Spaces:
Running
Running
File size: 18,940 Bytes
c2ea5ed |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 |
"""
Boundary Detection Module for Agent-Aware Semantic Splitting
This module identifies semantic boundaries in agent execution logs
to enable intelligent chunking that preserves agent interaction integrity.
"""
import re
from enum import Enum
from typing import Dict, List, Optional, Tuple
from dataclasses import dataclass
from abc import ABC, abstractmethod
from .log_type_detector import LogType, LogTypeDetector
class BoundaryType(Enum):
"""Types of semantic boundaries in agent logs."""
CREW_START = "crew_start"
CREW_END = "crew_end"
TASK_START = "task_start"
TASK_END = "task_end"
AGENT_ASSIGNMENT = "agent_assignment"
TOOL_CYCLE_START = "tool_cycle_start"
TOOL_CYCLE_END = "tool_cycle_end"
THINKING_START = "thinking_start"
THINKING_END = "thinking_end"
FINAL_ANSWER = "final_answer"
HUMAN_FEEDBACK = "human_feedback"
JSON_OBJECT_START = "json_object_start"
JSON_OBJECT_END = "json_object_end"
TRACE_START = "trace_start"
TRACE_END = "trace_end"
OBSERVATION_START = "observation_start"
OBSERVATION_END = "observation_end"
SEMANTIC_BREAK = "semantic_break"
@dataclass
class AgentBoundary:
"""Represents a detected boundary in agent logs."""
position: int
boundary_type: BoundaryType
pattern_matched: str
confidence_score: float
context_before: str
context_after: str
metadata: Dict[str, any]
@dataclass
class BoundaryConfidence:
"""Confidence scoring for boundary detection."""
pattern_confidence: float = 0.0
semantic_confidence: float = 0.0
context_confidence: float = 0.0
combined_score: float = 0.0
def is_valid_boundary(self, threshold: float = 0.7) -> bool:
"""Check if boundary meets confidence threshold."""
return self.combined_score >= threshold
class BaseBoundaryDetector(ABC):
"""Abstract base class for boundary detectors."""
@abstractmethod
def detect_boundaries(self, content: str) -> List[AgentBoundary]:
"""Detect boundaries in the given content."""
pass
@abstractmethod
def get_priority(self) -> int:
"""Get the priority of this detector (lower = higher priority)."""
pass
class FrameworkSpecificDetector(BaseBoundaryDetector):
"""Detector for framework-specific patterns (CrewAI, Langfuse, etc.)."""
def __init__(self):
self.crewai_patterns = {
BoundaryType.CREW_START: [
r'โญ.*Crew Execution Started.*โฎ',
r'๐ Crew: .*'
],
BoundaryType.CREW_END: [
r'โญ.*Crew Completion.*โฎ',
r'Crew Execution Completed'
],
BoundaryType.TASK_START: [
r'โโโ ๐ Task: [a-f0-9-]+',
r'Status: Executing Task\.\.\.'
],
BoundaryType.TASK_END: [
r'Status: โ
Completed',
r'โญ.*Task Completion.*โฎ'
],
BoundaryType.AGENT_ASSIGNMENT: [
r'# Agent: .*',
r'โโโ ๐ค Agent: .*'
],
BoundaryType.TOOL_CYCLE_START: [
r'## Using tool: .*',
r'โโโ ๐ง Using .*'
],
BoundaryType.TOOL_CYCLE_END: [
r'## Tool Output:',
r'โโโ ๐ง Used .*'
],
BoundaryType.THINKING_START: [
r'โโโ ๐ง Thinking\.\.\.',
r'## Thinking\.\.\.'
],
BoundaryType.FINAL_ANSWER: [
r'## Final Answer:',
r'## Final Result:'
],
BoundaryType.HUMAN_FEEDBACK: [
r'## HUMAN FEEDBACK:',
r'=====\n## HUMAN FEEDBACK:'
]
}
self.langfuse_patterns = {
BoundaryType.TRACE_START: [
r'"data": \{\s*"id": "[a-f0-9-]+"',
r'"trace_id": "[a-f0-9-]+"'
],
BoundaryType.OBSERVATION_START: [
r'"observations": \[',
r'"type": "(SPAN|GENERATION)"'
],
BoundaryType.JSON_OBJECT_START: [
r'^\s*\{',
r'^\s*\['
],
BoundaryType.JSON_OBJECT_END: [
r'\}\s*$',
r'\]\s*$'
]
}
def detect_boundaries(self, content: str) -> List[AgentBoundary]:
"""Detect framework-specific boundaries."""
boundaries = []
# Detect CrewAI boundaries
boundaries.extend(self._detect_pattern_boundaries(
content, self.crewai_patterns, "CrewAI"
))
# Detect Langfuse boundaries
boundaries.extend(self._detect_pattern_boundaries(
content, self.langfuse_patterns, "Langfuse"
))
return sorted(boundaries, key=lambda b: b.position)
def _detect_pattern_boundaries(self, content: str, patterns: Dict, framework: str) -> List[AgentBoundary]:
"""Detect boundaries using pattern matching."""
boundaries = []
for boundary_type, pattern_list in patterns.items():
for pattern in pattern_list:
for match in re.finditer(pattern, content, re.MULTILINE):
start_pos = match.start()
# Get context around the boundary
context_size = 100
context_start = max(0, start_pos - context_size)
context_end = min(len(content), start_pos + len(match.group()) + context_size)
context_before = content[context_start:start_pos]
context_after = content[start_pos + len(match.group()):context_end]
# Calculate confidence based on pattern specificity
confidence = self._calculate_pattern_confidence(pattern, match.group())
boundary = AgentBoundary(
position=start_pos,
boundary_type=boundary_type,
pattern_matched=pattern,
confidence_score=confidence,
context_before=context_before,
context_after=context_after,
metadata={
"framework": framework,
"matched_text": match.group(),
"pattern_type": "regex"
}
)
boundaries.append(boundary)
return boundaries
def _calculate_pattern_confidence(self, pattern: str, matched_text: str) -> float:
"""Calculate confidence score for pattern match."""
# Base confidence from pattern specificity
base_confidence = 0.7
# Increase confidence for longer, more specific patterns
if len(pattern) > 30:
base_confidence += 0.1
if len(pattern) > 50:
base_confidence += 0.1
# Increase confidence for exact character matches (emojis, special chars)
if re.search(r'[๐๐๐ค๐ง๐ง โ
โญโฎ]', pattern):
base_confidence += 0.15
# Increase confidence for UUID patterns
if re.search(r'[a-f0-9-]{36}', matched_text):
base_confidence += 0.1
return min(base_confidence, 1.0)
def get_priority(self) -> int:
"""Framework-specific has highest priority."""
return 1
class GenericAgentPatternDetector(BaseBoundaryDetector):
"""Detector for generic agent patterns across frameworks."""
def __init__(self):
self.generic_patterns = {
BoundaryType.AGENT_ASSIGNMENT: [
r'Agent: .*',
r'Role: .*',
r'Assistant: .*'
],
BoundaryType.TOOL_CYCLE_START: [
r'Tool: .*',
r'Action: .*',
r'Function: .*'
],
BoundaryType.TOOL_CYCLE_END: [
r'Result: .*',
r'Output: .*',
r'Response: .*'
],
BoundaryType.THINKING_START: [
r'Thought: .*',
r'Thinking: .*',
r'Reasoning: .*'
],
BoundaryType.FINAL_ANSWER: [
r'Answer: .*',
r'Conclusion: .*',
r'Final: .*'
]
}
def detect_boundaries(self, content: str) -> List[AgentBoundary]:
"""Detect generic agent pattern boundaries."""
return self._detect_pattern_boundaries(content, self.generic_patterns, "Generic")
def _detect_pattern_boundaries(self, content: str, patterns: Dict, framework: str) -> List[AgentBoundary]:
"""Detect boundaries using generic patterns."""
boundaries = []
for boundary_type, pattern_list in patterns.items():
for pattern in pattern_list:
for match in re.finditer(pattern, content, re.MULTILINE):
start_pos = match.start()
context_size = 50
context_start = max(0, start_pos - context_size)
context_end = min(len(content), start_pos + len(match.group()) + context_size)
context_before = content[context_start:start_pos]
context_after = content[start_pos + len(match.group()):context_end]
confidence = 0.6 # Lower confidence for generic patterns
boundary = AgentBoundary(
position=start_pos,
boundary_type=boundary_type,
pattern_matched=pattern,
confidence_score=confidence,
context_before=context_before,
context_after=context_after,
metadata={
"framework": framework,
"matched_text": match.group(),
"pattern_type": "generic"
}
)
boundaries.append(boundary)
return boundaries
def get_priority(self) -> int:
"""Generic patterns have medium priority."""
return 2
class StructuralDetector(BaseBoundaryDetector):
"""Detector for structural boundaries (JSON, sections, etc.)."""
def detect_boundaries(self, content: str) -> List[AgentBoundary]:
"""Detect structural boundaries."""
boundaries = []
# Detect JSON object boundaries
boundaries.extend(self._detect_json_boundaries(content))
# Detect section headers
boundaries.extend(self._detect_section_boundaries(content))
return sorted(boundaries, key=lambda b: b.position)
def _detect_json_boundaries(self, content: str) -> List[AgentBoundary]:
"""Detect JSON object start/end boundaries."""
boundaries = []
brace_stack = []
bracket_stack = []
for i, char in enumerate(content):
if char == '{':
brace_stack.append(i)
if len(brace_stack) == 1: # Start of top-level object
boundary = AgentBoundary(
position=i,
boundary_type=BoundaryType.JSON_OBJECT_START,
pattern_matched="{",
confidence_score=0.8,
context_before=content[max(0, i-20):i],
context_after=content[i+1:min(len(content), i+21)],
metadata={"structure_type": "json_object"}
)
boundaries.append(boundary)
elif char == '}':
if brace_stack:
brace_stack.pop()
if len(brace_stack) == 0: # End of top-level object
boundary = AgentBoundary(
position=i,
boundary_type=BoundaryType.JSON_OBJECT_END,
pattern_matched="}",
confidence_score=0.8,
context_before=content[max(0, i-20):i],
context_after=content[i+1:min(len(content), i+21)],
metadata={"structure_type": "json_object"}
)
boundaries.append(boundary)
return boundaries
def _detect_section_boundaries(self, content: str) -> List[AgentBoundary]:
"""Detect section header boundaries."""
boundaries = []
# Markdown-style headers
header_pattern = r'^#+\s+.*$'
for match in re.finditer(header_pattern, content, re.MULTILINE):
boundary = AgentBoundary(
position=match.start(),
boundary_type=BoundaryType.SEMANTIC_BREAK,
pattern_matched=header_pattern,
confidence_score=0.7,
context_before=content[max(0, match.start()-30):match.start()],
context_after=content[match.end():min(len(content), match.end()+30)],
metadata={"structure_type": "section_header", "header_text": match.group()}
)
boundaries.append(boundary)
return boundaries
def get_priority(self) -> int:
"""Structural detection has lower priority."""
return 3
class BoundaryDetector:
"""Main boundary detection coordinator."""
def __init__(self, log_type_detector: Optional[LogTypeDetector] = None):
"""Initialize with optional log type detector."""
self.log_type_detector = log_type_detector or LogTypeDetector()
# Initialize detectors in priority order
self.detectors = [
FrameworkSpecificDetector(),
GenericAgentPatternDetector(),
StructuralDetector()
]
# Sort by priority
self.detectors.sort(key=lambda d: d.get_priority())
def detect_boundaries(self, content: str, log_type: Optional[LogType] = None) -> List[AgentBoundary]:
"""
Detect all boundaries in content using multi-layer approach.
Args:
content: The content to analyze
log_type: Optional pre-detected log type
Returns:
List of detected boundaries sorted by position
"""
if not log_type:
detection_result = self.log_type_detector.detect_log_type(content)
log_type = detection_result.log_type
all_boundaries = []
# Run all detectors
for detector in self.detectors:
try:
boundaries = detector.detect_boundaries(content)
all_boundaries.extend(boundaries)
except Exception as e:
# Log error but continue with other detectors
print(f"Warning: Detector {detector.__class__.__name__} failed: {e}")
# Remove duplicate boundaries (same position, similar type)
deduplicated = self._deduplicate_boundaries(all_boundaries)
# Sort by position
return sorted(deduplicated, key=lambda b: b.position)
def _deduplicate_boundaries(self, boundaries: List[AgentBoundary]) -> List[AgentBoundary]:
"""Remove duplicate boundaries that are too close to each other."""
if not boundaries:
return []
# Sort by position first
sorted_boundaries = sorted(boundaries, key=lambda b: b.position)
deduplicated = [sorted_boundaries[0]]
for boundary in sorted_boundaries[1:]:
# Check if this boundary is too close to the last one
last_boundary = deduplicated[-1]
position_diff = boundary.position - last_boundary.position
# If boundaries are very close (within 10 characters), keep the higher confidence one
if position_diff < 10:
if boundary.confidence_score > last_boundary.confidence_score:
deduplicated[-1] = boundary
else:
deduplicated.append(boundary)
return deduplicated
def calculate_boundary_confidence(self, boundary: AgentBoundary, content: str) -> BoundaryConfidence:
"""
Calculate comprehensive confidence score for a boundary.
Args:
boundary: The boundary to analyze
content: The full content for context analysis
Returns:
BoundaryConfidence with detailed scoring
"""
confidence = BoundaryConfidence()
# Pattern confidence (from initial detection)
confidence.pattern_confidence = boundary.confidence_score
# Context confidence (analyze surrounding content)
confidence.context_confidence = self._analyze_context_confidence(boundary, content)
# Semantic confidence (placeholder for embedding-based analysis)
confidence.semantic_confidence = 0.7 # Will be enhanced with semantic analyzer
# Combined score (weighted average)
weights = {"pattern": 0.4, "context": 0.3, "semantic": 0.3}
confidence.combined_score = (
confidence.pattern_confidence * weights["pattern"] +
confidence.context_confidence * weights["context"] +
confidence.semantic_confidence * weights["semantic"]
)
return confidence
def _analyze_context_confidence(self, boundary: AgentBoundary, content: str) -> float:
"""Analyze context around boundary to determine confidence."""
# Check for consistent formatting around boundary
context_window = boundary.context_before + boundary.context_after
confidence = 0.5 # Base confidence
# Check for consistent indentation/formatting
if re.search(r'\n\s*\n', context_window): # Blank lines suggest section breaks
confidence += 0.2
# Check for consistent agent markers in context
agent_markers = len(re.findall(r'(Agent:|Task:|Tool:|๐|๐ค|๐|๐ง)', context_window))
if agent_markers > 0:
confidence += 0.1
# Check for timestamp patterns
if re.search(r'\d{2}:\d{2}:\d{2}', context_window):
confidence += 0.1
return min(confidence, 1.0) |