pdd-skills 3.1.13 → 3.2.2

package/tests/recorder.py

@@ -1,338 +1,406 @@
 """
-Testcase Recorder v1.0 - 交互式浏览器操作录制器
+Testcase Recorder v2.0 - 交互式标注录制器
 ==============================================
-从 testcase-ai.py 中提取的独立录制模块，遵循 SRP 原则。
+从"全自动录制"升级为"人机协作标注"模式。
 
-功能:
-  🎬 注入 JS 事件监听器，捕获用户在浏览器中的操作
-  📝 将操作事件自动转换为 YAML 测试步骤
-  💾 同时生成 .yaml 和 .env 文件
+核心理念:
+  - 录制负责"捕捉鼠标/键盘事件"
+  - 人类负责"理解业务语义"
+  - 重要节点弹出标注，普通操作自动处理
 
-用法:
-  from tests.recorder import run_record_mode
-  asyncio.run(run_record_mode("output.yaml"))
+v2.0 变更 (重大重构):
+  ✨ 新增: 智能节点检测算法
+  ✨ 新增: 交互式标注界面 (控制台)
+  ✨ 新增: 业务语义模板库
+  ✨ 改进: 步骤编号连续无跳跃
+  ✨ 改进: target 描述更准确
 
-  或通过 CLI:
-    python tests/testcase-ai.py --record [output_path]
+用法:
+  python tests/testcase-ai.py --record [output.yaml]
+  python tests/testcase-ai.py --record-annotate [output.yaml]  # 强制标注模式
 """
 
 import asyncio
 import json
 import os
+import re
+import shutil
+import subprocess
 import sys
 import time
-from typing import Any, Dict, List, Optional
+import tempfile
+from typing import Any, Dict, List, Optional, Tuple
 
 import yaml
-
-
-if sys.platform == "win32":
-    import io
-    sys.stdout = io.TextIOWrapper(sys.stdout.buffer, encoding="utf-8", errors="replace")
-    sys.stderr = io.TextIOWrapper(sys.stderr.buffer, encoding="utf-8", errors="replace")
+from mcp import ClientSession, StdioServerParameters
+from mcp.client.stdio import stdio_client
 
 
 # ============================================================
-# JS 注入脚本：注入到浏览器页面中捕获用户操作
+# 常量定义
 # ============================================================
 
-RECORDER_JS = """() => {
-  if (window.__recorder) {
-    return { ok: false, message: 'Already running', count: window.__recorder.events.length };
-  }
-  window.__recorder = { events: [], start: Date.now(), version: '1.0' };
-
-  function getXPath(el) {
-    const parts = [];
-    while (el && el.nodeType === 1) {
-      let idx = 1, sib = el.previousSibling;
-      while (sib) { if (sib.nodeType === 1) idx++; sib = sib.previousSibling; }
-      parts.unshift(el.tagName.toLowerCase() + '[' + idx + ']');
-      el = el.parentNode;
-    }
-    return '/' + parts.join('/');
-  }
-
+IMPORTANT_NODE_PATTERNS = {
+    "submit": ["登录", "login", "submit", "确定", "确认", "提交", "保存", "申请",
+              "下一步", "next", "完成", "finish", "发送", "send"],
+    "navigate": ["菜单", "menu", "首页", "home", "返回", "back", "退出", "logout",
+                "切换", "switch", "tab"],
+    "page_change": [],
+}
+
+MEANINGLESS_TARGETS = {"IMG", "SVG", "PATH", "I", "SPAN", "DIV", "A",
+                       "BUTTON", "INPUT", "FORM", "SECTION", "ARTICLE",
+                       "HEADER", "FOOTER", "NAV", "MAIN", "ASIDE"}
+
+BUSINESS_TEMPLATES = {
+    "login": {"desc": "登录系统", "action": "click", "category": "认证"},
+    "submit_form": {"desc": "提交表单", "action": "click", "category": "数据操作"},
+    "navigate_menu": {"desc": "导航到菜单", "action": "click", "category": "导航"},
+    "search_query": {"desc": "执行查询", "action": "click", "category": "查询"},
+    "new_tab": {"desc": "打开新标签页", "action": "switch_page", "category": "导航"},
+    "fill_field": {"desc": "填写字段", "action": "fill", "category": "数据输入"},
+    "select_option": {"desc": "选择选项", "action": "select_option", "category": "选择"},
+}
+
+
+def _find_chrome_executable() -> str:
+    candidates = [
+        r"C:\Program Files\Google\Chrome\Application\chrome.exe",
+        r"C:\Program Files (x86)\Google\Chrome\Application\chrome.exe",
+        os.path.expanduser(r"~\AppData\Local\Google\Chrome\Application\chrome.exe"),
+    ]
+    for path in candidates:
+        if os.path.exists(path):
+            return path
+    return shutil.which("chrome") or "chrome"
+
+
+def _launch_chrome_with_flags(port: int = 9222) -> subprocess.Popen:
+    chrome_exe = _find_chrome_executable()
+    user_data_dir = tempfile.mkdtemp(prefix="chrome_mcp_")
+    args = [
+        chrome_exe, f"--remote-debugging-port={port}", "--incognito",
+        "--no-first-run", "--no-default-browser-check", "--disable-default-apps",
+        "--disable-sync", "--disable-extensions",
+        "--disable-component-extensions-with-background-pages", "--disable-popup-blocking",
+        "--ignore-certificate-errors", "--ignore-certificate-errors-spki-list",
+        "--disable-web-security", "--allow-running-insecure-content",
+        "--unsafely-treat-insecure-origin-as-secure", "--window-size=1920,1080",
+        f"--user-data-dir={user_data_dir}", "--about:blank",
+    ]
+    proc = subprocess.Popen(args, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
+    time.sleep(2)
+    return proc
+
+
+def _create_incognito_server_params() -> StdioServerParameters:
+    npx_args = ["-y", "chrome-devtools-mcp@latest", "--browser-url=http://127.0.0.1:9222"]
+    return StdioServerParameters(name="Chrome DevTools MCP (Connected)", command="npx", args=npx_args, env=None)
+
+
+SERVER_PARAMS = _create_incognito_server_params()
+_global_chrome_process: Optional[subprocess.Popen] = None
+
+
+def _extract_text(result) -> str:
+    parts = []
+    if hasattr(result, 'content') and result.content:
+        for item in result.content:
+            if hasattr(item, 'text'):
+                raw = item.text
+                code_block = re.search(r'```(?:json)?\s*\n?(.*?)\n?```', raw, re.DOTALL)
+                if code_block: parts.append(code_block.group(1).strip())
+                else: parts.append(raw)
+            else: parts.append(str(item))
+    return "".join(parts)
+
+
+async def _call(session: ClientSession, tool: str, kwargs: dict = None) -> Any:
+    raw = await session.call_tool(tool, kwargs or {})
+    text = _extract_text(raw)
+    if not text or not text.strip(): return None
+    try: return json.loads(text)
+    except (json.JSONDecodeError, TypeError): return text
+
+
+RECORDER_JS = r"""() => {
+  if (window.__recorder) { return { ok: false, message: 'Already running', count: window.__recorder.events.length }; }
+  window.__recorder = { events: [], start: Date.now(), version: '2.0' };
   function capture(e, extra) {
-    const t = e.target;
-    window.__recorder.events.push({
-      type: extra.type || e.type,
-      ts: Date.now(),
-      tag: t.tagName,
-      id: t.id || '',
-      cls: (t.className || '').toString().slice(0, 60),
-      text: (t.textContent || '').trim().slice(0, 80),
-      ph: t.placeholder || '',
-      role: t.getAttribute('role') || '',
-      val: (t.value || '').slice(0, 100),
-      href: t.href || '',
-      name: t.name || '',
-      xpath: getXPath(t),
-      url: location.href,
-      ...extra
-    });
+    const t = e.target; const tag = t.tagName; const role = t.getAttribute('role') || '';
+    let desc = ''; const ph = t.placeholder || ''; const name = t.name || ''; const id = t.id || '';
+    const val = (t.value || '').slice(0, 100); const href = t.href || ''; const cls = (t.className || '').toString();
+    desc = ph || name || id || '';
+    if (tag === 'IMG' && !desc) { desc = t.alt || t.title || (t.src || '').split('/').pop().replace(/\.[^.]+$/, '') || ''; }
+    if ((tag === 'A' || role === 'link') && !desc && href) { desc = (t.textContent || '').trim().slice(0, 50) || href.slice(0, 50); }
+    if ((tag === 'BUTTON' || role === 'button') && !desc) { desc = (t.textContent || '').trim().slice(0, 50) || (t.value || '').slice(0, 50); }
+    const isSelect = (tag === 'SELECT' || cls.includes('el-select') || t.closest('.el-select') !== null || role === 'listbox' || role === 'option');
+    let selectedText = '';
+    if (isSelect && tag === 'SELECT') { selectedText = t.options[t.selectedIndex]?.text || ''; } else if (isSelect) { selectedText = (t.textContent || '').trim().slice(0, 80); }
+    window.__recorder.events.push({ type: extra.type || e.type, ts: Date.now(), tag, desc: desc || (t.textContent || '').trim().slice(0, 50), value: val, isSelect, selectedText, url: location.href, ...extra });
   }
-
   document.addEventListener('click', (e) => capture(e, { type: 'click' }), true);
   document.addEventListener('input', (e) => capture(e, { type: 'input' }), true);
   document.addEventListener('change', (e) => capture(e, { type: 'change' }), true);
-  document.addEventListener('keydown', (e) => {
-    if (e.key === 'Enter') capture(e, { type: 'enter' });
-  }, true);
-
-  const origPush = history.pushState;
-  history.pushState = function() {
-    window.__recorder.events.push({ type: 'navigate', ts: Date.now(), url: arguments[2] || location.href });
-    return origPush.apply(this, arguments);
-  };
-
-  return { ok: true, message: 'Recorder started', pid: performance.now().toString(36) };
+  document.addEventListener('keydown', (e) => { if (e.key === 'Enter') capture(e, { type: 'enter' }); }, true);
+  return { ok: true, message: 'Recorder v2.0 started', url: location.href };
 }"""
 
-POLL_JS = """() => {
-  if (!window.__recorder) return { ok: false };
-  const evts = window.__recorder.events;
-  const result = [...evts];
-  window.__recorder.events = [];
+POLL_JS = r"""() => {
+  if (!window.__recorder) return { ok: false, error: 'not_found' };
+  const evts = window.__recorder.events; const result = [...evts]; window.__recorder.events = [];
   return { ok: true, elapsed: Math.round((Date.now() - window.__recorder.start) / 1000), events: result };
 }"""
 
-STOP_JS = """() => {
+STOP_JS = r"""() => {
   if (!window.__recorder) return { ok: false, events: [] };
-  const all = window.__recorder.events;
-  window.__recorder = null;
+  const all = window.__recorder.events; window.__recorder = null;
   return { ok: true, totalEvents: all.length, events: all };
 }"""
 
 
-# ============================================================
-# 事件 → 步骤转换器
-# ============================================================
-
-def _event_to_step(evt: Dict, step_num: int) -> Optional[Dict]:
-    """将录制事件转换为 YAML 步骤
-
-    Args:
-        evt: 从浏览器捕获的原始事件数据
-        step_num: 当前步骤序号
+class NodeClassifier:
+    @staticmethod
+    def classify(evt: Dict) -> Tuple[str, str]:
+        etype = evt.get("type", ""); desc = evt.get("desc", "").lower(); tag = evt.get("tag", "").upper()
+        if etype == "click":
+            if tag in MEANINGLESS_TARGETS and not desc:
+                has_info = any([evt.get("imgAlt"), evt.get("imgTitle"), evt.get("selectedText"), evt.get("value")])
+                if not has_info: return ("skip", "meaningless_click")
+        if etype in ("click", "enter"):
+            if any(kw in desc for kw in IMPORTANT_NODE_PATTERNS["submit"]): return ("important", "submit")
+        if etype == "click":
+            if any(kw in desc for kw in IMPORTANT_NODE_PATTERNS["navigate"]): return ("important", "navigate")
+        if evt.get("isSelect") and evt.get("selectedText"): return ("semi_auto", "select")
+        if etype in ("input", "change"):
+            if evt.get("value"): return ("auto", "fill")
+        if etype == "click" and desc: return ("auto", "click")
+        return ("skip", "unknown")
+
+
+class InteractiveAnnotator:
+    def __init__(self, force_annotate: bool = False):
+        self.force_annotate = force_annotate; self.annotation_count = 0
+
+    def should_annotate(self, category: str, sub_type: str) -> bool:
+        if self.force_annotate: return True
+        return category in ("important",)
+
+    async def annotate(self, step_num: int, step: Dict, category: str, sub_type: str) -> Dict:
+        self.annotation_count += 1
+        action = step.get("action", ""); desc = step.get("desc", "")
+        target = step.get("target", "") or step.get("desc", "")
+        print(f"\n{'─' * 50}\n  📝 步骤 #{step_num} 需要标注\n{'─' * 50}")
+        print(f"  检测到的操作: [{action}] {desc}\n  目标元素: {target}\n")
+        templates = self._get_relevant_templates(sub_type, target)
+        print(f"  可选业务模板:")
+        for i, (key, tmpl) in enumerate(templates.items(), 1): print(f"    [{i}] {tmpl['desc']} ({tmpl['category']})")
+        print(f"    [c] 自定义描述\n    [s] 跳过此步骤\n    [d] 使用默认 (保持原样)\n")
+        choice = input(f"  请选择 [1-{len(templates)}/c/s/d] (默认d): ").strip().lower() or "d"
+        enhanced = dict(step)
+        if choice == "s": return None
+        elif choice == "c":
+            custom_desc = input(f"  请输入业务描述: ").strip()
+            if custom_desc: enhanced["desc"] = custom_desc; enhanced["_business_desc"] = custom_desc
+        elif choice.isdigit():
+            idx = int(choice) - 1; keys = list(templates.keys())
+            if 0 <= idx < len(keys):
+                selected_key = keys[idx]; tmpl = templates[selected_key]
+                enhanced["desc"] = tmpl["desc"]; enhanced["_business_type"] = selected_key; enhanced["_category"] = tmpl["category"]
+                if selected_key == "login":
+                    enhanced["wait_after"] = {"type": "time", "duration": 3000}
+                    enhanced["assertion"] = {"type": "url_contains", "expected": "", "confidence": "high", "note": "登录后预期URL特征"}
+                elif selected_key == "submit_form":
+                    enhanced["wait_after"] = {"type": "time", "duration": 2000}
+                    enhanced["assertions"] = [{"type": "toast_visible", "expected": "", "confidence": "high"}, {"type": "network_called", "url_pattern": "/api/*", "method": "POST", "response_code": 200}]
+        print()
+        return enhanced
+
+    def _get_relevant_templates(self, sub_type: str, target: str) -> Dict:
+        relevant = {}; target_lower = target.lower()
+        if any(kw in target_lower for kw in ["登录", "login"]): relevant["login"] = BUSINESS_TEMPLATES["login"]
+        if any(kw in target_lower for kw in ["提交", "保存", "申请", "确定", "确认"]): relevant["submit_form"] = BUSINESS_TEMPLATES["submit_form"]
+        if any(kw in target_lower for kw in ["查询", "搜索", "search"]): relevant["search_query"] = BUSINESS_TEMPLATES["search_query"]
+        if any(kw in target_lower for kw in ["菜单", "menu", "首页", "home"]): relevant["navigate_menu"] = BUSINESS_TEMPLATES["navigate_menu"]
+        if sub_type == "fill" and "fill_field" not in relevant: relevant["fill_field"] = BUSINESS_TEMPLATES["fill_field"]
+        if sub_type == "select" and "select_option" not in relevant: relevant["select_option"] = BUSINESS_TEMPLATES["select_option"]
+        if sub_type == "navigate" and "navigate_menu" not in relevant: relevant["navigate_menu"] = BUSINESS_TEMPLATES["navigate_menu"]
+        if not relevant: relevant["generic"] = {"desc": f"{target}", "action": "click", "category": "其他"}
+        return relevant
 
-    Returns:
-        YAML 步骤字典，或 None（跳过该事件）
-    """
-    etype = evt.get("type", "")
 
+def _event_to_step(evt: Dict, step_num: int) -> Optional[Dict]:
+    etype = evt.get("type", ""); desc = evt.get("desc", "") or ""
+    is_select = evt.get("isSelect", False); selected_text = evt.get("selectedText", "")
+    value = evt.get("value", ""); tag = evt.get("tag", "").upper()
     if etype == "click":
-        text = evt.get("text", "") or ""
-        ph = evt.get("ph", "") or ""
-        href = evt.get("href", "")
-        tag = evt.get("tag", "")
-
-        target = ph or text or href or f"{tag}元素"
-        if len(target) > 50:
-            target = target[:47] + "..."
-
-        step = {
-            "step": step_num,
-            "desc": f"点击'{target}'",
-            "action": "click",
-            "target": target,
-        }
-        if href and href.startswith("http"):
-            step["wait_after"] = {"type": "navigation", "timeout": 8000}
+        target = desc or f"{tag}"
+        if tag in MEANINGLESS_TARGETS and not desc and not evt.get("imgAlt"): return None
+        if len(target) > 60: target = target[:57] + "..."
+        step = {"step": step_num, "desc": f"点击'{target}'", "action": "click", "target": target}
         return step
-
     elif etype in ("input", "change"):
-        val = evt.get("val", "")
-        ph = evt.get("ph", "") or ""
-        target = ph or evt.get("name") or f"{evt.get('tag','')}输入框"
-        if not val:
-            return None
-
-        is_number = val.replace(".", "").replace("-", "").isdigit()
-        action = "fill"
-
-        step = {
-            "step": step_num,
-            "desc": f"填写'{target}'为'{val}'",
-            "action": action,
-            "target": target,
-            "value": val,
-        }
-        if is_number:
-            step["desc"] = f"填写数值'{val}'到{target}"
-        return step
-
+        if not value: return None
+        target = desc or f"字段{step_num}"
+        if is_select and selected_text: return {"step": step_num, "desc": f"选择'{selected_text}'", "action": "select_option", "target": target, "option": selected_text}
+        if value.replace(".", "").replace("-", "").isdigit():
+            return {"step": step_num, "desc": f"填写数值'{value}'到{target}", "action": "fill", "target": target, "value": value}
+        return {"step": step_num, "desc": f"填写'{target}'为'{value}'", "action": "fill", "target": target, "value": value}
     elif etype == "navigate":
-        return {
-            "step": step_num,
-            "desc": f"导航至 '{evt.get('url', '')}'",
-            "action": "navigate",
-            "url": evt.get("url", ""),
-            "wait_after": {"type": "time", "duration": 2000},
-        }
-
-    elif etype == "enter":
-        return None
-
+        return {"step": step_num, "desc": f"页面导航至 '{evt.get('url', '')[:50]}'", "action": "navigate", "url": evt.get("url", ""), "wait_after": {"type": "time", "duration": 2000}}
     return None
 
 
-# ============================================================
-# 录制模式主流程
-# ============================================================
-
-async def run_record_mode(output_path: str):
-    """交互式录制模式：监听用户操作 → 自动生成 YAML + .env
-
-    流程:
-      1. 通过 LoginManager 连接 Chrome DevTools MCP
-      2. 向页面注入 RECORDER_JS 事件监听器
-      3. 轮询 POLL_JS 获取捕获的事件，实时转换为步骤显示
-      4. 用户按 Enter 停止，调用 STOP_JS 收尾
-      5. 生成 YAML 用例文件和 .env 环境变量文件
-
-    Args:
-        output_path: 输出 YAML 文件路径（.env 同目录同名生成）
-    """
-    from .login_manager import LoginManager
-
-    print("=" * 60)
-    print("  Testcase Recorder v1.0")
-    print("  Interactive Browser Operation Recorder")
-    print("=" * 60)
-
-    lm = LoginManager()
-    session = await lm.create_session()
-    if not session:
-        print("[ERROR] Cannot connect to Chrome DevTools MCP service")
-        sys.exit(1)
-
-    print(f"\n[1/4] Browser connected... OK")
-
-    try:
-        await session.call_tool("evaluate_script", {"function": RECORDER_JS})
-        print("[2/4] Event listeners injected... OK")
-    except Exception as e:
-        print(f"[ERROR] Injection failed: {e}")
-        sys.exit(1)
-
-    print("""
-+--------------------------------------------------+
-|           You can operate in browser now!          |
-|                                                    |
-|   - Click buttons, links, inputs                   |
-|   - Fill forms, select dropdowns                   |
-|   - Navigate to different pages                     |
-|                                                    |
-|   Press Enter to stop recording                    |
-+--------------------------------------------------+
-""")
-
-    all_events: List[Dict] = []
-    step_num = 0
-    last_url = ""
-
-    import msvcrt
-    print("\n[3/4] Recording... (Press Enter to stop)\n")
-
-    while True:
-        await asyncio.sleep(1.5)
-        try:
-            raw = await session.call_tool("evaluate_script", {"function": POLL_JS})
-            data = json.loads(raw[0]["text"]) if isinstance(raw, list) else raw
-            if data.get("ok") and data.get("events"):
-                new_evts = data["events"]
-                for evt in new_evts:
-                    step_num += 1
-                    step = _event_to_step(evt, step_num)
-                    if step:
-                        all_events.append(step)
-                        action_icon = {"click": "[click]", "fill": "[fill]", "navigate": "[nav]"}.get(step["action"], "[act]")
-                        val_info = f" -> '{step.get('value','')}'" if step.get("value") else ""
-                        print(f"   [{step_num:02d}] {action_icon} {step['desc']}{val_info}")
-
-                if new_evts:
-                    last_url = new_evts[-1].get("url", last_url)
-        except Exception:
-            pass
-
-        if msvcrt.kbhit():
-            key = msvcrt.getwch()
-            if key == '\r':
-                break
-
-    print(f"\n[4/4] Stopped. Captured {len(all_events)} operation steps\n")
-
+def _merge_input_events(steps: List[Dict]) -> List[Dict]:
+    merged = []; i = 0
+    while i < len(steps):
+        s = steps[i]
+        if s.get("action") in ("input", "change", "fill") and s.get("value"):
+            last_val = s["value"]; last_target = s["target"]; j = i + 1
+            while j < len(steps):
+                nxt = steps[j]
+                if (nxt.get("action") in ("input", "change", "fill") and nxt.get("target") == last_target and nxt.get("value")):
+                    last_val = nxt["value"]; j += 1
+                else: break
+            merged.append({**s, "value": last_val}); i = j
+        else: merged.append(s); i += 1
+    return merged
+
+
+def _enrich_steps(steps: List[Dict]) -> List[Dict]:
+    enriched = []
+    for s in steps:
+        action = s.get("action", ""); step_copy = dict(s)
+        if action in ("fill", "click", "select_option"):
+            target = s.get("target", "")
+            safe_key = re.sub(r'[^a-zA-Z0-9]', '_', target).lower().strip('_') or f"step_{s['step']}"
+            step_copy["locator"] = {"uid_cache_key": f"{action}_{safe_key}"}
+        if "assertion" not in step_copy and "assertions" not in step_copy:
+            if action == "fill":
+                target_lower = s.get("target", "").lower()
+                if any(kw in target_lower for kw in ("密码", "password", "pass")): step_copy["isSensitive"] = True
+                step_copy["assertion"] = {"type": "field_filled", "expected": "已填写", "confidence": "high"}
+            elif action == "select_option": step_copy["assertion"] = {"type": "element_text", "expected": s.get("option", ""), "confidence": "high"}
+            elif action == "click": step_copy["assertion"] = {"type": "element_visible", "expected": "", "confidence": "medium", "note": "请验证"}
+            elif action == "navigate": step_copy["assertion"] = {"type": "text_contains", "expected": "", "confidence": "high", "note": "页面特征"}
+        for key in list(step_copy.keys()):
+            if key.startswith("_"): del step_copy[key]
+        enriched.append(step_copy)
+    return enriched
+
+
+async def run_record_mode(output_path: str, force_annotate: bool = False):
+    def log(msg: str): print(msg, flush=True)
+    log("=" * 60); log("  Testcase Recorder v2.0"); log("  模式: 交互式标注录制"); log("=" * 60)
+    global _global_chrome_process
+    log("\n[0/5] 启动 Chrome Incognito...")
+    _global_chrome_process = _launch_chrome_with_flags(port=9222)
+    if _global_chrome_process and _global_chrome_process.poll() is None: log(f"       Chrome 已启动 (PID: {_global_chrome_process.pid})")
+    else: log("       [WARN] Chrome 启动失败")
+    annotator = InteractiveAnnotator(force_annotate=force_annotate)
     try:
-        raw = await session.call_tool("evaluate_script", {"function": STOP_JS})
-        final_data = json.loads(raw[0]["text"]) if isinstance(raw, list) else raw
-        remaining = final_data.get("events", [])
-        for evt in remaining:
-            step_num += 1
-            step = _event_to_step(evt, step_num)
-            if step:
-                all_events.append(step)
-    except Exception:
-        pass
-
-    if not all_events:
-        print("[WARN] No operations captured, no files generated.")
-        return
-
-    yaml_data = {
-        "test_id": f"REC-{int(time.time())}-recorded",
-        "title": "Browser Operation Recording (auto-generated)",
-        "priority": "P1",
-        "tags": ["recorded", "auto-generated"],
-        "author": "Testcase Recorder v1.0",
-        "context_check": {
-            "login_url": last_url.split("/")[0] + "//" + last_url.split("/")[2] if "://" in last_url else "",
-            "home_indicator": "",
-            "credentials": {"username": "${TEST_USER}", "password": "${TEST_PASS}"},
-            "captcha_required": False,
-        },
-        "steps": all_events,
-        "teardown": [
-            {"action": "screenshot", "name": "recorded-result-{timestamp}.png", "fullPage": True}
-        ],
-    }
-
-    env_vars = set()
-    for step in all_events:
-        val = str(step.get("value", ""))
-        if "${" in val:
-            env_vars.add(val[2:-1])
-
-    out_dir = os.path.dirname(output_path)
-    os.makedirs(out_dir, exist_ok=True)
-
-    with open(output_path, 'w', encoding='utf-8') as f:
-        yaml.dump(yaml_data, f, allow_unicode=True, default_flow_style=False, sort_keys=False)
-    print(f"   OK  YAML: {output_path}")
-
-    env_path = output_path.rsplit('.', 1)[0] + ".env"
-    if env_vars:
-        with open(env_path, 'w', encoding='utf-8') as f:
-            f.write("# Recorded test case environment variables (auto-generated)\n")
-            for var in sorted(env_vars):
-                f.write(f"{var}=<please fill>\n")
-        print(f"   OK  ENV:  {env_path} ({len(env_vars)} vars pending)")
-    else:
-        with open(env_path, 'w', encoding='utf-8') as f:
-            f.write("# Recorded test case environment variables (auto-generated)\nTEST_USER=yuanye\nTEST_PASS=yuanye\n")
-        print(f"   OK  ENV:  {env_path}")
-
-    print(f"\n{'-' * 50}")
-    print(f"Recording complete!")
-    print(f"   Case ID : {yaml_data['test_id']}")
-    print(f"   Steps   : {len(all_events)}")
-    print(f"   Output:")
-    print(f"     - {output_path}")
-    print(f"     - {env_path}")
-    print(f"\nNext steps:")
-    print(f"   1. Edit .env file to fill real test data")
-    print(f"   2. Review and optimize YAML step descriptions and assertions")
-    print(f"   3. Run: python tests/testcase-ai.py {output_path}")
-
-
-__all__ = ["run_record_mode", "_event_to_step", "RECORDER_JS", "POLL_JS", "STOP_JS"]
+        async with stdio_client(SERVER_PARAMS) as (_read, _write):
+            async with ClientSession(_read, _write) as session:
+                await session.initialize(); log("\n[1/5] Browser connected... OK")
+                current_url = await _prepare_page(session, log)
+                if not current_url: return
+                inject_data = await _call(session, "evaluate_script", {"function": RECORDER_JS})
+                if isinstance(inject_data, dict) and inject_data.get("ok"): log("[2/5] Event listeners injected... OK")
+                log("\n+--------------------------------------------------+\n|  🎬 录制中...                                       |\n|  📌 重要操作时会弹出标注提示                           |\n|  ⌨️  Ctrl+C 结束录制                                  |\n+--------------------------------------------------\n")
+                all_events: List[Dict] = []; step_num = 0; poll_count = 0; consecutive_errors = 0; MAX_ERRORS = 5
+                log("[3/5] Recording...\n")
+                try:
+                    while True:
+                        await asyncio.sleep(1.5); poll_count += 1
+                        data = await _call(session, "evaluate_script", {"function": POLL_JS})
+                        if data is None:
+                            if poll_count % 10 == 0: log(f"   [poll#{poll_count}] (无响应)")
+                            consecutive_errors += 1
+                            if consecutive_errors >= MAX_ERRORS: log("\n   [!] 连接可能丢失"); break
+                            continue
+                        consecutive_errors = 0
+                        if not isinstance(data, dict) or not data.get("ok"):
+                            if poll_count % 20 == 0: log(f"   [poll#{poll_count}] Error: {data.get('error','?') if isinstance(data,dict) else '?'}")
+                            continue
+                        evts = data.get("events", [])
+                        if evts:
+                            for evt in evts:
+                                raw_step = _event_to_step(evt, step_num + 1)
+                                if not raw_step: continue
+                                category, sub_type = NodeClassifier.classify(evt)
+                                if category == "skip": log(f"   [-] 跳过: {raw_step.get('desc', '?')[:40]}"); continue
+                                step_num += 1; raw_step["step"] = step_num
+                                if annotator.should_annotate(category, sub_type):
+                                    annotated = await annotator.annotate(step_num, raw_step, category, sub_type)
+                                    if annotated is None: log(f"   [{step_num:02d}] ⏭️  用户跳过"); step_num -= 1; continue
+                                    final_step = annotated; log(f"   [{step_num:02d}] ✅ 已标注: {final_step.get('desc', '?')}")
+                                else:
+                                    final_step = raw_step
+                                    icon = {"click": "[click]", "fill": "[fill]", "select": "[sel]", "nav": "[nav]"}.get(final_step["action"], "[act]")
+                                    val = f" = '{final_step.get('value', '')}'" if final_step.get("value") else ""
+                                    opt = f" → {final_step.get('option', '')}" if final_step.get("option") else ""
+                                    log(f"   [{step_num:02d}] {icon} {final_step['desc']}{val}{opt}")
+                                all_events.append(final_step)
+                        elif poll_count % 15 == 0: log(f"   [poll#{poll_count}] ... ({data.get('elapsed','?')}s)")
+                except (KeyboardInterrupt, asyncio.CancelledError): log("\n\n[4/5] Recording stopped by user")
+                final_data = await _call(session, "evaluate_script", {"function": STOP_JS})
+                if isinstance(final_data, dict) and final_data.get("events"):
+                    for evt in final_data["events"]:
+                        step_num += 1; step = _event_to_step(evt, step_num)
+                        if step: step["step"] = step_num; all_events.append(step)
+                log(f"\n       Captured: {len(all_events)} steps")
+                if not all_events: log("[WARN] No operations captured."); return
+                merged = _merge_input_events(all_events); final_steps = _enrich_steps(merged)
+                nav_step = {"step": 1, "desc": "打开目标页面", "action": "navigate", "url": current_url, "wait_after": {"type": "time", "duration": 2000}, "assertion": {"type": "text_contains", "expected": "", "confidence": "high", "note": "页面标题或关键文本"}}
+                for s in final_steps: s["step"] = s["step"] + 1
+                final_steps.insert(0, nav_step)
+                log(f"       Final: {len(final_steps)} steps\n")
+                base_url = ""; parts = current_url.split("/") if "://" in current_url else []
+                if len(parts) >= 3: base_url = parts[0] + "//" + parts[2]
+                yaml_data = {"test_id": f"REC-{int(time.time())}-annotated", "title": f"交互式标注测试用例 - {os.path.basename(output_path)}", "priority": "P1", "tags": ["recorded", "annotated", "v2.0"], "author": "Testcase Recorder v2.0 (Interactive)", "context_check": {"login_url": base_url, "home_indicator": "", "credentials": {"username": "${TEST_USER}", "password": "${TEST_PASS}"}, "captcha_required": False}, "steps": final_steps, "teardown": [{"action": "screenshot", "name": "result-{timestamp}.png", "fullPage": True}], "metadata": {"created_by": "testcase-recorder v2.0", "annotation_count": annotator.annotation_count, "total_steps": len(final_steps), "note": "使用交互式标注模式生成，重要节点已由人工确认"}}
+                out_dir = os.path.dirname(output_path); os.makedirs(out_dir, exist_ok=True)
+                with open(output_path, 'w', encoding='utf-8') as f: yaml.dump(yaml_data, f, allow_unicode=True, default_flow_style=False, sort_keys=False)
+                log(f"   OK  YAML: {output_path}")
+                env_path = output_path.rsplit('.', 1)[0] + ".env"; extracted_env: Dict[str, str] = {}
+                for s in final_steps:
+                    val = s.get("value", ""); target = s.get("target", "").lower()
+                    if not val or "${" in val: continue
+                    if any(kw in target for kw in ("用户名", "username")): extracted_env["TEST_USER"] = val
+                    elif any(kw in target for kw in ("密码", "password")): extracted_env["TEST_PASS"] = val
+                with open(env_path, 'w', encoding='utf-8') as f:
+                    f.write(f"# Generated by Testcase Recorder v2.0 at {time.strftime('%Y-%m-%d %H:%M:%S')}\n\n")
+                    for k, v in sorted(extracted_env.items()): f.write(f"{k}={v}\n")
+                log(f"   OK  ENV:  {env_path}")
+                log(f"\n{'=' * 50}\nRecording complete!\n   Steps  : {len(final_steps)}\n   Annotated: {annotator.annotation_count} nodes\n   Output :\n     - {output_path}\n     - {env_path}")
+    except Exception as e: log(f"\n[FATAL] {type(e).__name__}: {e}"); import traceback; traceback.print_exc()
+    finally:
+        if _global_chrome_process and _global_chrome_process.poll() is None:
+            log("\n[清理] 关闭 Chrome..."); _global_chrome_process.terminate()
+
+
+async def _prepare_page(session: ClientSession, log) -> Optional[str]:
+    pages_raw = await session.call_tool("list_pages", {})
+    pages_text = _extract_text(pages_raw)
+    pages_data = json.loads(pages_text) if isinstance(pages_text, str) else (pages_text if isinstance(pages_text, list) else None)
+    current_url = "about:blank"
+    if isinstance(pages_data, list) and len(pages_data) > 0:
+        for p in pages_data:
+            if isinstance(p, dict):
+                url = p.get("url", ""); title = p.get("title", "") or url; pid = p.get("pageId", 0)
+                is_blank = "about:blank" in url; log(f"       Tab[{pid}] {title}{' ← 当前' if is_blank else ''}")
+                if not is_blank and url.startswith("http"): current_url = url
+    if current_url == "about:blank":
+        target_url = input("\n       [!] 当前空白页，请输入目标 URL\n       URL> ").strip()
+        if target_url:
+            if not target_url.startswith(("http://", "https://")): target_url = "http://" + target_url
+            log(f"\n       导航至: {target_url}")
+            await _call(session, "evaluate_script", {"function": f"() => {{ location.href = '{target_url}'; return location.href; }}"})
+            log("       等待加载..."); await asyncio.sleep(3)
+            check = await _call(session, "evaluate_script", {"function": "() => location.href"})
+            if isinstance(check, str): current_url = check
+    verify = await _call(session, "evaluate_script", {"function": "() => location.href"})
+    if isinstance(verify, str): current_url = verify
+    log(f"\n[1/5] Target: {current_url}"); return current_url
+
+
+__all__ = ["run_record_mode"]