feat: fsapp reconnect loop, tui branch/rewind, simphtml flow-child fix, sop updates

frontends/fsapp.py

@@ -257,13 +257,16 @@ def _card(text):
 
 
 def _send_raw(receive_id, payload, msg_type, rtype):
-    body = CreateMessageRequest.builder().receive_id_type(rtype).request_body(
+    try:
-        CreateMessageRequestBody.builder().receive_id(receive_id).msg_type(msg_type).content(payload).build()
+        body = CreateMessageRequest.builder().receive_id_type(rtype).request_body(
-    ).build()
+            CreateMessageRequestBody.builder().receive_id(receive_id).msg_type(msg_type).content(payload).build()
-    r = client.im.v1.message.create(body)
+        ).build()
-    if r.success():
+        r = client.im.v1.message.create(body)
-        return r.data.message_id if r.data else None
+        if r.success():
-    print(f"发送失败: {r.code}, {r.msg}")
+            return r.data.message_id if r.data else None
+        print(f"发送失败: {r.code}, {r.msg}")
+    except Exception as e:
+        print(f"[ERROR] _send_raw 网络异常: {e}")
     return None
 
 
@@ -272,14 +275,18 @@ def _patch_card(message_id, card_json):
 
 
 def _patch_card_result(message_id, card_json):
-    body = PatchMessageRequest.builder().message_id(message_id).request_body(
+    try:
-        PatchMessageRequestBody.builder().content(card_json).build()
+        body = PatchMessageRequest.builder().message_id(message_id).request_body(
-    ).build()
+            PatchMessageRequestBody.builder().content(card_json).build()
-    r = client.im.v1.message.patch(body)
+        ).build()
-    if not r.success():
+        r = client.im.v1.message.patch(body)
-        print(f"[ERROR] patch_card 失败: {r.code}, {r.msg}")
+        if not r.success():
-    msg = f"{getattr(r, 'code', '')} {getattr(r, 'msg', '')}".lower()
+            print(f"[ERROR] patch_card 失败: {r.code}, {r.msg}")
-    return r.success(), ("230099" in msg or "11310" in msg or "element exceeds the limit" in msg)
+        msg = f"{getattr(r, 'code', '')} {getattr(r, 'msg', '')}".lower()
+        return r.success(), ("230099" in msg or "11310" in msg or "element exceeds the limit" in msg)
+    except Exception as e:
+        print(f"[ERROR] _patch_card 网络异常: {e}")
+        return False, False
 
 
 def send_message(receive_id, content, msg_type="text", use_card=False, receive_id_type="open_id"):
@@ -680,9 +687,22 @@ def main():
         sys.exit(1)
     client = create_client()
     handler = lark.EventDispatcherHandler.builder("", "").register_p2_im_message_receive_v1(handle_message).build()
-    cli = lark.ws.Client(APP_ID, APP_SECRET, event_handler=handler, log_level=lark.LogLevel.INFO)
     print("=" * 50 + "\n飞书 Agent 已启动（长连接模式）\n" + f"App ID: {APP_ID}\n等待消息...\n" + "=" * 50)
-    cli.start()
+    retry_delay = 5
+    while True:
+        try:
+            cli = lark.ws.Client(APP_ID, APP_SECRET, event_handler=handler, log_level=lark.LogLevel.INFO)
+            cli.start()
+        except Exception as e:
+            print(f"[WARN] 飞书长连接断开或启动失败: {e}")
+        print(f"[INFO] {retry_delay}s 后重连...")
+        time.sleep(retry_delay)
+        retry_delay = min(retry_delay * 2, 120)
+        # 重连时刷新 client
+        try:
+            client = create_client()
+        except Exception:
+            pass
 
 
 if __name__ == "__main__":

frontends/stapp.py

@@ -253,7 +253,6 @@ if prompt := st.chat_input("any task?"):
             with live.container(): render_segments([segs[i]])
             if i < len(segs) - 1: live = st.empty()
     st.session_state.messages.append({"role": "assistant", "content": response})
-    st.session_state.messages = st.session_state.messages[-50:]
     st.session_state.last_reply_time = int(time.time())
 
 if st.session_state.autonomous_enabled:

frontends/tuiapp.py

@@ -6,7 +6,6 @@ Run from the project root:
 
 Useful options:
 
-    python frontends/tuiapp.py --demo   # no LLM/key usage; fake streaming agent for smoke tests
     python frontends/tuiapp.py --help
 
 MVP design notes:
@@ -147,65 +146,11 @@ def parse_local_command(raw: str) -> tuple[str, list[str]] | None:
     name, *rest = text.split(maxsplit=1)
     cmd = name[1:].lower()
     args = rest[0].split() if rest else []
-    if cmd in {"help", "status", "new", "switch", "sessions", "stop", "llm", "quit", "exit"}:
+    if cmd in {"help", "status", "new", "switch", "sessions", "stop", "llm", "branch", "rewind", "quit", "exit"}:
         return cmd, args
     return None
 
 
-class DemoAgent:
-    """Small fake agent used by --demo and tests; mimics put_task/run/display_queue."""
-
-    def __init__(self) -> None:
-        self.task_queue: queue.Queue = queue.Queue()
-        self.is_running = False
-        self.stop_sig = False
-        self.inc_out = True
-        self.verbose = False
-        self._llm_no = 0
-
-    def run(self) -> None:
-        while True:
-            task = self.task_queue.get()
-            raw = task["query"]
-            dq = task["output"]
-            self.is_running = True
-            self.stop_sig = False
-            chunks = [
-                "LLM Running (Turn 1) ...\n<summary>demo planning</summary>\n",
-                f"Echo: {raw}\n",
-                "LLM Running (Turn 2) ...\nFinal demo answer.\n",
-            ]
-            full = ""
-            for chunk in chunks:
-                if self.stop_sig:
-                    break
-                time.sleep(0.08)
-                full += chunk
-                dq.put({"next": chunk, "source": "demo"})
-            if self.stop_sig:
-                full += "\n[Interrupted]"
-            dq.put({"done": full, "source": "demo"})
-            self.is_running = False
-            self.task_queue.task_done()
-
-    def put_task(self, query: str, source: str = "user", images: Optional[list[Any]] = None) -> queue.Queue:
-        dq: queue.Queue = queue.Queue()
-        self.task_queue.put({"query": query, "source": source, "images": images or [], "output": dq})
-        return dq
-
-    def abort(self) -> None:
-        self.stop_sig = True
-
-    def list_llms(self) -> list[tuple[int, str, bool]]:
-        return [(0, "DemoAgent/demo", True)]
-
-    def next_llm(self, n: int = -1) -> None:
-        self._llm_no = 0
-
-    def get_llm_name(self, b: Any = None, model: bool = False) -> str:
-        return "demo"
-
-
 def default_agent_factory() -> Any:
     from agentmain import GenericAgent
 
@@ -237,13 +182,12 @@ class GenericAgentTUI(App[None]):
         Binding("ctrl+right", "next_session", "Next→", show=True, priority=True),
     ]
 
-    def __init__(self, agent_factory: Optional[AgentFactory] = None, *, demo: bool = False) -> None:
+    def __init__(self, agent_factory: Optional[AgentFactory] = None) -> None:
         super().__init__()
-        self.agent_factory: AgentFactory = agent_factory or (DemoAgent if demo else default_agent_factory)
+        self.agent_factory: AgentFactory = agent_factory or default_agent_factory
         self.sessions: dict[int, AgentSession] = {}
         self.current_id: Optional[int] = None
         self._ids = count(1)
-        self.demo = demo
         self.fold_mode: bool = True
         self._last_stream_refresh: float = 0.0
         self._stream_throttle_ms: float = 0.15  # seconds between streaming UI refreshes
@@ -255,7 +199,7 @@ class GenericAgentTUI(App[None]):
             with Vertical(id="main"):
                 yield Static("", id="status")
                 yield RichLog(id="log", wrap=True, highlight=True, markup=True)
-        yield Input(placeholder="Message, or /help  /new name  /switch 1  /sessions  /stop  /llm", id="prompt")
+        yield Input(placeholder="Message, or /help  /new  /branch  /rewind  /switch  /stop  /llm  /resume", id="prompt")
         yield Footer()
 
     def on_mount(self) -> None:
@@ -263,6 +207,10 @@ class GenericAgentTUI(App[None]):
         self._system("Welcome to GenericAgent TUI. Type /help for commands.")
         self.query_one("#prompt", Input).focus()
 
+    def on_resize(self, event) -> None:
+        narrow = self.size.width < 70
+        self.query_one("#sidebar").styles.display = "none" if narrow else "block"
+
     @property
     def current(self) -> AgentSession:
         if self.current_id is None:
@@ -340,6 +288,8 @@ class GenericAgentTUI(App[None]):
             "sessions": self._cmd_sessions,
             "stop": self._cmd_stop,
             "llm": self._cmd_llm,
+            "branch": self._cmd_branch,
+            "rewind": self._cmd_rewind,
             "quit": lambda _args: self.exit(),
             "exit": lambda _args: self.exit(),
         }
@@ -412,13 +362,18 @@ class GenericAgentTUI(App[None]):
                 break
         else:
             session.messages.append(ChatMessage("assistant", text, task_id=task_id, done=done))
-        self._refresh_all()
+        if agent_id == self.current_id:
+            self._refresh_all()
+        else:
+            self._refresh_sidebar()
 
     def _cmd_help(self, args: list[str]) -> None:
         self._system(
             "Commands:\n"
             "/help - show this help\n"
             "/new [name] - create and switch to a new agent session\n"
+            "/branch [name] - fork current session (copies LLM history + display)\n"
+            "/rewind - list rewindable turns; /rewind <n> to truncate history\n"
             "/switch <id|name> - switch active session\n"
             "/sessions - list sessions\n"
             "/status - show current/all status\n"
@@ -434,6 +389,85 @@ class GenericAgentTUI(App[None]):
         session = self.add_session(name)
         self._system(f"Created session #{session.agent_id} {session.name!r}. Shared temp/memory are not isolated.")
 
+    def _cmd_branch(self, args: list[str]) -> None:
+        import copy
+        old_session = self.current
+        name = " ".join(args).strip() or f"{old_session.name}-branch"
+        new_session = self.add_session(name)
+        # Copy LLM backend history
+        try:
+            new_session.agent.llmclient.backend.history = copy.deepcopy(
+                old_session.agent.llmclient.backend.history
+            )
+        except Exception as e:
+            self._system(f"Branch warning: failed to copy history: {e}")
+            return
+        # Copy TUI display messages
+        new_session.messages = copy.deepcopy(old_session.messages)
+        new_session.task_seq = old_session.task_seq
+        n = len(new_session.agent.llmclient.backend.history)
+        self._system(f"Branched from #{old_session.agent_id} → #{new_session.agent_id} ({n} messages inherited).")
+
+    def _cmd_rewind(self, args: list[str]) -> None:
+        session = self.current
+        if session.status == "running":
+            self._system("Cannot rewind while running. /stop first.")
+            return
+        history = session.agent.llmclient.backend.history
+        # Find real user turn boundaries — skip tool_result messages
+        turns = []  # list of (index_in_history, preview_text)
+        for i, msg in enumerate(history):
+            if msg.get("role") != "user":
+                continue
+            content = msg.get("content")
+            # Pure string content is always a real user message
+            if isinstance(content, str):
+                turns.append((i, content[:60]))
+                continue
+            if isinstance(content, list):
+                # Skip if content is purely tool_result blocks
+                has_tool_result = any(b.get("type") == "tool_result" for b in content if isinstance(b, dict))
+                if has_tool_result:
+                    continue
+                texts = [b.get("text", "") for b in content if isinstance(b, dict) and b.get("type") == "text"]
+                if texts and any(t.strip() for t in texts):
+                    turns.append((i, (texts[0] or "")[:60]))
+        if not turns:
+            self._system("No rewindable turns in history.")
+            return
+        # Reverse numbering: 1 = most recent turn, 2 = second most recent, etc.
+        # /rewind without args: show list
+        if not args:
+            lines = [f"Rewindable turns ({len(turns)} total, showing last 10):"]
+            show = turns[-10:]
+            for offset, (_, preview) in enumerate(reversed(show), 1):
+                lines.append(f"  {offset}) {preview!r}")
+            lines.append("/rewind <n> to rewind n turns (1 = undo last turn).")
+            self._system("\n".join(lines))
+            return
+        # /rewind <n>: truncate last n turns
+        try:
+            n = int(args[0])
+        except ValueError:
+            self._system("Usage: /rewind <n> (1 = undo last turn)")
+            return
+        if n < 1 or n > len(turns):
+            self._system(f"Invalid: range is 1-{len(turns)}")
+            return
+        # cut_at = index of the n-th turn from the end
+        cut_at = turns[-n][0]
+        removed = len(history) - cut_at
+        history[:] = history[:cut_at]
+        # Sync TUI messages: keep only messages before the corresponding user message
+        real_user_indices = [i for i, msg in enumerate(session.messages) if msg.role == "user"]
+        if n <= len(real_user_indices):
+            cut_msg = real_user_indices[-n]
+            session.messages = session.messages[:cut_msg]
+        # Mark rewind in agentmain's working memory history
+        try: session.agent.history.append(f"[USER]: /rewind {n}")
+        except Exception: pass
+        self._system(f"Rewound {n} turn(s). Removed {removed} history entries.")
+
     def _cmd_switch(self, args: list[str]) -> None:
         if not args:
             self._system("Usage: /switch <id|name>")
@@ -568,9 +602,8 @@ class GenericAgentTUI(App[None]):
             model = session.agent.get_llm_name(model=True)
         except Exception:
             model = "unknown"
-        mode = "demo" if self.demo else "real"
         status.update(
-            f"[b]#{session.agent_id} {session.name}[/b]  status={session.status}  task={session.current_task_id}  model={model}  mode={mode}\n"
+            f"[b]#{session.agent_id} {session.name}[/b]  status={session.status}  task={session.current_task_id}  model={model}\n"
             "Enter message or /help. Per-task queue streaming is enabled (inc_out=True)."
         )
 
@@ -625,13 +658,12 @@ class GenericAgentTUI(App[None]):
 
 def build_arg_parser() -> argparse.ArgumentParser:
     parser = argparse.ArgumentParser(description="Textual TUI for GenericAgent")
-    parser.add_argument("--demo", action="store_true", help="use a fake streaming agent; no LLM/API keys required")
     return parser
 
 
 def main(argv: Optional[list[str]] = None) -> int:
     args = build_arg_parser().parse_args(argv)
-    app = GenericAgentTUI(demo=args.demo)
+    app = GenericAgentTUI()
     app.run()
     return 0
 

memory/tmwebdriver_sop.md

@@ -11,7 +11,8 @@
 ## 限制(isTrusted)
 - JS事件`isTrusted=false`，敏感操作（如文件上传/部分按钮）可能被拦截；这类场景首选**CDP桥**
 - ⚠JS点击按钮打不开新tab→可能是浏览器弹窗拦截，换CDP点击试试
-- 文件上传：JS无法填充`<input type=file>`；首选CDP batch：getDocument→querySelector→DOM.setFileInputFiles，备选ljqCtrl物理点击
+- Vue3自定义组件(Select/Dropdown)：⭐优先vnode实例调用(无视口限制)→见**vue3_component_sop**；CDP坐标点击仅适合选项少且可见的场景
+- 文件上传：⭐首选**DataTransfer API**（纯JS，无CDP依赖）：`new File([content],name,{type}) → new DataTransfer().items.add(file) → input.files=dt.files → dispatch input+change`；CDP `DOM.setFileInputFiles` 在tmwd桥环境nodeId跨调用失效，不推荐；备选ljqCtrl物理点击
 - 需转物理坐标时：`physX = (screenX + rect中心x) * dpr`，`physY = (screenY + chromeH + rect中心y) * dpr`；其中 `chromeH = outerHeight - innerHeight`
 
 ## 导航
@@ -73,10 +74,20 @@ web_execute_js script='{"cmd": "batch", "commands": [...]}'
   - ⚠tabId：CDP默认sender.tab.id(当前注入页)，跨tab需显式tabId或先batch内tabs查
 - ⭐跨tab无需前台：指定tabId即可操作后台标签页
 
-## CDP点击完整生命周期（未验证，BBS#23）
+## CDP点击完整生命周期（✅已验证）
 - 通用点击需**三事件序列**：mouseMoved → mousePressed → mouseReleased（间隔50-100ms）
   - 省略mouseMoved会导致MUI Tooltip/Ant Design Dropdown等hover依赖组件失效
   - ⚠autofill释放是特例，只需mousePressed即可（见下方autofill章节）
+- ⭐**坐标系结论**：稳定状态下 CDP坐标 = `getBoundingClientRect()` 坐标，**无需修正**
+  - ⚠**首次attach陷阱**：CDP debugger首次attach时Chrome弹出infobar("正在受自动化控制"，~20px高)，页面内容被推下
+  - 如果在attach前测量坐标、attach后发送点击 → 坐标偏移！（之前Currency下拉失败的根因）
+  - ✅**解决**：确保测量坐标在CDP已attach稳定之后（即infobar已出现后再getBoundingClientRect）
+  - 实践：首次CDP操作前先发一个无害的`mouseMoved(0,0)`预热，之后坐标系就稳定了
+- ⭐**下拉框(Vue3 oxd-select等)CDP操作流程**：
+  1. 获取select元素rect → CDP点击打开下拉
+  2. 获取option元素rect → CDP点击选中（option是动态DOM，打开后才能测量）
+  - 已验证：CDP点击对自定义下拉框有效，无isTrusted问题
+  - ⚠**限制**：选项多时底部option超出视口，CDP坐标够不着→此时应优先vnode方案(见vue3_component_sop)
 - 坐标修正（页面有transform:scale/zoom时）：
   ```js
   var scale = window.visualViewport ? window.visualViewport.scale : 1;

simphtml.py

@@ -83,21 +83,24 @@ function createEnhancedDOMCopy() {
       const hasVisualBg = bgAlpha > 0.1 || style.backgroundImage !== 'none' || (style.backdropFilter && style.backdropFilter !== 'none') || style.boxShadow !== 'none';
       
       if (!hasVisualBg && childrenInfos.length > 0) {
-        let minL = Infinity, minT = Infinity, maxR = -Infinity, maxB = -Infinity;
+        // Skip fixed/absolute children when computing parent's merged rect (they're out of flow)
-        for (const cInfo of childrenInfos) {
+        const flowChildren = childrenInfos.filter(cInfo => cInfo.style && cInfo.style.position !== 'fixed' && cInfo.style.position !== 'absolute');
-          minL = Math.min(minL, cInfo.rect.left);
+        if (flowChildren.length > 0) {
-          minT = Math.min(minT, cInfo.rect.top);
+          let minL = Infinity, minT = Infinity, maxR = -Infinity, maxB = -Infinity;
-          maxR = Math.max(maxR, cInfo.rect.right);
+          for (const cInfo of flowChildren) {
-          maxB = Math.max(maxB, cInfo.rect.bottom);
+            minL = Math.min(minL, cInfo.rect.left);
+            minT = Math.min(minT, cInfo.rect.top);
+            maxR = Math.max(maxR, cInfo.rect.right);
+            maxB = Math.max(maxB, cInfo.rect.bottom);
+          }
+          info.rect = { left: minL, top: minT, right: maxR, bottom: maxB, width: maxR - minL, height: maxB - minT };
+          info.area = info.rect.width * info.rect.height;
+        } else {
+          const maxC = childrenInfos.filter(i => i.isVisible).sort((a, b) => b.area - a.area)[0];
+          if (maxC && maxC.area > 10000 && (!isVisible || maxC.area > info.area * 5)) info = maxC;
         }
-        info.rect = { left: minL, top: minT, right: maxR, bottom: maxB, width: maxR - minL, height: maxB - minT };
-        info.area = info.rect.width * info.rect.height;
-      } else {
-        const maxC = childrenInfos.filter(i => i.isVisible).sort((a, b) => b.area - a.area)[0];
-        if (maxC && maxC.area > 10000 && (!isVisible || maxC.area > info.area * 5)) info = maxC;
       }
     }
-    nodeInfo.set(clone, info);
 
     if (sourceNode.nodeType === 1 && sourceNode.tagName === 'DIV') {    
       if (!hasValidChildren && !sourceNode.textContent.trim()) return null;