse JSON
json_str = response.text.strip()
if json_str.startswith("```json"):
json_str = json_str[7:]
if json_str.endswith("```"):
json_str = json_str[:-3]
result_data = json.loads(json_str.strip())
# Basic validation
if "bias_detected" not in result_data or "findings" not in result_data:
raise ValueError("LLM response missing required keys: bias_detected, findings")
if not isinstance(result_data["findings"], list):
raise ValueError("LLM response findings key is not a list")
logger.info(f"Bias detection check completed. Bias detected: {result_data.get('bias_detected')}")
return result_data
except json.JSONDecodeError as json_err:
logger.error(f"Failed to parse JSON response from LLM: {json_err}. Response text: {response.text}")
return {"error": f"Failed to parse LLM JSON response: {json_err}"}
except ValueError as val_err:
logger.error(f"Invalid JSON structure from LLM: {val_err}. Response text: {response.text}")
return {"error": f"Invalid JSON structure from LLM: {val_err}"}
except Exception as e:
logger.error(f"LLM call failed during bias detection: {e}", exc_info=True)
return {"error": f"Error during bias detection: {e}"}
# Note: fact_check_with_search primarily prepares the request for research_agent.
def fact_check_with_search(claim: str) -> Dict[str, str]:
"""Prepares a request to fact-check a specific claim using external search.
This tool does not perform the search itself but structures the request
for handoff to the research_agent.
Args:
claim (str): The specific factual claim to be checked.
Returns:
Dict: A dictionary indicating the need for handoff and the query.
"""
logger.info(f"Preparing fact-check request for claim: {claim[:150]}...")
# This tool signals the need for handoff to the research agent.
# The agent's prompt should guide it to use this tool's output
# to formulate the handoff message/query.
return {
"action": "handoff",
"target_agent": "research_agent",
"query": f"Fact-check the following claim: {claim}. Provide supporting or contradicting evidence from reliable sources.",
"tool_name": "fact_check_with_search" # For context
}
# --- Tool Definitions ---
cross_reference_tool = FunctionTool.from_defaults(
fn=cross_reference_check,
name="cross_reference_check",
description=(
"Verifies a claim against a list of provided source contents (text). "
"Input: claim (str), sources_content (List[Dict[str, str]] with 'content' key). "
"Output: Dict summarizing findings per source or error."
),
)
logical_consistency_tool = FunctionTool.from_defaults(
fn=logical_consistency_check,
name="logical_consistency_check",
description=(
"Analyzes text for internal logical contradictions or fallacies. "
"Input: text (str). Output: Dict with 'consistent' (bool) and 'findings' (List[str]) or error."
),
)
bias_detection_tool = FunctionTool.from_defaults(
fn=bias_detection,
name="bias_detection",
description=(
"Examines text for potential biases (cognitive, presentation). "
"Input: text (str), Optional: source_context (str). "
"Output: Dict with 'bias_detected' (bool) and 'findings' (List[Dict]) or error."
),
)
fact_check_tool = FunctionTool.from_defaults(
fn=fact_check_with_search,
name="fact_check_with_search",
description=(
"Prepares a request to fact-check a specific claim using external search via the research_agent. "
"Input: claim (str). Output: Dict indicating handoff parameters for research_agent."
),
)
# --- Agent Initialization ---
def initialize_advanced_validation_agent() -> ReActAgent:
"""Initializes the Advanced Validation Agent."""
logger.info("Initializing AdvancedValidationAgent...")
# Configuration for the agent's main LLM
agent_llm_model = os.getenv("VALIDATION_AGENT_LLM_MODEL", "gemini-2.5-pro-preview-03-25") # Use Pro for main agent logic
gemini_api_key = os.getenv("GEMINI_API_KEY")
if not gemini_api_key:
logger.error("GEMINI_API_KEY not found for AdvancedValidationAgent.")
raise ValueError("GEMINI_API_KEY must be set for AdvancedValidationAgent")
try:
llm = GoogleGenAI(api_key=gemini_api_key, model="gemini-2.5-pro-preview-03-25", temperature=0.05)
logger.info(f"Using agent LLM: {agent_llm_model}")
# Load system prompt
default_system_prompt = ("You are AdvancedValidationAgent... [Default prompt content - replace with actual]" # Placeholder
)
system_prompt = load_prompt_from_file("../prompts/advanced_validation_agent_prompt.txt", default_system_prompt)
if system_prompt == default_system_prompt:
logger.warning("Using default/fallback system prompt for AdvancedValidationAgent.")
# Define available tools
tools = [
cross_reference_tool,
logical_consistency_tool,
bias_detection_tool,
fact_check_tool # Tool to initiate handoff for external search
]
# Define valid handoff targets
valid_handoffs = [
"research_agent", # For fact-checking requiring external search
"planner_agent", # To return results
"reasoning_agent" # To return results
]
agent = ReActAgent(
name="advanced_validation_agent",
description=(
"Critically evaluates information for accuracy, consistency, and bias using specialized tools. "
"Can cross-reference claims, check logic, detect bias, and initiate external fact-checks via research_agent."
),
tools=tools,
llm=llm,
system_prompt=system_prompt,
can_handoff_to=valid_handoffs,
)
logger.info("AdvancedValidationAgent initialized successfully.")
return agent
except Exception as e:
logger.error(f"Error during AdvancedValidationAgent initialization: {e}", exc_info=True)
raise
# Example usage (for testing if run directly)
if __name__ == "__main__":
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
logger.info("Running advanced_validation_agent.py directly for testing...")
# Check required keys
required_keys = ["GEMINI_API_KEY"]
missing_keys = [key for key in required_keys if not os.getenv(key)]
if missing_keys:
print(f"Error: Required environment variable(s) not set: {', '.join(missing_keys)}. Cannot run test.")
else:
try:
# Test cross-reference tool
print("\nTesting cross_reference_check...")
test_claim = "The Eiffel Tower is located in Berlin."
test_sources = [
{"url": "wiki/paris", "content": "Paris is the capital of France, known for the Eiffel Tower."},
{"url": "wiki/berlin", "content": "Berlin is the capital of Germany, featuring the Brandenburg Gate."}
]
cross_ref_result = cross_reference_check(test_claim, test_sources)
print(f"Cross-reference Result:\n{json.dumps(cross_ref_result, indent=2)}")
# Test logical consistency tool
print("\nTesting logical_consistency_check...")
inconsistent_text = "All birds can fly. Penguins are birds. Therefore, penguins can fly."
consistency_result = logical_consistency_check(inconsistent_text)
print(f"Consistency Result:\n{json.dumps(consistency_result, indent=2)}")
# Test bias detection tool
print("\nTesting bias_detection...")
biased_text = "The revolutionary new policy is clearly the only sensible path forward, despite what uninformed critics might claim."
bias_result = bias_detection(biased_text)
print(f"Bias Detection Result:\n{json.dumps(bias_result, indent=2)}")
# Test fact_check tool (prepares handoff)
print("\nTesting fact_check_with_search...")
fact_check_prep = fact_check_with_search("Is the Earth flat?")
print(f"Fact Check Prep Result:\n{json.dumps(fact_check_prep, indent=2)}")
# Initialize the agent (optional)
# test_agent = initialize_advanced_validation_agent()
# print("\nAdvanced Validation Agent initialized successfully for testing.")
except Exception as e:
print(f"Error during testing: {e}")
|