Spaces:
Sleeping
Sleeping
File size: 18,220 Bytes
1ea26af |
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 |
#
import os
import re
import shutil
import urllib.request
from contextlib import contextmanager
from concurrent.futures import ThreadPoolExecutor
from ..agents.agent import MultiStepAgent, register_template, ActionResult
from ..agents.model import LLM
from ..agents.utils import zwarn, rprint, have_images_in_messages
from ..agents.tool import SimpleSearchTool
from .utils import WebEnv
from .playwright_utils import PlaywrightWebEnv
from .prompts import PROMPTS as WEB_PROMPTS
# --
# pre-defined actions: simply convert things to str
def web_click(id: int, link_name=""): return ActionResult(f"click [{id}] {link_name}")
def web_type(id: int, content: str, enter=True): return ActionResult(f"type [{id}] {content}" if enter else f"type [{id}] {content}[NOENTER]")
def web_scroll_up(): return ActionResult(f"scroll up")
def web_scroll_down(): return ActionResult(f"scroll down")
def web_wait(): return ActionResult(f"wait")
def web_goback(): return ActionResult(f"goback")
def web_restart(): return ActionResult(f"restart")
def web_goto(url: str): return ActionResult(f"goto {url}")
class ThreadedWebEnv:
"""A thin proxy that runs the builtin PlaywrightWebEnv entirely on a dedicated thread.
Ensures sync Playwright APIs never execute on an asyncio event-loop thread.
"""
def __init__(self, **kwargs):
self._executor = ThreadPoolExecutor(max_workers=1, thread_name_prefix="ck_web_env")
self._env = None
def _create():
# Import here so tests can monkeypatch ck_pro.ck_web.playwright_utils.PlaywrightWebEnv
from .playwright_utils import PlaywrightWebEnv as _PWE
return _PWE(**kwargs)
# Construct the real env on the dedicated thread
self._env = self._executor.submit(_create).result()
def _call(self, fn_name, *args, **kwargs):
def _invoke():
env = self._env
return getattr(env, fn_name)(*args, **kwargs)
return self._executor.submit(_invoke).result()
# Public methods used by WebAgent
def get_state(self):
return self._call("get_state")
def step_state(self, action_string: str) -> str:
return self._call("step_state", action_string)
def sync_files(self):
return self._call("sync_files")
def stop(self):
# Cleanup the underlying env on its own thread, then shutdown the executor
def _cleanup():
env = self._env
if env is not None:
try:
env.stop()
finally:
bp = getattr(env, "browser_pool", None)
if bp:
try:
bp.stop()
finally:
pass
self._env = None
try:
self._executor.submit(_cleanup).result()
finally:
self._executor.shutdown(wait=True)
# def web_stop(answer, summary): return ActionResult(f"stop [{answer}] ({summary})") # use self-defined function!
# --
class WebAgent(MultiStepAgent):
def __init__(self, settings=None, logger=None, **kwargs):
# note: this is a little tricky since things will get re-init again in super().__init__
feed_kwargs = dict(
name="web_agent",
description="A web agent helping to browse and operate web pages to solve a specific task.",
templates={"plan": "web_plan", "action": "web_action", "end": "web_end"}, # template names
max_steps=16,
)
feed_kwargs.update(kwargs)
self.logger = logger # 接收外部传入的日志器
self.settings = settings # Store settings reference
self.web_env_kwargs = {} # kwargs for web env
self.check_nodiff_steps = 3 # if for 3 steps, we have the same web page, then explicitly indicating this!
self.html_md_budget = 0 # budget in bytes (around 4 bytes per token, for example: 2K bytes ~ 500 tokens; 0 means not using this)
self.use_multimodal = "auto" # no: always no, yes: always yes, auto: let the agent decide
# Use same model config as main model for multimodal (if provided); otherwise lazy init
multimodal_kwargs = kwargs.get('model', {}).copy() if kwargs.get('model') else None
if multimodal_kwargs:
self.model_multimodal = LLM(**multimodal_kwargs)
else:
# Lazy/default init to avoid validation errors when not needed
self.model_multimodal = LLM(_default_init=True)
# Fuse mechanism is fully automatic - no manual configuration needed
# self.searcher = SimpleSearchTool(max_results=16, list_enum=False) # use more!
# --
register_template(WEB_PROMPTS) # add web prompts
super().__init__(**feed_kwargs)
self.web_envs = {} # session_id -> ENV
self.ACTIVE_FUNCTIONS.update(click=web_click, type=web_type, scroll_up=web_scroll_up, scroll_down=web_scroll_down, wait=web_wait, goback=web_goback, restart=web_restart, goto=web_goto)
# self.ACTIVE_FUNCTIONS.update(stop=self._my_stop, save=self._my_save, search=self._my_search)
self.ACTIVE_FUNCTIONS.update(stop=self._my_stop, save=self._my_save, screenshot=self._my_screenshot)
# --
# note: a specific stop function!
def _my_stop(self, answer: str = None, summary: str = None, output: str = None):
if output:
ret = f"Final answer: [{output}] ({summary})"
else:
ret = f"Final answer: [{answer}] ({summary})"
self.put_final_result(ret) # mark end and put final result
return ActionResult("stop", ret)
# note: special save
def _my_save(self, remote_path: str, local_path: str):
try:
_dir = os.path.dirname(local_path)
if _dir:
os.makedirs(_dir, exist_ok=True)
if local_path != remote_path:
remote_path = remote_path.strip()
if remote_path.startswith("http://") or remote_path.startswith("https://"): # retrieve from the web
urllib.request.urlretrieve(remote_path, local_path)
else: # simply copy!
shutil.copyfile(remote_path, local_path)
ret = f"Save Succeed: from remote_path = {remote_path} to local_path = {local_path}"
except Exception as e:
ret = f"Save Failed with {e}: from remote_path = {remote_path} to local_path = {local_path}"
return ActionResult("save", ret)
# note: whether use the screenshot mode
def _my_screenshot(self, flag: bool, save_path=""):
return ActionResult(f"screenshot {int(flag)} {save_path}")
def get_function_definition(self, short: bool):
if short:
return "- def web_agent(task: str, target_url: str = None) -> Dict: # Employs a web browser to navigate and interact with web pages to accomplish a specific task. Note that the web agent is limited to downloading files and cannot process or analyze them."
else:
return """- web_agent
```python
def web_agent(task: str) -> dict:
\""" Employs a web browser to navigate and interact with web pages to accomplish a specific task.
Args:
task (str): A detailed description of the task to perform. This may include:
- The target website(s) to visit (include valid URLs).
- Specific output formatting requirements.
- Instructions to download files (specify desired output path if needed).
Returns:
dict: A dictionary with the following structure:
{
'output': <str> # The well-formatted answer, strictly following any specified output format.
'log': <str> # Additional notes, such as steps taken, issues encountered, or relevant context.
}
Notes:
- If the `task` specifies an output format, ensure the 'output' field matches it exactly.
- The web agent can download files, but cannot process or analyze them. If file analysis is required, save the file to a local path and return control to an external planner or file agent for further processing.
Example:
>>> answer = web_agent(task="What is the current club of Messi? (Format your output directly as 'club_name'.)")
>>> print(answer) # directly print the full result dictionary
\"""
```"""
def __call__(self, task: str, **kwargs): # allow *args styled calling
return super().__call__(task, **kwargs)
def init_run(self, session):
super().init_run(session)
_id = session.id
assert _id not in self.web_envs
_kwargs = self.web_env_kwargs.copy()
if session.info.get("target_url"):
_kwargs["starting_target_url"] = session.info["target_url"]
_kwargs["logger"] = self.logger # 传递 logger 给 WebEnv
# 自动选择Web环境实现:优先HTTP API,失败则使用内置Playwright
web_ip = _kwargs.get("web_ip", "localhost:3000")
if self._test_web_ip_connection(web_ip):
if self.logger:
self.logger.info("[WEB_AGENT] Using HTTP API (web_ip: %s)", web_ip)
self.web_envs[_id] = WebEnv(**_kwargs)
else:
if self.logger:
self.logger.info("[WEB_AGENT] HTTP API unavailable, using builtin")
# 使用内置实现
builtin_kwargs = {k: v for k, v in _kwargs.items()
if k in ["starting_target_url", "logger", "headless", "max_browsers", "web_timeout"]}
# Run builtin PlaywrightWebEnv entirely on a dedicated thread to avoid asyncio-loop conflicts
self.web_envs[_id] = ThreadedWebEnv(**builtin_kwargs)
def _test_web_ip_connection(self, web_ip: str) -> bool:
"""测试web_ip连接性"""
try:
import requests
response = requests.get(f"http://{web_ip}/health", timeout=5)
return response.status_code == 200
except Exception:
return False
def end_run(self, session):
ret = super().end_run(session)
_id = session.id
self.web_envs[_id].stop()
del self.web_envs[_id] # remove web env
return ret
def step_call(self, messages, session, model=None):
_use_multimodal = session.info.get("use_multimodal", False) or have_images_in_messages(messages)
if model is None:
model = self.model_multimodal if _use_multimodal else self.model # use which model?
response = model(messages)
return response
def step_prepare(self, session, state):
_input_kwargs, _extra_kwargs = super().step_prepare(session, state)
_web_env = self.web_envs[session.id]
_web_state = _web_env.get_state()
_this_page_info = self._prep_page(_web_state)
_input_kwargs.update(_this_page_info) # update for the current one
if session.num_of_steps() > 1: # has previous step
_prev_step = session.get_specific_step(-2) # the step before
_input_kwargs.update(self._prep_page(_prev_step["action"]["web_state_before"], suffix="_old"))
else:
_input_kwargs["web_page_old"] = "N/A"
_input_kwargs["html_md"] = self._prep_html_md(_web_state)
# --
# check web page differences
if session.num_of_steps() >= self.check_nodiff_steps and self.check_nodiff_steps > 1:
_check_pages = [self._prep_page(z["action"]["web_state_before"]) for z in session.get_latest_steps(count=self.check_nodiff_steps-1)] + [_this_page_info]
if all(z==_check_pages[0] for z in _check_pages): # error
# 埋点:检测到卡在同一页面的错误
if self.logger:
self.logger.warning("[WEB_FALLBACK] Trigger: stuck_same_page | Method: stop_function | Result: error_message_added | Impact: task_termination")
_input_kwargs["web_page"] = _input_kwargs["web_page"] + "\n(* Error: Notice that we have been stuck at the same page for many steps, use the `stop` function to terminate and report related errors!!)"
elif _check_pages[-1] == _check_pages[-2]: # warning
# 埋点:检测到页面未变化的警告
if self.logger:
self.logger.debug("[WEB_DECISION] page_unchanged -> warning_message")
_input_kwargs["web_page"] = _input_kwargs["web_page"] + "\n(* Warning: Notice that the web page has not been changed.)"
# --
_extra_kwargs["web_env"] = _web_env
return _input_kwargs, _extra_kwargs
def step_action(self, action_res, action_input_kwargs, web_env=None, **kwargs):
action_res["web_state_before"] = web_env.get_state() # inplace storage of the web-state before the action
_rr = super().step_action(action_res, action_input_kwargs) # get action from code execution
if isinstance(_rr, ActionResult):
action_str, action_result = _rr.action, _rr.result
else:
action_str = self.get_obs_str(None, obs=_rr, add_seq_enum=False)
action_str, action_result = "nop", action_str.strip() # no-operation
# 埋点:浏览器动作执行前
if self.logger:
current_state = web_env.get_state()
current_url = current_state.get('current_url', 'unknown')
self.logger.info("[WEB_BROWSER] Executing: %s", action_str)
self.logger.debug("[WEB_STATE] Before_URL: %s", current_url)
# state step
try: # execute the action on the browser
step_result = web_env.step_state(action_str)
ret = action_result if action_result is not None else step_result # use action result if there are direct ones
web_env.sync_files()
# 埋点:浏览器动作执行后
if self.logger:
new_state = web_env.get_state()
new_url = new_state.get('current_url', 'unknown')
self.logger.info("[WEB_BROWSER] Result: success | URL: %s", new_url)
if new_url != current_url:
self.logger.info("[WEB_STATE] URL_Changed: %s -> %s", current_url, new_url)
except Exception as e:
zwarn("web_env execution error!")
ret = f"Browser error: {e}"
# 埋点:浏览器动作执行错误
if self.logger:
self.logger.error("[WEB_BROWSER] Error: %s", str(e))
return ret
# --
# other helpers
def _prep_page(self, web_state, suffix=""):
_ss = web_state
_ret = _ss["current_accessibility_tree"]
if _ss["error_message"]:
_ret = _ret + "\n(Note: " + _ss["error_message"] + ")"
elif _ss["current_has_cookie_popup"]:
_ret = _ret + "\n(Note: There is a cookie banner on the page, please accept the cookie banner.)"
ret = {"web_page": _ret, "downloaded_file_path": _ss["downloaded_file_path"]}
# --
if self.use_multimodal == 'on': # always on
ret["screenshot"] = _ss["boxed_screenshot"]
elif self.use_multimodal == 'off':
ret["screenshot_note"] = "The current system does not support webpage screenshots. Please refer to the accessibility tree to understand the current webpage."
else: # adaptive decision
if web_state.get("curr_screenshot_mode"): # currently on
ret["screenshot"] = _ss["boxed_screenshot"]
else:
ret["screenshot_note"] = "The current system's screenshot mode is off. If you need the screenshots, please use the corresponding action to turn it on."
# --
if suffix:
ret = {k+suffix: v for k, v in ret.items()}
return ret
def _prep_html_md(self, web_state):
_IGNORE_LINE_LEN = 7 # ignore md line if <= this
_LOCAL_WINDOW = 2 # -W -> +W
_budget = self.html_md_budget
if _budget <= 0:
return ""
# --
axtree, html_md = web_state["current_accessibility_tree"], web_state.get("html_md", "")
# first locate raw texts from axtree
axtree_texts = []
for line in axtree.split("\n"):
m = re.findall(r"(?:StaticText|link)\s+'(.*)'", line)
axtree_texts.extend(m)
# then locate to the html ones
html_lines = [z for z in html_md.split("\n") if z.strip() and len(z) > _IGNORE_LINE_LEN]
hit_lines = set()
_last_hit = 0
for one_t in axtree_texts:
_curr = _last_hit
while _curr < len(html_lines):
if one_t in html_lines[_curr]: # hit
hit_lines.update([ii for ii in range(_curr-_LOCAL_WINDOW, _curr+_LOCAL_WINDOW+1) if ii>=0 and ii<len(html_lines)]) # add local window
_last_hit = _curr
break
_curr += 1
# get the contents
_last_idx = -1
_all_addings = []
_all_adding_lines = []
for line_idx in sorted(hit_lines):
if _budget < 0:
break
_line = html_lines[line_idx].rstrip()
adding = f"...\n{_line}" if (line_idx > _last_idx+1) else _line
_all_addings.append(adding)
_all_adding_lines.append(line_idx)
_budget -= len(adding.encode()) # with regard to bytes!
_last_idx = line_idx
while _budget > 0: # add more lines if we still have budget
_last_idx = _last_idx + 1
if _last_idx >= len(html_lines):
break
_line = html_lines[_last_idx].rstrip()
_all_addings.append(_line)
_all_adding_lines.append(_last_idx)
_budget -= len(_line.encode()) # with regard to bytes!
if _last_idx < len(html_lines):
_all_addings.append("...")
final_ret = "\n".join(_all_addings)
return final_ret
def set_multimodal(self, use_multimodal):
if use_multimodal is not None:
self.use_multimodal = use_multimodal
def get_multimodal(self):
return self.use_multimodal
|