Spaces:
Paused
Paused
File size: 18,038 Bytes
f39464e | 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 | """
workflow_engine.py β Motor de workflow YAWL-inspired para a Trindade Pipeline.
ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
Conceitos YAWL implementados
βββββββββββββββββββββββββββββ
Task : unidade atΓ΄mica de execuΓ§Γ£o (um turno de LLM ou operaΓ§Γ£o)
Condition : expressΓ£o Python avaliada contra o ExecutionContext
XOR-split : exatamente um arco de saΓda dispara (primeira condiΓ§Γ£o verdadeira)
AND-split : todos os arcos disparam (execuΓ§Γ£o paralela β futuro)
OR-split : um ou mais arcos disparam (subconjunto verdadeiro)
Loop : task repete enquanto condiΓ§Γ£o for verdadeira
SubNet : task que encapsula um sub-workflow inteiro
Fluxo de controle
ββββββββββββββββββ
WorkflowEngine.start() β retorna a primeira TaskDef
WorkflowEngine.advance(ctx) β avalia transiΓ§Γ΅es do nΓ³ atual, retorna prΓ³ximo
WorkflowEngine.is_terminal() β True se chegou ao nΓ³ END
O ExecutionContext Γ© um dict simples que o pipeline preenche apΓ³s cada task.
O motor lΓͺ esse dict para avaliar as condiΓ§Γ΅es β nΓ£o conhece LLMs, HTTP, nem JSON.
SeguranΓ§a do eval
ββββββββββββββββββ
CondiΓ§Γ΅es sΓ£o avaliadas com eval() em namespace restrito:
β’ Apenas builtins seguros (len, int, float, bool, str, min, max, abs, round)
β’ Mais todas as chaves do ExecutionContext
β’ Nenhum acesso a __import__, open, os, sys, etc.
ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
"""
from __future__ import annotations
import copy
from dataclasses import dataclass, field
from enum import Enum
from typing import Any, Dict, List, Optional
import yaml
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# TIPOS DE SPLIT
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
class SplitType(str, Enum):
XOR = "XOR" # exatamente um arco (default)
AND = "AND" # todos os arcos (paralelo)
OR = "OR" # um ou mais arcos
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# TRANSIΓΓO (arco de saΓda de um nΓ³)
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
@dataclass
class Transition:
target: str
condition: Optional[str] = None # None = default/unconditional
label: str = ""
def evaluate(self, ctx: Dict[str, Any]) -> bool:
"""Avalia a condiΓ§Γ£o contra o contexto. None β sempre True."""
if self.condition is None:
return True
return _safe_eval(self.condition, ctx)
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# TASK DEFINITION
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
@dataclass
class TaskDef:
"""
DefiniΓ§Γ£o declarativa de uma task no workflow.
Campos relevantes para o executor (pipeline_v33.py):
id : identificador ΓΊnico do nΓ³
task_type : "reasoning" | "final" | "audit_reasoning" |
"audit_final" | "normalize" | "validate" |
"loop_recovery" | "end"
phase : chave em PAYLOAD_CONFIG (ex. "STEP1", "AUDIT_REASONING")
provider : slug do provider (ex. "groq", "openrouter")
model : model id (sobrescreve o default do PAYLOAD_CONFIG)
stop_tokens : se True, usa STOP_TOKENS_REASONING
is_final : agente final (escreve JSON completo)
loop : TaskLoop se esta task pode ser repetida
O executor nΓ£o interpreta nenhum outro campo β passa como `task.params`.
"""
id: str
task_type: str
phase: str
provider: str = "groq"
model: Optional[str] = None
stop_tokens: bool = False
is_final: bool = False
loop: Optional["TaskLoop"] = None
transitions: List[Transition] = field(default_factory=list)
params: Dict[str, Any] = field(default_factory=dict)
description: str = ""
prompt: Optional[str] = None
instructions: Optional[str] = None
output_keys: List[str] = field(default_factory=list)
@property
def is_terminal(self) -> bool:
return self.task_type == "end"
def next_tasks(self, ctx: Dict[str, Any], split: SplitType = SplitType.XOR) -> List[str]:
"""Retorna ids dos prΓ³ximos nΓ³s dado o contexto atual."""
results: List[str] = []
for t in self.transitions:
if t.evaluate(ctx):
results.append(t.target)
if split == SplitType.XOR:
break # XOR: para no primeiro verdadeiro
return results
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# LOOP DEFINITION
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
@dataclass
class TaskLoop:
"""
Define comportamento de loop para uma task.
while_condition : expressΓ£o avaliada ANTES de cada iteraΓ§Γ£o
until_condition : expressΓ£o avaliada APΓS cada iteraΓ§Γ£o (do-while)
max_iterations : teto de seguranΓ§a (evita loop infinito)
"""
while_condition: Optional[str] = None
until_condition: Optional[str] = None
max_iterations: int = 3
def should_enter(self, ctx: Dict[str, Any]) -> bool:
if self.while_condition:
return _safe_eval(self.while_condition, ctx)
return True
def should_continue(self, ctx: Dict[str, Any], iteration: int) -> bool:
if iteration >= self.max_iterations:
return False
if self.until_condition:
return not _safe_eval(self.until_condition, ctx)
if self.while_condition:
return _safe_eval(self.while_condition, ctx)
return False
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# EXECUTION CONTEXT
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
class ExecutionContext(dict):
"""
Dict especializado que representa o estado de execuΓ§Γ£o do workflow.
O motor lΓͺ este dict para avaliar condiΓ§Γ΅es.
O executor escreve nele apΓ³s cada task.
Chaves convencionais (o motor conhece)
βββββββββββββββββββββββββββββββββββββββ
reasoning_chars : int β chars de reasoning da ΓΊltima fase intermediΓ‘ria
json_valid : bool β True se o JSON final passou parse
audit_passed : bool β True se o ΓΊltimo ciclo de audit retornou AUDIT_PASS
loop_count : int β tentativas de recovery de loop
atom_count : int β total de Γ‘tomos extraΓdos
correction_count : int β ciclos de correΓ§Γ£o do audit
schema_errors : int β erros de schema apΓ³s validate_and_fix
phase_failed : bool β True se a fase atual falhou (content=None)
current_phase : str β id do nΓ³ em execuΓ§Γ£o
"""
def update_phase(self, phase_id: str, **kwargs: Any) -> None:
self["current_phase"] = phase_id
self.update(kwargs)
def increment(self, key: str, by: int = 1) -> int:
self[key] = self.get(key, 0) + by
return self[key]
def snapshot(self) -> Dict[str, Any]:
return dict(self)
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# WORKFLOW ENGINE
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
class WorkflowEngine:
"""
MΓ‘quina de estados finitos YAWL-inspired.
Uso tΓpico no pipeline_v33.py
βββββββββββββββββββββββββββββββ
engine = WorkflowEngine.from_yaml("trindade_workflow.yaml")
ctx = ExecutionContext(defaults)
task = engine.start()
while not task.is_terminal:
result = executor.run(task, memory)
ctx.update(result)
task = engine.advance(ctx)
"""
def __init__(self, tasks: Dict[str, TaskDef], start_id: str) -> None:
self._tasks = tasks
self._start_id = start_id
self._current = start_id
self._wrapper_key = "output" # sobrescrito por from_dict
self._id_key = "id" # sobrescrito por from_dict
@property
def wrapper_key(self) -> str:
"""Chave de wrapper do resultado final (ex: 'manifestacao_juridica')."""
return self._wrapper_key
@property
def id_key(self) -> str:
"""Chave de id dentro do wrapper (ex: 'id_manifestacao')."""
return self._id_key
# ββ NavegaΓ§Γ£o βββββββββββββββββββββββββββββββββββββββββββββββββ
def start(self) -> TaskDef:
self._current = self._start_id
return self._tasks[self._current]
def current(self) -> TaskDef:
return self._tasks[self._current]
def advance(self, ctx: ExecutionContext) -> TaskDef:
"""
Avalia as transiΓ§Γ΅es do nΓ³ atual e move para o prΓ³ximo.
Retorna a TaskDef do prΓ³ximo nΓ³ (pode ser END).
"""
current_task = self._tasks[self._current]
split_type = SplitType(current_task.params.get("split", SplitType.XOR.value))
next_ids = current_task.next_tasks(ctx, split=split_type)
if not next_ids:
# Sem transiΓ§Γ£o β END implΓcito
self._current = "__END__"
return TaskDef(id="__END__", task_type="end", phase="END")
# Para XOR e OR, executa o primeiro target (AND Γ© futuro)
next_id = next_ids[0]
if next_id not in self._tasks:
raise KeyError(f"NΓ³ '{next_id}' referenciado mas nΓ£o definido no workflow")
self._current = next_id
ctx["current_phase"] = next_id
return self._tasks[next_id]
def peek_next(self, ctx: ExecutionContext) -> Optional[str]:
"""Retorna o id do prΓ³ximo nΓ³ sem avanΓ§ar o estado."""
task = self._tasks[self._current]
ids = task.next_tasks(ctx)
return ids[0] if ids else None
def reset(self) -> None:
self._current = self._start_id
# ββ Factory βββββββββββββββββββββββββββββββββββββββββββββββββββ
@classmethod
def from_yaml(cls, path: str) -> "WorkflowEngine":
"""Carrega um workflow de um arquivo YAML."""
with open(path, encoding="utf-8") as f:
spec = yaml.safe_load(f)
return cls.from_dict(spec)
@classmethod
def from_dict(cls, spec: Dict[str, Any]) -> "WorkflowEngine":
"""ConstrΓ³i o engine a partir de um dict (jΓ‘ parseado)."""
wf = spec["workflow"]
start_id = wf["start"]
tasks: Dict[str, TaskDef] = {}
# ββ Defaults do workflow (base para todos os tasks) ββββββββ
wf_defaults = wf.get("defaults", {})
# Campos top-level da TaskDef β nΓ£o vΓ£o para params
_top_level = {"id", "type", "phase", "provider", "model", "stop_tokens",
"is_final", "loop", "transitions", "description",
"prompt", "instructions", "output_keys", "params"}
# ββ ConfiguraΓ§Γ£o de saΓda agnΓ³stica ββββββββββββββββββββββββ
out_cfg = wf.get("output", {})
wrapper_key = out_cfg.get("wrapper_key", "output")
id_key = out_cfg.get("id_key", "id")
for raw in wf["tasks"]:
task_id = raw["id"]
# ββ Loop ββββββββββββββββββββββββββββββββββββββββββββββ
loop = None
if "loop" in raw:
lraw = raw["loop"]
loop = TaskLoop(
while_condition = lraw.get("while"),
until_condition = lraw.get("until"),
max_iterations = lraw.get("max_iterations", 3),
)
# ββ Transitions βββββββββββββββββββββββββββββββββββββββ
transitions = []
for t in raw.get("transitions", []):
transitions.append(Transition(
target = t["target"],
condition = t.get("condition"),
label = t.get("label", ""),
))
# ββ Params: defaults do YAML β sobrescritos por task ββ
params = {k: v for k, v in wf_defaults.items() if k not in _top_level}
params.update({k: v for k, v in raw.items() if k not in _top_level})
tasks[task_id] = TaskDef(
id = task_id,
task_type = raw.get("type", "task"),
phase = raw.get("phase", task_id),
provider = raw.get("provider", wf_defaults.get("provider", "groq")),
model = raw.get("model", wf_defaults.get("model")),
stop_tokens = raw.get("stop_tokens", False),
is_final = raw.get("is_final", False),
loop = loop,
transitions = transitions,
params = params,
description = raw.get("description", ""),
prompt = raw.get("prompt"),
instructions = raw.get("instructions"),
output_keys = raw.get("output_keys", []),
)
engine = cls(tasks=tasks, start_id=start_id)
engine._wrapper_key = wrapper_key
engine._id_key = id_key
return engine
# ββ InspeΓ§Γ£o ββββββββββββββββββββββββββββββββββββββββββββββββββ
def task_ids(self) -> List[str]:
return list(self._tasks.keys())
def describe(self) -> str:
"""Retorna representaΓ§Γ£o textual do grafo para debug."""
lines = [f"WorkflowEngine start={self._start_id} nodes={len(self._tasks)}"]
for tid, t in self._tasks.items():
arrow = " β ".join(
f"{tr.target}[{tr.condition or 'default'}]"
for tr in t.transitions
) or "(terminal)"
lines.append(f" {tid:30s} ({t.task_type:20s}) {arrow}")
return "\n".join(lines)
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# SAFE EVAL β avalia condiΓ§Γ΅es em namespace restrito
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
_SAFE_BUILTINS = {
"len": len,
"int": int,
"float": float,
"bool": bool,
"str": str,
"min": min,
"max": max,
"abs": abs,
"round": round,
"True": True,
"False": False,
"None": None,
"all": all,
"any": any,
}
def _safe_eval(expr: str, ctx: Dict[str, Any]) -> bool:
"""
Avalia `expr` (string Python) contra `ctx`.
Namespace: builtins seguros + ctx.
Qualquer exceΓ§Γ£o β False (condiΓ§Γ£o nΓ£o satisfeita).
Exemplos vΓ‘lidos de expressΓ΅es:
"reasoning_chars > 100"
"audit_passed == True"
"loop_count < 2 and not phase_failed"
"atom_count >= 3"
"schema_errors == 0"
"""
namespace = {**_SAFE_BUILTINS, **ctx}
try:
result = eval(expr, {"__builtins__": {}}, namespace) # noqa: S307
return bool(result)
except Exception:
return False
|