File size: 16,653 Bytes
5008b9d | 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 | """
ReAct Agent Framework β Reasoning and Acting
=============================================
Implements the ReAct prompting technique (Yao et al., 2023) for all
agent interactions in the AgentIC multi-agent pipeline.
ReAct Pattern:
Thought β Action β Observation β Thought β Action β ...
Each agent step follows this loop:
1. THOUGHT: Reason about the current state and what needs to happen
2. ACTION: Choose and execute one of the available tools
3. OBSERVATION: Observe the result of the action
4. Repeat until the task is complete or max steps reached
This replaces ad-hoc LLM prompting with structured, traceable reasoning.
"""
import json
import re
import time
import logging
from dataclasses import dataclass, field, asdict
from typing import Any, Callable, Dict, List, Optional, Tuple
from enum import Enum
logger = logging.getLogger(__name__)
# βββ Data Structures βββββββββββββββββββββββββββββββββββββββββββββββββ
class StepStatus(Enum):
THOUGHT = "THOUGHT"
ACTION = "ACTION"
OBSERVATION = "OBSERVATION"
FINAL_ANSWER = "FINAL_ANSWER"
ERROR = "ERROR"
@dataclass
class ReActStep:
"""A single step in the ReAct reasoning chain."""
step_num: int
status: StepStatus
content: str
action_name: str = ""
action_input: str = ""
observation: str = ""
timestamp: float = field(default_factory=time.time)
duration_s: float = 0.0
def to_dict(self) -> dict:
return {
"step": self.step_num,
"status": self.status.value,
"content": self.content,
"action_name": self.action_name,
"action_input": self.action_input[:500],
"observation": self.observation[:500],
}
@dataclass
class ReActTrace:
"""Complete trace of a ReAct agent run."""
task_description: str
steps: List[ReActStep] = field(default_factory=list)
final_answer: str = ""
success: bool = False
total_steps: int = 0
total_duration_s: float = 0.0
error: str = ""
def to_json(self) -> str:
return json.dumps({
"task": self.task_description[:200],
"success": self.success,
"total_steps": self.total_steps,
"total_duration_s": round(self.total_duration_s, 2),
"steps": [s.to_dict() for s in self.steps],
"final_answer": self.final_answer[:2000],
"error": self.error,
}, indent=2)
# βββ Tool Registry βββββββββββββββββββββββββββββββββββββββββββββββββββ
@dataclass
class ToolDef:
"""Definition of a tool available to the ReAct agent."""
name: str
description: str
function: Callable
parameters: Dict[str, str] = field(default_factory=dict) # param_name β description
class ToolRegistry:
"""Registry of tools available to ReAct agents."""
def __init__(self):
self._tools: Dict[str, ToolDef] = {}
def register(self, name: str, description: str, func: Callable,
parameters: Optional[Dict[str, str]] = None):
self._tools[name] = ToolDef(
name=name,
description=description,
function=func,
parameters=parameters or {},
)
def get(self, name: str) -> Optional[ToolDef]:
return self._tools.get(name)
def list_tools(self) -> str:
"""Format tools for the ReAct prompt."""
lines = []
for name, tool in self._tools.items():
params = ", ".join(f"{k}: {v}" for k, v in tool.parameters.items())
lines.append(f" {name}({params}) β {tool.description}")
return "\n".join(lines)
def execute(self, name: str, input_str: str) -> str:
"""Execute a tool by name with the given input string."""
tool = self._tools.get(name)
if not tool:
return f"ERROR: Unknown tool '{name}'. Available: {', '.join(self._tools.keys())}"
try:
result = tool.function(input_str)
return str(result) if result is not None else "OK"
except Exception as e:
return f"ERROR: {name} failed: {str(e)}"
# βββ ReAct Prompt Templates ββββββββββββββββββββββββββββββββββββββββββ
REACT_SYSTEM_PROMPT = """\
You are an expert VLSI agent using the ReAct (Reasoning and Acting) framework.
On each turn you must output ONE of:
Thought: <your reasoning about the current state>
Action: <tool_name>(<input>)
Final Answer: <your complete answer>
RULES:
1. Always start with a Thought before taking any Action.
2. After each Action, wait for the Observation before your next Thought.
3. You MUST use the available tools β do not hallucinate tool outputs.
4. When you have enough information, produce a Final Answer.
5. Maximum {max_steps} steps β be efficient.
6. If an action fails, reason about WHY and try a different approach.
Available Tools:
{tools}
TASK: {task}
"""
REACT_OBSERVATION_PROMPT = """\
Observation: {observation}
Continue with your next Thought or provide your Final Answer.
"""
# βββ ReAct Agent ββββββββββββββββββββββββββββββββββββββββββββββββββββββ
class ReActAgent:
"""
General-purpose ReAct agent for the AgentIC pipeline.
Uses the ReAct (Reasoning + Acting) prompting technique to provide
structured, traceable reasoning for all agent interactions.
Usage:
agent = ReActAgent(llm, role="RTL Debugger")
agent.register_tool("syntax_check", "Check Verilog syntax", syntax_check_fn)
agent.register_tool("read_file", "Read a file", read_file_fn)
trace = agent.run("Fix the syntax error in counter.v")
print(trace.final_answer)
"""
def __init__(
self,
llm, # CrewAI LLM instance
role: str = "VLSI Agent",
max_steps: int = 10,
verbose: bool = False,
):
self.llm = llm
self.role = role
self.max_steps = max_steps
self.verbose = verbose
self.tools = ToolRegistry()
self._conversation: List[Dict[str, str]] = []
def register_tool(self, name: str, description: str, func: Callable,
parameters: Optional[Dict[str, str]] = None):
"""Register a tool available to this agent."""
self.tools.register(name, description, func, parameters)
def run(self, task: str, context: str = "") -> ReActTrace:
"""
Execute the ReAct loop for the given task.
Args:
task: Natural language task description
context: Additional context (RTL code, error logs, etc.)
Returns:
ReActTrace with complete reasoning chain and final answer.
"""
trace = ReActTrace(task_description=task)
start_time = time.time()
# Build system prompt
system_prompt = REACT_SYSTEM_PROMPT.format(
max_steps=self.max_steps,
tools=self.tools.list_tools(),
task=task,
)
if context:
system_prompt += f"\n\nCONTEXT:\n{context[:8000]}"
self._conversation = [{"role": "system", "content": system_prompt}]
step_num = 0
while step_num < self.max_steps:
step_num += 1
step_start = time.time()
# Get LLM response
try:
response = self._call_llm()
except Exception as e:
trace.steps.append(ReActStep(
step_num=step_num,
status=StepStatus.ERROR,
content=f"LLM call failed: {str(e)}",
))
trace.error = str(e)
break
# Parse the response
thought, action_name, action_input, final_answer = self._parse_response(response)
# Handle FINAL ANSWER
if final_answer:
trace.steps.append(ReActStep(
step_num=step_num,
status=StepStatus.FINAL_ANSWER,
content=final_answer,
duration_s=time.time() - step_start,
))
trace.final_answer = final_answer
trace.success = True
break
# Handle THOUGHT
if thought:
trace.steps.append(ReActStep(
step_num=step_num,
status=StepStatus.THOUGHT,
content=thought,
duration_s=time.time() - step_start,
))
if self.verbose:
logger.info(f"[ReAct:{self.role}] Thought: {thought[:200]}")
# Handle ACTION
if action_name:
# Execute the tool
observation = self.tools.execute(action_name, action_input)
trace.steps.append(ReActStep(
step_num=step_num,
status=StepStatus.ACTION,
content=f"{action_name}({action_input[:200]})",
action_name=action_name,
action_input=action_input,
observation=observation[:2000],
duration_s=time.time() - step_start,
))
if self.verbose:
logger.info(f"[ReAct:{self.role}] Action: {action_name} β {observation[:200]}")
# Feed observation back
obs_prompt = REACT_OBSERVATION_PROMPT.format(
observation=observation[:4000]
)
self._conversation.append({"role": "assistant", "content": response})
self._conversation.append({"role": "user", "content": obs_prompt})
elif not thought and not final_answer:
# LLM produced something unparseable β nudge it
self._conversation.append({"role": "assistant", "content": response})
self._conversation.append({
"role": "user",
"content": (
"Your response didn't follow the ReAct format. "
"Please respond with one of:\n"
" Thought: <reasoning>\n"
" Action: <tool_name>(<input>)\n"
" Final Answer: <answer>"
),
})
trace.total_steps = step_num
trace.total_duration_s = time.time() - start_time
if not trace.success:
trace.error = trace.error or "Max steps reached without Final Answer"
# Use last thought/action as fallback answer
for step in reversed(trace.steps):
if step.content:
trace.final_answer = step.content
break
return trace
def _call_llm(self) -> str:
"""Call the LLM with the current conversation."""
from crewai import Agent, Task, Crew
# Build a single prompt from conversation history
prompt_parts = []
for msg in self._conversation:
if msg["role"] == "system":
prompt_parts.append(msg["content"])
elif msg["role"] == "user":
prompt_parts.append(f"\n{msg['content']}")
elif msg["role"] == "assistant":
prompt_parts.append(f"\nAssistant: {msg['content']}")
full_prompt = "\n".join(prompt_parts)
agent = Agent(
role=self.role,
goal="Follow the ReAct framework to complete the task",
backstory=f"Expert {self.role} using structured ReAct reasoning.",
llm=self.llm,
verbose=False, # We handle our own logging
)
task = Task(
description=full_prompt[-12000:], # Truncate to fit context
expected_output="A Thought, Action, or Final Answer following ReAct format",
agent=agent,
)
result = str(Crew(agents=[agent], tasks=[task]).kickoff())
return result
def _parse_response(self, response: str) -> Tuple[str, str, str, str]:
"""
Parse a ReAct response into (thought, action_name, action_input, final_answer).
Returns empty strings for components not present in the response.
"""
thought = ""
action_name = ""
action_input = ""
final_answer = ""
# Check for Final Answer
fa_match = re.search(r'Final\s+Answer\s*:\s*(.+)', response, re.DOTALL | re.IGNORECASE)
if fa_match:
final_answer = fa_match.group(1).strip()
return thought, action_name, action_input, final_answer
# Check for Thought
th_match = re.search(r'Thought\s*:\s*(.+?)(?=Action\s*:|Final\s+Answer\s*:|$)',
response, re.DOTALL | re.IGNORECASE)
if th_match:
thought = th_match.group(1).strip()
# Check for Action
act_match = re.search(r'Action\s*:\s*(\w+)\s*\((.+?)\)\s*$',
response, re.MULTILINE | re.IGNORECASE)
if not act_match:
# Try alternative format: Action: tool_name\nAction Input: input
act_match2 = re.search(r'Action\s*:\s*(\w+)', response, re.IGNORECASE)
inp_match = re.search(r'Action\s+Input\s*:\s*(.+?)(?=\n|$)', response,
re.DOTALL | re.IGNORECASE)
if act_match2:
action_name = act_match2.group(1).strip()
action_input = inp_match.group(1).strip() if inp_match else ""
else:
action_name = act_match.group(1).strip()
action_input = act_match.group(2).strip()
return thought, action_name, action_input, final_answer
# βββ Pre-built ReAct Agents for AgentIC Pipeline βββββββββββββββββββββ
def create_rtl_debugger_agent(llm, tools_dict: Dict[str, Callable],
verbose: bool = False) -> ReActAgent:
"""Create a ReAct agent pre-configured for RTL debugging."""
agent = ReActAgent(llm, role="RTL Debugger", max_steps=8, verbose=verbose)
default_tools = {
"syntax_check": ("Check Verilog syntax of a file path", {}),
"read_file": ("Read contents of a file path", {}),
"run_simulation": ("Run Icarus Verilog simulation for a design name", {}),
"trace_signal": ("Back-trace a signal through the RTL AST", {}),
}
for name, func in tools_dict.items():
desc, params = default_tools.get(name, (f"Execute {name}", {}))
agent.register_tool(name, desc, func, params)
return agent
def create_formal_debugger_agent(llm, tools_dict: Dict[str, Callable],
verbose: bool = False) -> ReActAgent:
"""Create a ReAct agent pre-configured for formal verification debugging."""
agent = ReActAgent(llm, role="Formal Verification Debugger", max_steps=10, verbose=verbose)
default_tools = {
"run_formal": ("Run SymbiYosys formal verification on a .sby file", {}),
"read_file": ("Read contents of a file path", {}),
"analyze_signal": ("Run balanced for-and-against analysis on a signal", {}),
"build_causal_graph": ("Build causal graph from RTL and failure", {}),
}
for name, func in tools_dict.items():
desc, params = default_tools.get(name, (f"Execute {name}", {}))
agent.register_tool(name, desc, func, params)
return agent
def create_architect_agent(llm, tools_dict: Dict[str, Callable],
verbose: bool = False) -> ReActAgent:
"""Create a ReAct agent pre-configured for architectural decomposition."""
agent = ReActAgent(llm, role="Spec2RTL Architect", max_steps=6, verbose=verbose)
default_tools = {
"decompose_spec": ("Decompose a natural language spec into JSON SID", {}),
"validate_sid": ("Validate a Structured Information Dictionary", {}),
"read_spec": ("Read a specification file (text or PDF)", {}),
}
for name, func in tools_dict.items():
desc, params = default_tools.get(name, (f"Execute {name}", {}))
agent.register_tool(name, desc, func, params)
return agent
|