refactor: update agent loop and tool extraction

agent_loop.py

@@ -28,10 +28,7 @@ class BaseHandler:
             yield f"未知工具: {tool_name}\n"
             return StepOutcome(None, next_prompt=f"未知工具 {tool_name}", should_exit=False)
 
-def json_default(o):
+def json_default(o): return list(o) if isinstance(o, set) else str(o)
-    if isinstance(o, set): return list(o)
-    return str(o) 
-
 def exhaust(g):
     try: 
         while True: next(g)
@@ -47,7 +44,7 @@ def agent_runner_loop(client, system_prompt, user_input, handler, tools_schema,
         {"role": "system", "content": system_prompt},
         {"role": "user", "content": initial_user_content if initial_user_content is not None else user_input}
     ]
-    turn = 0; handler._done_hooks = [];  handler.max_turns = max_turns
+    turn = 0;  handler.max_turns = max_turns
     while turn < handler.max_turns:
         turn += 1; md = '**' if verbose else ''
         yield f"{md}LLM Running (Turn {turn}) ...{md}\n\n"
@@ -65,7 +62,7 @@ def agent_runner_loop(client, system_prompt, user_input, handler, tools_schema,
         else: tool_calls = [{'tool_name': tc.function.name, 'args': json.loads(tc.function.arguments), 'id': tc.id}
                           for tc in response.tool_calls]
        
-        tool_results = []; next_prompts = set(); exit_reason = None
+        tool_results = []; next_prompts = set(); exit_reason = {}
         for ii, tc in enumerate(tool_calls):
             tool_name, args, tid = tc['tool_name'], tc['args'], tc.get('id', '')
             if tool_name == 'no_tool': pass
@@ -92,7 +89,7 @@ def agent_runner_loop(client, system_prompt, user_input, handler, tools_schema,
                 tool_results.append({'tool_use_id': tid, 'content': datastr})
             next_prompts.add(outcome.next_prompt)
         if len(next_prompts) == 0 or exit_reason:
-            if len(handler._done_hooks) == 0: break
+            if len(handler._done_hooks) == 0 or exit_reason.get('result', '') == 'EXITED': break
             next_prompts.add(handler._done_hooks.pop(0))
         next_prompt = handler.turn_end_callback(response, tool_calls, tool_results, turn, '\n'.join(next_prompts), exit_reason)
         messages = [{"role": "user", "content": next_prompt, "tool_results": tool_results}]   # just new message, history is kept in *Session

agentmain.py

@@ -77,13 +77,15 @@ class GeneraticAgent:
         self.llmclient = self.llmclients[self.llm_no]
         self.llmclient.backend.history = lastc.backend.history
         self.llmclient.last_tools = ''
-        name = self.get_llm_name().lower()
+        name = self.get_llm_name(model=True)
         if 'glm' in name or 'minimax' in name or 'kimi' in name: load_tool_schema('_cn')
         else: load_tool_schema()
     def list_llms(self): return [(i, self.get_llm_name(b), i == self.llm_no) for i, b in enumerate(self.llmclients)]
-    def get_llm_name(self, b=None):
+    def get_llm_name(self, b=None, model=False):
         b = self.llmclient if b is None else b
-        return f"{type(b.backend).__name__}/{b.backend.name}" if not isinstance(b, dict) else "BADCONFIG_MIXIN"
+        if isinstance(b, dict): return 'BADCONFIG_MIXIN'
+        if model: return b.backend.model.lower()
+        return f"{type(b.backend).__name__}/{b.backend.name}"
 
     def abort(self):
         if not self.is_running: return
@@ -135,11 +137,10 @@ class GeneraticAgent:
             user_input = raw_query
             if source == 'feishu' and len(self.history) > 1:   # 如果有历史记录且来自飞书，注入到首轮 user_input 中（支持/restore恢复上下文）
                 user_input = handler._get_anchor_prompt() + f"\n\n### 用户当前消息\n{raw_query}"
-            initial_user_content = None
+            if 'gpt' in self.get_llm_name(model=True): handler._done_hooks.append('请确定用户任务是否完成，如未完成需要继续工具调用直到完成任务，确实需要问用户应使用ask_user工具')
             # although new handler, the **full** history is in llmclient, so it is full history!
             gen = agent_runner_loop(self.llmclient, sys_prompt, user_input, 
-                                handler, TOOLS_SCHEMA, max_turns=40, verbose=self.verbose,
+                                handler, TOOLS_SCHEMA, max_turns=40, verbose=self.verbose)
-                                initial_user_content=initial_user_content)
             try:
                 full_resp = ""; last_pos = 0
                 for chunk in gen:

frontends/wechatapp.py

@@ -11,8 +11,8 @@ from agentmain import GeneraticAgent
 API = 'https://ilinkai.weixin.qq.com'
 TOKEN_FILE = Path.home() / '.wxbot' / 'token.json'
 TOKEN_FILE.parent.mkdir(exist_ok=True)
-VER, MSG_USER, MSG_BOT, ITEM_TEXT, STATE_FINISH = '0.2.5', 1, 2, 1, 2
+VER, MSG_USER, MSG_BOT, ITEM_TEXT, STATE_FINISH = '2.1.8', 1, 2, 1, 2
-ITEM_FILE = 4
+ITEM_IMAGE, ITEM_FILE, ITEM_VIDEO = 2, 4, 5
 CDN_BASE = 'https://novac2c.cdn.weixin.qq.com/c2c'
 
 def _uin():
@@ -94,48 +94,72 @@ class WxBotClient:
             'to_user_id': to_user_id, 'typing_ticket': typing_ticket,
             'typing_status': 2 if cancel else 1, 'base_info': {'channel_version': VER}})
 
-    def send_file(self, to_user_id, file_path, context_token=''):
+    def _enc(self, raw, aes_key):
-        """Send a file to user via CDN upload."""
+        pad = 16 - (len(raw) % 16)
+        return AES.new(aes_key, AES.MODE_ECB).encrypt(raw + bytes([pad] * pad))
+
+    def _upload(self, filekey, upload_param, raw, aes_key, timeout=120, upload_url=''):
+        url = upload_url.strip() if upload_url else f'{CDN_BASE}/upload?encrypted_query_param={quote(upload_param)}&filekey={filekey}'
+        data = self._enc(raw, aes_key)
+        last_err = None
+        for attempt in range(1, 4):
+            try:
+                r = requests.post(url, data=data, headers={'Content-Type': 'application/octet-stream'}, timeout=timeout)
+                if 400 <= r.status_code < 500:
+                    msg = r.headers.get('x-error-message') or r.text[:300]
+                    raise RuntimeError(f'CDN upload client error {r.status_code}: {msg}')
+                if r.status_code != 200:
+                    msg = r.headers.get('x-error-message') or f'status {r.status_code}'
+                    raise RuntimeError(f'CDN upload server error: {msg}')
+                eq = r.headers.get('x-encrypted-param', '')
+                if not eq: raise RuntimeError('CDN upload response missing x-encrypted-param header')
+                return {'encrypt_query_param': eq,
+                        'aes_key': base64.b64encode(aes_key.hex().encode()).decode(), 'encrypt_type': 1}
+            except Exception as e:
+                last_err = e
+                if 'client error' in str(e) or attempt >= 3: break
+                print(f'[WX] CDN upload retry {attempt}: {e}', file=sys.__stdout__)
+        raise last_err
+
+    def _send_media(self, to_user_id, file_path, media_type, item_type, item_key, context_token=''):
         fp = Path(file_path)
         raw = fp.read_bytes()
-        rawsize = len(raw)
-        rawfilemd5 = hashlib.md5(raw).hexdigest()
-        aes_key = os.urandom(16)
         filekey = uuid.uuid4().hex
-        ciphertext_size = ((rawsize // 16) + 1) * 16
+        aes_key = os.urandom(16)
-        # 1. get upload url
+        ciphertext_size = ((len(raw) // 16) + 1) * 16
-        resp = self._post('ilink/bot/getuploadurl', {
+        body = {
-            'filekey': filekey, 'media_type': 3, 'to_user_id': to_user_id,
+            'filekey': filekey, 'media_type': media_type, 'to_user_id': to_user_id,
-            'rawsize': rawsize, 'rawfilemd5': rawfilemd5,
+            'rawsize': len(raw), 'rawfilemd5': hashlib.md5(raw).hexdigest(),
             'filesize': ciphertext_size, 'no_need_thumb': True,
-            'aeskey': aes_key.hex(),
+            'aeskey': aes_key.hex(), 'base_info': {'channel_version': VER}}
-            'base_info': {'channel_version': VER}})
+        resp = self._post('ilink/bot/getuploadurl', body)
         upload_param = resp.get('upload_param', '')
-        if not upload_param:
+        upload_url = resp.get('upload_full_url', '')
-            raise RuntimeError(f'getuploadurl failed: {resp}')
+        if not (upload_param or upload_url): raise RuntimeError(f'getuploadurl failed: {resp}')
-        # 2. AES-128-ECB encrypt (PKCS7)
+        media = self._upload(filekey, upload_param, raw, aes_key=aes_key, upload_url=upload_url)
-        cipher = AES.new(aes_key, AES.MODE_ECB)
+        item = {'media': media}
-        pad_len = 16 - (rawsize % 16)
+        if item_key == 'file_item':
-        ciphertext = cipher.encrypt(raw + bytes([pad_len] * pad_len))
+            item.update({'file_name': fp.name, 'len': str(len(raw))})
-        # 3. upload to CDN
+        elif item_key == 'image_item':
-        upload_url = (f'{CDN_BASE}/upload?encrypted_query_param='
+            item.update({'mid_size': ciphertext_size})
-                      f'{quote(upload_param)}&filekey={filekey}')
+        elif item_key == 'video_item':
-        r = requests.post(upload_url, data=ciphertext, headers={'Content-Type': 'application/octet-stream'}, timeout=120)
+            item.update({'video_size': ciphertext_size})
-        r.raise_for_status()
-        download_param = r.headers.get('x-encrypted-param', '')
-        if not download_param:
-            raise RuntimeError(f'CDN upload: no x-encrypted-param. status={r.status_code}')
-        # 4. send message with file attachment
         msg = {'from_user_id': '', 'to_user_id': to_user_id,
                'client_id': f'pyclient-{uuid.uuid4().hex[:16]}',
                'message_type': MSG_BOT, 'message_state': STATE_FINISH,
-               'item_list': [{'type': ITEM_FILE, 'file_item': {
+               'item_list': [{'type': item_type, item_key: item}]}
-                   'media': {'encrypt_query_param': download_param,
-                             'aes_key': base64.b64encode(aes_key.hex().encode()).decode(), 'encrypt_type': 1},
-                   'file_name': fp.name, 'len': str(rawsize)}}]}
         if context_token: msg['context_token'] = context_token
         return self._post('ilink/bot/sendmessage', {'msg': msg, 'base_info': {'channel_version': VER}})
 
+    def send_file(self, to_user_id, file_path, context_token=''):
+        return self._send_media(to_user_id, file_path, 3, ITEM_FILE, 'file_item', context_token)
+
+    def send_image(self, to_user_id, file_path, context_token=''):
+        return self._send_media(to_user_id, file_path, 1, ITEM_IMAGE, 'image_item', context_token)
+
+    def send_video(self, to_user_id, file_path, context_token=''):
+        return self._send_media(to_user_id, file_path, 2, ITEM_VIDEO, 'video_item', context_token)
+
     @staticmethod
     def extract_text(msg):
         return '\n'.join(it['text_item'].get('text', '')
@@ -190,7 +214,7 @@ def _dl_media(items):
 agent = GeneraticAgent()
 agent.verbose = False
 
-_TAG_PATS = [r'<' + t + r'>.*?</' + t + r'>' for t in ('thinking', 'summary', 'tool_use')]
+_TAG_PATS = [r'<' + t + r'>.*?</' + t + r'>' for t in ('thinking', 'tool_use')]
 _TAG_PATS.append(r'<file_content>.*?</file_content>')
 
 def _strip_md(t):
@@ -213,8 +237,11 @@ def _strip_md(t):
     return re.sub(r'\n{3,}', '\n\n', t).strip()
 
 def _clean(t):
+    t = re.sub(r'^\s*LLM Running \(Turn \d+\) \.{3}\s*$', '', t, flags=re.M)
+    t = re.sub(r'^\s*🛠️\s*[A-Za-z_][A-Za-z0-9_]*\(.*$', '', t, flags=re.M)
     for p in _TAG_PATS:
         t = re.sub(p, '', t, flags=re.DOTALL)
+    t = re.sub(r'</?summary>', '', t)
     return re.sub(r'\n{3,}', '\n\n', _strip_md(t)).strip() or '...'
 
 def _split(text, limit=1800):
@@ -270,7 +297,8 @@ def on_message(bot, msg):
                 if 'done' in item: result = item['done']; break
         except queue.Empty: result = '[超时]'
         files = re.findall(r'\[FILE:([^\]]+)\]', result)
-        files = [f for f in files if (f if os.path.isabs(f) else os.path.join(_TEMP_DIR, f)) not in media_paths]
+        bad = {'filepath', '<filepath>', 'path', '<path>', 'file_path', '<file_path>', '...'}
+        files = [f for f in files if f.strip().lower() not in bad and (f if os.path.isabs(f) else os.path.join(_TEMP_DIR, f)) not in media_paths]
         show = _clean(result)
         chunks = _split(show)
         _MAX_MSGS = 6
@@ -285,9 +313,12 @@ def on_message(bot, msg):
             if not os.path.isabs(fpath): fpath = os.path.join(_TEMP_DIR, fpath)
             try:
                 if not os.path.exists(fpath): raise FileNotFoundError(f"文件不存在: {fpath}")
-                bot.send_file(uid, fpath, context_token=ctx)
+                ext = os.path.splitext(fpath)[1].lower()
-                print(f'[WX] sent file: {fpath}', file=sys.__stdout__)
+                sender = bot.send_video if ext in {'.mp4', '.mov', '.m4v', '.webm'} else \
-            except Exception as e: print(f'[WX] send_file err: {e}', file=sys.__stdout__)
+                         bot.send_image if ext in {'.jpg', '.jpeg', '.png', '.gif', '.webp', '.bmp'} else bot.send_file
+                sender(uid, fpath, context_token=ctx)
+                print(f'[WX] sent media: {fpath}', file=sys.__stdout__)
+            except Exception as e: print(f'[WX] send media err: {e}', file=sys.__stdout__)
 
     threading.Thread(target=_handle, daemon=True).start()
 

ga.py

@@ -17,7 +17,7 @@ def code_run(code, code_type="python", timeout=60, cwd=None, code_cwd=None, stop
     yield f"[Action] Running {code_type} in {os.path.basename(cwd)}: {preview}\n"
     script_dir = os.path.dirname(os.path.abspath(__file__))
     cwd = cwd or os.path.join(script_dir, 'temp'); tmp_path = None
-    if code_type == "python":
+    if code_type in ["python", "py"]:
         tmp_file = tempfile.NamedTemporaryFile(suffix=".ai.py", delete=False, mode='w', encoding='utf-8', dir=code_cwd)
         cr_header = os.path.join(script_dir, 'assets', 'code_run_header.py')
         if os.path.exists(cr_header): tmp_file.write(open(cr_header, encoding='utf-8').read())
@@ -25,7 +25,7 @@ def code_run(code, code_type="python", timeout=60, cwd=None, code_cwd=None, stop
         tmp_path = tmp_file.name
         tmp_file.close()
         cmd = [sys.executable, "-X", "utf8", "-u", tmp_path]   
-    elif code_type in ["powershell", "bash"]:
+    elif code_type in ["powershell", "bash", "sh", "shell", "ps1", "pwsh"]:
         if os.name == 'nt': cmd = ["powershell", "-NoProfile", "-NonInteractive", "-Command", code]
         else: cmd = ["bash", "-c", code]
     else:
@@ -110,12 +110,10 @@ def first_init_driver():
         time.sleep(3)
 
 def web_scan(tabs_only=False, switch_tab_id=None, text_only=False):
-    """
+    """获取当前页面的简化HTML内容和标签页列表。注意：简化过程会过滤边栏、浮动元素等非主体内容。
-    获取当前页面的简化HTML内容和标签页列表。注意：简化过程会过滤边栏、浮动元素等非主体内容。
     tabs_only: 仅返回标签页列表，不获取HTML内容（节省token）。
     switch_tab_id: 可选参数，如果提供，则在扫描前切换到该标签页。
-    应当多用execute_js，少全量观察html。
+    应当多用execute_js，少全量观察html"""
-    """
     global driver
     try:
         if driver is None: first_init_driver()
@@ -265,13 +263,15 @@ class GenericAgentHandler(BaseHandler):
         self.cwd = cwd;  self.current_turn = 0
         self.history_info = last_history if last_history else []
         self.code_stop_signal = []
+        self._done_hooks = []
 
     def _get_abs_path(self, path):
         if not path: return ""
         return os.path.abspath(os.path.join(self.cwd, path))   
 
     def _extract_code_block(self, response, code_type):
-        matches = re.findall(rf"```{code_type}\n(.*?)\n```", response.content, re.DOTALL)
+        code_type = {'python':'python|py', 'powershell':'powershell|ps1|pwsh', 'bash':'bash|sh|shell'}.get(code_type, re.escape(code_type))
+        matches = re.findall(rf"```(?:{code_type})\n(.*?)\n```", response.content, re.DOTALL)
         return matches[-1].strip() if matches else None
 
     def do_code_run(self, args, response):
@@ -280,7 +280,7 @@ class GenericAgentHandler(BaseHandler):
         code = args.get("code") or args.get("script")
         if not code:
             code = self._extract_code_block(response, code_type)
-            if not code: return StepOutcome("[Error] Code missing. Use ```{code_type} block or 'script' arg.", next_prompt="\n")
+            if not code: return StepOutcome("[Error] Code missing. Must use reply code block or 'script' arg.", next_prompt="\n")
         timeout = args.get("timeout", 60)
         raw_path = os.path.join(self.cwd, args.get("cwd", './'))
         cwd = os.path.normpath(os.path.abspath(raw_path))
@@ -309,8 +309,7 @@ class GenericAgentHandler(BaseHandler):
     def do_web_scan(self, args, response):
         '''获取当前页面内容和标签页列表。也可用于切换标签页。
         注意：HTML经过简化，边栏/浮动元素等可能被过滤。如需查看被过滤的内容请用execute_js。
-        tabs_only=true时仅返回标签页列表，不获取HTML（省token）。
+        tabs_only=true时仅返回标签页列表，不获取HTML（省token）'''
-        '''
         tabs_only = args.get("tabs_only", False)
         switch_tab_id = args.get("switch_tab_id", None)
         text_only = args.get("text_only", False)
@@ -523,7 +522,7 @@ class GenericAgentHandler(BaseHandler):
             clean_args = {k: v for k, v in args.items() if not k.startswith('_')}
             summary = f"调用工具{tool_name}, args: {clean_args}"
             if tool_name == 'no_tool': summary = "直接回答了用户问题"
-            next_prompt += "\n[DANGER] 上一轮遗漏了<summary>，已根据物理动作自动补全。在下次回复中记得<summary>协议。" 
+            next_prompt += "\n[DANGER] 上一轮遗漏了<summary>，需要按协议在<summary>中输出极简单行摘要！" 
         summary = smart_format(summary, max_str_len=100)
         self.history_info.append(f'[Agent] {summary}')
         if turn % 35 == 0 and 'plan' not in str(self.working.get('related_sop')):