release: runtime_env marker, SKILL v1.0.51

SKILL.md

@@ -1,7 +1,7 @@
 ---
 name: 账号管理
 description: 多平台多账号管理。管理各平台账号与Chrome Profile的对应关系，供publisher类Skill调用获取账号信息。
-version: 1.0.4
+version: 1.0.51
 author: 深圳匠厂科技有限公司
 metadata:
   openclaw:
@@ -14,6 +14,10 @@ allowed-tools:
 
 # 账号管理
 
+## JIANGCHANG 共享 Python（匠厂桌面）
+
+依赖由宿主安装到 `{JIANGCHANG_DATA_ROOT}/python-runtime/.venv`（技能市场安装时自动 `uv sync`）。勿在技能目录内创建 `.venv`。Playwright 浏览器目录：`{JIANGCHANG_DATA_ROOT}/playwright-browsers`（`PLAYWRIGHT_BROWSERS_PATH`）。
+
 ## 使用时机
 
 当用户发送以下内容的时候触发本Skill：
@@ -35,7 +39,7 @@ python3 {baseDir}/scripts/main.py add 知乎 13800138000
 python3 {baseDir}/scripts/main.py open <id>
 ```
 
-### 登录并自动检测、写入数据库
+### 登录并自动检测（成功则刷新该账号 `updated_at`，不在库中保存「是否已登录」）
 ```bash
 python3 {baseDir}/scripts/main.py login <id>
 ```
@@ -53,14 +57,13 @@ python3 {baseDir}/scripts/main.py list all
 python3 {baseDir}/scripts/main.py list 搜狐号
 ```
 
-### 跨技能：单行 JSON（get / list-json / pick-web / pick-logged-in / set-login-status）
+### 跨技能：单行 JSON（get / list-json / pick-web）
 ```bash
 python3 {baseDir}/scripts/main.py get <id>
 python3 {baseDir}/scripts/main.py list-json all --limit 50
 python3 {baseDir}/scripts/main.py pick-web 搜狐号
-python3 {baseDir}/scripts/main.py set-login-status <id> 1
 ```
-约定与字段说明见 `references/INTEGRATION.md`；错误前缀见 `references/ERRORS.md`。
+约定与字段说明见 `references/INTEGRATION.md`；错误前缀见 `references/ERRORS.md`。是否已登录由调用方网页会话处理，本 skill 不做编排用预检。
 
 ## 扩展文档（便于 AI 精确调用）
 

assets/README.md

@@ -1,6 +1,6 @@
 # assets
 
 - `examples/`：CLI 成功输出形状示例（虚构路径与数据）。
-- `schemas/`：`get` / `list-json` / `pick-*` 单条对象的轻量 JSON Schema（`additionalProperties: true` 允许扩展字段）。
+- `schemas/`：`get` / `list-json` / `pick-web` 单条对象的轻量 JSON Schema（`additionalProperties: true` 允许扩展字段）。
 
 详细命令与错误码见仓库根目录 `references/`。

assets/examples/get-response.json

@@ -5,8 +5,6 @@
   "phone": "13800138000",
   "profile_dir": "D:/jiangchang-data/_anon/account-manager/profiles/搜狐号/13800138000",
   "url": "https://mp.sohu.com",
-  "login_status": 1,
-  "last_login_at": "2026-04-01T12:00:00",
   "created_at": "2026-03-01T10:00:00",
   "updated_at": "2026-04-01T12:00:00"
 }

assets/examples/list-json-array.json

@@ -6,8 +6,6 @@
     "phone": "13900001111",
     "profile_dir": "D:/jiangchang-data/_anon/account-manager/profiles/知乎/13900001111",
     "url": "https://www.zhihu.com",
-    "login_status": 0,
-    "last_login_at": null,
     "created_at": "2026-04-02T09:00:00",
     "updated_at": "2026-04-02T09:00:00"
   },
@@ -18,8 +16,6 @@
     "phone": "13800138000",
     "profile_dir": "D:/jiangchang-data/_anon/account-manager/profiles/搜狐号/13800138000",
     "url": "https://mp.sohu.com",
-    "login_status": 1,
-    "last_login_at": "2026-04-01T12:00:00",
     "created_at": "2026-03-01T10:00:00",
     "updated_at": "2026-04-01T12:00:00"
   }

assets/schemas/account-record.schema.json

@@ -2,9 +2,9 @@
   "$schema": "https://json-schema.org/draft/2020-12/schema",
   "$id": "https://openclaw.local/account-manager/account-record.schema.json",
   "title": "AccountRecord",
-  "description": "account-manager get / list-json / pick-* 返回的单条账号对象（核心字段；extra_json 合并键不枚举）",
+  "description": "account-manager get / list-json / pick-web 返回的单条账号对象（核心字段；extra_json 合并键不枚举）",
   "type": "object",
-  "required": ["id", "name", "platform", "phone", "profile_dir", "url", "login_status", "created_at", "updated_at"],
+  "required": ["id", "name", "platform", "phone", "profile_dir", "url", "created_at", "updated_at"],
   "properties": {
     "id": { "type": "integer" },
     "name": { "type": "string" },
@@ -12,11 +12,6 @@
     "phone": { "type": "string", "description": "归一后的数字串，可能为空字符串" },
     "profile_dir": { "type": "string", "description": "Playwright 用户数据目录绝对路径" },
     "url": { "type": "string", "format": "uri" },
-    "login_status": { "type": "integer", "enum": [0, 1] },
-    "last_login_at": {
-      "type": ["string", "null"],
-      "description": "本地时区 ISO8601 字符串；未登录过为 null"
-    },
     "created_at": { "type": ["string", "null"] },
     "updated_at": { "type": ["string", "null"] }
   },

references/CLI.md

@@ -6,7 +6,7 @@
 
 - **平台参数**：可为 `scripts/util/platforms.py` 中配置的**中文展示名、别名或英文键**（如 `sohu`、`搜狐号`）。详见 [PLATFORMS.md](PLATFORMS.md)。
 - **数据与路径**：库文件、profile 目录由环境变量决定，网关与本地终端不一致时会「看不到同一份数据」。详见 [RUNTIME.md](RUNTIME.md)。
-- **机器可读子命令**：`get`、`list-json`、`pick-logged-in`、`pick-web`、`set-login-status` 的成功/失败约定见 [INTEGRATION.md](INTEGRATION.md)。
+- **机器可读子命令**：`get`、`list-json`、`pick-web` 的成功/失败约定见 [INTEGRATION.md](INTEGRATION.md)。
 - **错误码前缀**：见 [ERRORS.md](ERRORS.md)。
 
 > 说明：当前版本 **未实现** `health` / `version` 子命令；与工作区其它技能的统一约定可对齐后另补。
@@ -56,33 +56,15 @@ python main.py list-json <平台|all|全部> [--limit N]
 - 默认 `limit=200`，实际 clamp 为 `1..500`。
 - 排序：按 `updated_at` 倒序。
 
-### `pick-logged-in` — 该平台**已登录**的一条账号（单行 JSON）
-
-```bash
-python main.py pick-logged-in <平台>
-```
-
-- 筛选：`login_status = 1`，按 `last_login_at` 优先。
-- 失败：首行以 `ERROR:` 开头（**勿当 JSON 解析**）。
-
 ### `pick-web` — 网页自动化候选账号（单行 JSON）
 
 ```bash
 python main.py pick-web <平台>
 ```
 
-- 优先同 `pick-logged-in`；若无已登录，则取该平台 `updated_at` 最新一条。
+- 按该平台 `updated_at`、`created_at` 倒序选一条；**不**表示当前浏览器已登录。
 - 失败：首行 `ERROR:`。
 
-### `set-login-status` — 回写登录态（跨技能）
-
-```bash
-python main.py set-login-status <id> <0|1>
-```
-
-- 成功：`stdout` 首行 `OK:SET_LOGIN_STATUS` 或 `OK:CLEARED_LOGIN_STATUS`，退出码 `0`。
-- 失败：`stdout` 首行 `ERROR:...`，退出码 `1`。
-
 ### `open` — 仅打开浏览器（**不写库**）
 
 ```bash
@@ -91,13 +73,15 @@ python main.py open <id>
 
 - 需本机 Chrome/Edge + Playwright；详见运行环境说明。
 
-### `login` — 打开浏览器并**自动检测登录**，写回 `login_status`
+### `login` — 打开浏览器并**自动检测登录**（成功则刷新 `updated_at`）
 
 ```bash
 python main.py login <id>
 ```
 
 - 依赖 Playwright；超时与轮询间隔等见 [RUNTIME.md](RUNTIME.md)。
+- **不**在库中持久化「是否已登录」；仅刷新 `updated_at` 供 `pick-web` 等排序参考。
+- 跨技能编排（如 llm-manager 网页生成）**不再**通过本 skill 单独做「登录预检」；由业务侧在同一浏览器会话内处理未登录态。
 
 ### `delete` — 删除库记录并尽量删除 profile 目录
 

references/ERRORS.md

@@ -1,13 +1,6 @@
 # 终端前缀约定（account-manager）
 
-约定：**机器解析只依赖 `stdout` 首行**时，应先判断首行是否以 `ERROR:` 或 `OK:` 开头；多行说明可忽略或仅作展示。
-
-## `OK:`（成功标记）
-
-| 首行 | 场景 |
-|------|------|
-| `OK:SET_LOGIN_STATUS` | `set-login-status <id> 1` 成功 |
-| `OK:CLEARED_LOGIN_STATUS` | `set-login-status <id> 0` 成功 |
+约定：**机器解析只依赖 `stdout` 首行**时，应先判断首行是否以 `ERROR:` 开头；多行说明可忽略或仅作展示。
 
 ## `ERROR:` — CLI 参数 / 路由
 
@@ -19,9 +12,7 @@
 | `ERROR:CLI_LOGIN_MISSING_ID` | `login` 缺少 id |
 | `ERROR:CLI_DELETE_MISSING_ARGS` | `delete` 参数不足 |
 | `ERROR:CLI_DELETE_BAD_MODE` | `delete` 第二段不是 `id`/`platform` |
-| `ERROR:CLI_SET_LOGIN_STATUS_MISSING_ARGS` | `set-login-status` 参数不足 |
 | `ERROR:CLI_LIST_JSON_MISSING_PLATFORM` | `list-json` 缺少平台参数 |
-| `ERROR:CLI_PICK_LOGGED_IN_MISSING_ARGS` | `pick-logged-in` 缺少平台 |
 | `ERROR:CLI_PICK_WEB_MISSING_ARGS` | `pick-web` 缺少平台 |
 | `ERROR:CLI_UNKNOWN_OR_BAD_ARGS` | 无法识别的子命令或参数组合 |
 
@@ -38,10 +29,7 @@
 | `ERROR:PHONE_INVALID` | 手机号格式不符合大陆 11 位规则 |
 | `ERROR:DUPLICATE_PHONE_PLATFORM` | 同平台下山手机号已存在 |
 | `ERROR:DELETE_INVALID_ID` | 删除 id 非法 |
-| `ERROR:NO_LOGGED_IN_ACCOUNT` | `pick-logged-in` 无 `login_status=1` 的账号 |
 | `ERROR:NO_ACCOUNT` | `pick-web` 该平台无任何记录 |
-| `ERROR:INVALID_ACCOUNT_ID` | `set-login-status` 的 id 非数字 |
-| `ERROR:INVALID_STATUS` | `set-login-status` 状态不是 0/1 |
 
 ## `ERROR:` — 浏览器 / 环境
 
@@ -53,4 +41,4 @@
 ## 非前缀输出
 
 - **成功添加账号**、**删除成功**等可能包含 `✅` 等人类可读行，**不宜**作为稳定协议解析。
-- **`get` / `list-json` / `pick-*` 成功**：`stdout` 应为**单行 JSON**（失败时首行 `ERROR:`，勿当 JSON）。
+- **`get` / `list-json` / `pick-web` 成功**：`stdout` 应为**单行 JSON**（失败时首行 `ERROR:`，勿当 JSON）。

references/INTEGRATION.md

@@ -10,30 +10,27 @@
 |------|------|----------------|
 | 取单条账号 JSON | `get <id>` | 单行 JSON 对象 |
 | 批量账号 JSON | `list-json <平台或all> [--limit N]` | 单行 JSON 数组 |
-| 只要已登录 | `pick-logged-in <平台>` | 单行 JSON 对象 |
-| 自动化候选（优先已登录） | `pick-web <平台>` | 单行 JSON 对象 |
-| 回写登录态 | `set-login-status <id> <0或1>` | 首行 `OK:...` |
+| 网页自动化候选账号 | `pick-web <平台>` | 单行 JSON 对象（按 `updated_at` 选一条，**不**表示当前已登录） |
 
 ## 解析协议（强建议）
 
 1. 读取子进程 **stdout** 全文或首行。
 2. 若首行以 **`ERROR:`** 开头 → 失败，不要 `json.loads` 整段 stdout。
-3. 若首行以 **`OK:`** 开头 → `set-login-status` 成功（可再检查退出码 `0`）。
-4. **`get` / `list-json` / `pick-logged-in` / `pick-web`**：成功时应对**第一行**做 `json.loads`；后续行应不存在，若有则视为非协议内容。
+3. **`get` / `list-json` / `pick-web`**：成功时应对**第一行**做 `json.loads`；后续行应不存在，若有则视为非协议内容。
 
 ## JSON 对象字段（与 `get` 一致）
 
 典型字段见仓库 `assets/examples/get-response.json` 与 `assets/schemas/account-record.schema.json`。
 
-常见键名包括：`id`, `name`, `platform`, `phone`, `profile_dir`, `url`, `login_status`, `last_login_at`, `created_at`, `updated_at`；`extra_json` 列中的扁平扩展可能并入同一对象。
+常见键名包括：`id`, `name`, `platform`, `phone`, `profile_dir`, `url`, `created_at`, `updated_at`；`extra_json` 列中的扁平扩展可能并入同一对象。
+
+**库中不再持久化「是否已登录」**；编排侧在网页自动化中自行处理登录（与 `pick-web` / `get` 返回的 `profile_dir` 同一浏览器会话）。
 
 ## Publisher / 自动化流水线示例
 
 1. `pick-web 搜狐号`（或目标平台）→ 解析 JSON → 取 `id`、`profile_dir`、`url`。
-2. 若需确认已登录：检查 `login_status == 1`，否则可引导用户执行 `login <id>` 或先 `open <id>` 人工登录。
-3. 外部脚本在确认登录后，可调用 `set-login-status <id> 1` 保持与库一致（若你们自有检测逻辑）。
+2. 使用 `profile_dir` 启动持久化浏览器上下文，在业务页面内完成或等待登录。
 
 ## 退出码
 
-- **`set-login-status`**：失败时进程以 **`1`** 退出（见实现）。
-- 其它子命令多数为 **`0`**；若需严格编排，请以 **stdout 首行前缀** 与 **JSON 可解析性** 为准，不要仅依赖退出码。
+请以 **stdout 首行前缀** 与 **JSON 可解析性** 为准；不要仅依赖退出码。

references/README.md

@@ -7,7 +7,7 @@
 | [CLI.md](CLI.md) | 子命令、参数、行为说明 |
 | [PLATFORMS.md](PLATFORMS.md) | 平台英文键、中文名、别名 |
 | [RUNTIME.md](RUNTIME.md) | 环境变量、数据路径、日志、登录调参 |
-| [ERRORS.md](ERRORS.md) | `ERROR:` / `OK:` 前缀一览 |
+| [ERRORS.md](ERRORS.md) | `ERROR:` 前缀一览 |
 | [INTEGRATION.md](INTEGRATION.md) | 跨 Skill 调用与 JSON 解析约定 |
 
 示例 JSON 与 Schema 见仓库 `assets/`。

scripts/cli/app.py

@@ -9,11 +9,9 @@ from service.account_service import (
     cmd_get,
     cmd_list,
     cmd_list_json,
-    cmd_pick_logged_in,
     cmd_pick_web,
-    cmd_set_login_status,
 )
-from service.browser_service import cmd_login, cmd_open, cmd_wait_login
+from service.browser_service import cmd_login, cmd_open
 from util.constants import LOG_LOGGER_NAME, SKILL_SLUG
 from util.logging_config import get_skill_logger, setup_skill_logging
 from util.platforms import _platform_list_cn_for_help, resolve_platform_key
@@ -25,14 +23,11 @@ def _cli_print_full_usage() -> None:
     print("用法概览（将 main.py 换为你的路径）：")
     print("  python main.py add <平台中文名> <手机号>")
     print("  python main.py list [平台|all|全部]")
-    print("  python main.py pick-logged-in <平台>   # 跨技能用：仅已登录账号，一行 JSON 或 ERROR")
-    print("  python main.py pick-web <平台>         # 跨技能用：优先已登录，否则最新一条账号")
+    print("  python main.py pick-web <平台>         # 跨技能用：按最近更新选一条账号，一行 JSON")
     print("  python main.py list-json <平台|all> [--limit N]   # 跨技能用：一行 JSON 数组，元素同 get")
     print("  python main.py get <id>")
-    print("  python main.py set-login-status <id> <0|1>   # 跨技能回写登录态")
     print("  python main.py open <id>")
     print("  python main.py login <id>")
-    print("  python main.py wait-login <id> [--force]   # 跨技能：限时内等待网页登录（默认 180s）；--force 忽略库内已登录标记")
     print("  python main.py delete id <id>")
     print("  python main.py delete platform <平台>")
     print("  python main.py delete platform <平台> <手机号>")
@@ -109,13 +104,6 @@ def main() -> None:
             _cli_fail_need_account_id("get")
             sys.exit(1)
         cmd_get(sys.argv[2])
-    elif cmd == "set-login-status":
-        if len(sys.argv) < 4:
-            print("ERROR:CLI_SET_LOGIN_STATUS_MISSING_ARGS")
-            print("用法：python main.py set-login-status <account_id> <0|1>")
-            print("说明：供 llm-manager 等脚本回写登录态；成功时 stdout 首行以 OK: 开头。")
-            sys.exit(1)
-        cmd_set_login_status(sys.argv[2], sys.argv[3])
     elif cmd == "list-json":
         if len(sys.argv) < 3:
             print("ERROR:CLI_LIST_JSON_MISSING_PLATFORM")
@@ -133,18 +121,11 @@ def main() -> None:
                 except ValueError:
                     pass
         cmd_list_json(platform_arg, limit=lim)
-    elif cmd == "pick-logged-in":
-        if len(sys.argv) < 3:
-            print("ERROR:CLI_PICK_LOGGED_IN_MISSING_ARGS")
-            print("用法：python main.py pick-logged-in <平台中文名或英文键>")
-            print("说明：供 llm-manager 等技能查询该平台已登录账号；成功时 stdout 仅一行 JSON。")
-            sys.exit(1)
-        cmd_pick_logged_in(sys.argv[2])
     elif cmd == "pick-web":
         if len(sys.argv) < 3:
             print("ERROR:CLI_PICK_WEB_MISSING_ARGS")
             print("用法：python main.py pick-web <平台中文名或英文键>")
-            print("说明：供 llm-manager 等；优先已登录账号，否则返回该平台最新一条账号（未登录也可用于打开浏览器登录）。")
+            print("说明：按 updated_at 选一条账号；是否已登录由调用方网页会话自行处理。")
             sys.exit(1)
         cmd_pick_web(sys.argv[2])
     elif cmd == "open":
@@ -152,17 +133,6 @@ def main() -> None:
             _cli_fail_need_account_id("open")
             sys.exit(1)
         cmd_open(sys.argv[2])
-    elif cmd == "wait-login":
-        av = [a for a in sys.argv[2:] if a]
-        force = "--force" in av or "-f" in av
-        av = [a for a in av if a not in ("--force", "-f")]
-        if len(av) < 1:
-            print("ERROR:CLI_WAIT_LOGIN_MISSING_ID")
-            print("用法：python main.py wait-login <账号 id> [--force]")
-            print("说明：供 llm-manager / 搜狐发布等编排；未登录时打开浏览器，限时内等待登录或用户关窗即结束。")
-            print("--force：即使库内已标为登录也重新打开浏览器检测（Cookie 失效时用）。")
-            sys.exit(1)
-        sys.exit(cmd_wait_login(av[0], force=force))
     elif cmd == "login":
         if len(sys.argv) < 3:
             _cli_fail_need_account_id("login")

scripts/db/accounts_repo.py

@@ -46,7 +46,7 @@ def get_account_by_id(account_id):
         cur = conn.cursor()
         cur.execute(
             """
-            SELECT id, name, platform, phone, profile_dir, url, login_status, last_login_at, extra_json,
+            SELECT id, name, platform, phone, profile_dir, url, extra_json,
                    created_at, updated_at
             FROM accounts WHERE id = ?
             """,
@@ -62,14 +62,12 @@ def get_account_by_id(account_id):
             "phone": row[3] or "",
             "profile_dir": (row[4] or "").strip(),
             "url": row[5] or PLATFORM_URLS.get(row[2], ""),
-            "login_status": int(row[6] or 0),
-            "last_login_at": _unix_to_iso_local(row[7]),
-            "created_at": _unix_to_iso_local(row[9]),
-            "updated_at": _unix_to_iso_local(row[10]),
+            "created_at": _unix_to_iso_local(row[7]),
+            "updated_at": _unix_to_iso_local(row[8]),
         }
-        if row[8]:
+        if row[6]:
             try:
-                extra = json.loads(row[8])
+                extra = json.loads(row[6])
                 if isinstance(extra, dict):
                     acc.update(extra)
             except Exception:
@@ -79,21 +77,13 @@ def get_account_by_id(account_id):
         conn.close()
 
 
-def mark_login_status(account_id, success: bool):
+def touch_account_updated_at(account_id) -> None:
+    """登录 DOM 检测成功等场景下刷新 updated_at，供 pick-web 等按最近活跃选号。"""
     now = _now_unix()
     conn = get_conn()
     try:
         cur = conn.cursor()
-        if success:
-            cur.execute(
-                "UPDATE accounts SET login_status = 1, last_login_at = ?, updated_at = ? WHERE id = ?",
-                (now, now, account_id),
-            )
-        else:
-            cur.execute(
-                "UPDATE accounts SET login_status = 0, updated_at = ? WHERE id = ?",
-                (now, account_id),
-            )
+        cur.execute("UPDATE accounts SET updated_at = ? WHERE id = ?", (now, account_id))
         conn.commit()
     finally:
         conn.close()
@@ -105,8 +95,8 @@ def fetch_list_rows(platform_key: str, limit: int):
     try:
         cur = conn.cursor()
         sql = (
-            "SELECT id, name, platform, phone, profile_dir, url, login_status, "
-            "last_login_at, extra_json, created_at, updated_at FROM accounts "
+            "SELECT id, name, platform, phone, profile_dir, url, "
+            "extra_json, created_at, updated_at FROM accounts "
         )
         if platform_key == "all":
             cur.execute(sql + "ORDER BY created_at DESC, id DESC LIMIT ?", (int(limit),))
@@ -140,50 +130,21 @@ def fetch_ids_for_list_json(platform_key: str, lim: int):
         conn.close()
 
 
-def fetch_pick_logged_in_id(platform_key: str) -> Optional[int]:
-    conn = get_conn()
-    try:
-        cur = conn.cursor()
-        cur.execute(
-            """
-            SELECT id FROM accounts
-            WHERE platform = ? AND login_status = 1
-            ORDER BY (last_login_at IS NULL), last_login_at DESC
-            LIMIT 1
-            """,
-            (platform_key,),
-        )
-        row = cur.fetchone()
-        return int(row[0]) if row else None
-    finally:
-        conn.close()
-
-
 def fetch_pick_web_candidate_id(platform_key: str) -> Optional[int]:
+    """该平台一条账号：按 updated_at、created_at 倒序，与是否登录无关。"""
     conn = get_conn()
     try:
         cur = conn.cursor()
         cur.execute(
             """
             SELECT id FROM accounts
-            WHERE platform = ? AND login_status = 1
-            ORDER BY (last_login_at IS NULL), last_login_at DESC
+            WHERE platform = ?
+            ORDER BY updated_at DESC, created_at DESC, id DESC
             LIMIT 1
             """,
             (platform_key,),
         )
         row = cur.fetchone()
-        if not row:
-            cur.execute(
-                """
-                SELECT id FROM accounts
-                WHERE platform = ?
-                ORDER BY updated_at DESC, id DESC
-                LIMIT 1
-                """,
-                (platform_key,),
-            )
-            row = cur.fetchone()
         return int(row[0]) if row else None
     finally:
         conn.close()
@@ -216,8 +177,8 @@ def insert_account_row(
     cur = conn.cursor()
     cur.execute(
         """
-        INSERT INTO accounts (name, platform, phone, profile_dir, url, login_status, last_login_at, extra_json, created_at, updated_at)
-        VALUES (?, ?, ?, ?, ?, 0, NULL, NULL, ?, ?)
+        INSERT INTO accounts (name, platform, phone, profile_dir, url, extra_json, created_at, updated_at)
+        VALUES (?, ?, ?, ?, ?, NULL, ?, ?)
         """,
         (name, platform_key, phone_store, profile_dir, url, now, now),
     )

scripts/db/schema.py

@@ -1,7 +1,7 @@
 # SQLite 无独立 DATETIME 类型: 时间统一存 INTEGER Unix 秒 (UTC), 查询/JSON 再转 ISO8601.
 # 下列 DDL 含注释, 会原样写入 sqlite_master, 便于 Navicat / DBeaver 等查看建表语句.
 ACCOUNTS_TABLE_SQL = """
-/* 多平台账号表: 与本地 Chromium/Edge 用户数据目录 (profile) 绑定, 供发布类技能读取登录态 */
+/* 多平台账号表: 与本地 Chromium/Edge 用户数据目录 (profile) 绑定 */
 CREATE TABLE accounts (
     id INTEGER PRIMARY KEY AUTOINCREMENT, -- 账号主键 (自增)
     name TEXT NOT NULL, -- 展示名称, 如「搜狐1号」
@@ -9,8 +9,6 @@ CREATE TABLE accounts (
     phone TEXT, -- 可选绑定手机号
     profile_dir TEXT, -- Playwright 用户数据目录 (绝对路径); 默认可读结构 profiles/<平台展示名>/<手机号>/
     url TEXT, -- 平台入口或登录页 URL
-    login_status INTEGER NOT NULL DEFAULT 0, -- 展示/排序用尽力同步；编排上请以 Playwright DOM 检测为准，勿单独依赖本字段
-    last_login_at INTEGER, -- 最近一次登录成功时间, Unix 秒 UTC; 未登录过为 NULL
     extra_json TEXT, -- 扩展字段 JSON
     created_at INTEGER NOT NULL, -- 记录创建时间, Unix 秒 UTC
     updated_at INTEGER NOT NULL -- 记录最后更新时间, Unix 秒 UTC

scripts/jiangchang_skill_core/runtime_env.py

@@ -2,6 +2,11 @@
 JIANGCHANG_* 数据根与用户目录：解析规则 + 可选本地 CLI 默认值。
 
 各技能 scripts/jiangchang_skill_core/ 为发布用副本，与 kit 保持同步。
+
+宿主在 skills.entries 与 Gateway 中注入：
+- PATH 前缀：{JIANGCHANG_DATA_ROOT}/python-runtime/.venv/(Scripts|bin)
+- PLAYWRIGHT_BROWSERS_PATH、VIRTUAL_ENV、JIANGCHANG_PYTHON_EXE
+技能勿在仓库内维护独立 .venv；依赖以 jiangchang-platform-kit/python-runtime 锁文件为准。
 """
 
 from __future__ import annotations
@@ -18,6 +23,9 @@ DEFAULT_LOCAL_USER_ID = "10032"
 # Windows 下未设置 JIANGCHANG_DATA_ROOT 时的默认盘路径
 WIN_DEFAULT_DATA_ROOT = r"D:\jiangchang-data"
 
+# 匠厂桌面宿主应用根（未设置 JIANGCHANG_APP_ROOT 时 Windows 兜底；技能一般在 {APP}/skills/ 下并列）
+WIN_DEFAULT_JIANGCHANG_APP_ROOT = r"D:\AI\jiangchang"
+
 
 def platform_default_data_root() -> str:
     if sys.platform == "win32":
@@ -36,6 +44,81 @@ def get_user_id() -> str:
     return (os.getenv("JIANGCHANG_USER_ID") or "").strip() or "_anon"
 
 
+def _looks_like_skills_root(path: str) -> bool:
+    if not path or not os.path.isdir(path):
+        return False
+    for marker in (
+        "llm-manager",
+        "content-manager",
+        "account-manager",
+        "sohu-publisher",
+        "toutiao-publisher",
+        "logistics-tracker",
+        "api-key-vault",
+    ):
+        if os.path.isdir(os.path.join(path, marker)):
+            return True
+    return False
+
+
+def get_skills_root() -> str:
+    """
+    并列技能安装目录（其下为「技能 slug」子目录）。
+
+    优先级：
+    1) JIANGCHANG_SKILLS_ROOT
+    2) CLAW_SKILLS_ROOT
+    3) JIANGCHANG_APP_ROOT：若 {APP}/skills 存在且像技能根则用之；否则若 APP 下已有技能子目录则 APP 即根
+    4) Windows：默认 D:\\AI\\jiangchang 同上规则
+    5) 其他平台：~/.openclaw/skills
+    """
+    for key in ("JIANGCHANG_SKILLS_ROOT", "CLAW_SKILLS_ROOT"):
+        v = (os.getenv(key) or "").strip()
+        if v:
+            return os.path.normpath(v)
+
+    app = (os.getenv("JIANGCHANG_APP_ROOT") or "").strip()
+    if sys.platform == "win32" and not app:
+        app = WIN_DEFAULT_JIANGCHANG_APP_ROOT
+    if app:
+        nested = os.path.join(app, "skills")
+        if _looks_like_skills_root(nested):
+            return os.path.normpath(nested)
+        if _looks_like_skills_root(app):
+            return os.path.normpath(app)
+
+    if sys.platform == "win32":
+        nested = os.path.join(WIN_DEFAULT_JIANGCHANG_APP_ROOT, "skills")
+        if _looks_like_skills_root(nested):
+            return os.path.normpath(nested)
+        if _looks_like_skills_root(WIN_DEFAULT_JIANGCHANG_APP_ROOT):
+            return os.path.normpath(WIN_DEFAULT_JIANGCHANG_APP_ROOT)
+
+    return os.path.normpath(os.path.join(os.path.expanduser("~"), ".openclaw", "skills"))
+
+
+def get_sibling_skills_root(skill_scripts_dir: str | None = None) -> str:
+    """
+    编排子进程调用兄弟技能时用：先根据本技能 scripts 目录推断并列根；若目录下能识别出兄弟技能
+    则用之（开发仓 D:\\OpenClaw 与安装目录 ~/.openclaw/skills 均满足「技能根之上一级 = 并列根」），
+    避免全局 JIANGCHANG_SKILLS_ROOT / CLAW_SKILLS_ROOT 指向与当前检出不一致时错调兄弟进程。
+    推断失败时再读上述环境变量，最后回落 get_skills_root()。
+    """
+    if skill_scripts_dir:
+        scripts = os.path.abspath(skill_scripts_dir)
+        skill_root = os.path.dirname(scripts)
+        inferred = os.path.dirname(skill_root)
+        if _looks_like_skills_root(inferred):
+            return os.path.normpath(inferred)
+
+    for key in ("JIANGCHANG_SKILLS_ROOT", "CLAW_SKILLS_ROOT"):
+        v = (os.getenv(key) or "").strip()
+        if v:
+            return os.path.normpath(v)
+
+    return get_skills_root()
+
+
 def apply_cli_local_defaults() -> None:
     """
     在 CLI 最早阶段调用（main.py 在 import 业务包之前）。

scripts/service/account_service.py

@@ -1,4 +1,4 @@
-"""账号用例：list / get / add / delete / pick / set-login-status（含终端输出，供 CLI 调用）。"""
+"""账号用例：list / get / add / delete / pick-web（含终端输出，供 CLI 调用）。"""
 import json
 import os
 import shutil
@@ -15,13 +15,11 @@ from db.accounts_repo import (
     fetch_ids_for_list_json,
     fetch_ids_profile_for_platform_conn,
     fetch_list_rows,
-    fetch_pick_logged_in_id,
     fetch_pick_web_candidate_id,
     fetch_row_platform_phone_conn,
     get_account_by_id,
     has_duplicate_phone_conn,
     insert_account_row,
-    mark_login_status,
     normalize_account_id,
 )
 from db.connection import get_conn, init_db
@@ -78,8 +76,6 @@ def cmd_list(platform="all", limit: int = 10):
                 phone,
                 profile_dir,
                 url,
-                login_status,
-                last_login_at,
                 extra_json,
                 created_at,
                 updated_at,
@@ -90,8 +86,6 @@ def cmd_list(platform="all", limit: int = 10):
             print(f"phone：{phone or ''}")
             print(f"profile_dir：{profile_dir or ''}")
             print(f"url：{url or ''}")
-            print(f"login_status：{int(login_status) if login_status is not None else ''}")
-            print(f"last_login_at：{int(last_login_at) if last_login_at is not None else ''}")
             print(f"extra_json：{extra_json or ''}")
             print(f"created_at：{int(created_at) if created_at is not None else ''}")
             print(f"updated_at：{int(updated_at) if updated_at is not None else ''}")
@@ -141,40 +135,10 @@ def cmd_list_json(platform_input: str, limit: int = 200) -> None:
     print(json.dumps(out, ensure_ascii=False))
 
 
-def cmd_pick_logged_in(platform_input: str):
-    """
-    机器可读跨技能接口：查询指定平台下「已登录」的一条账号（login_status=1，按 last_login_at 优先）。
-    成功：stdout 仅输出一行 JSON，结构与 get 子命令一致。
-    失败：stdout 首行以 ERROR: 开头（由调用方判断，勿解析为 JSON）。
-    """
-    get_skill_logger().info("pick_logged_in platform_input=%r", platform_input)
-    key = resolve_platform_key((platform_input or "").strip())
-    if not key:
-        print("ERROR:INVALID_PLATFORM 无法识别的平台名称。")
-        print("支持：" + _platform_list_cn_for_help())
-        return
-
-    init_db()
-    picked = fetch_pick_logged_in_id(key)
-
-    if picked is None:
-        print("ERROR:NO_LOGGED_IN_ACCOUNT 该平台暂无已登录账号（login_status=1）。")
-        print("请先 list 查看账号 id，再执行：python main.py login <id>")
-        return
-
-    acc = get_account_by_id(picked)
-    if not acc:
-        print("ERROR:ACCOUNT_NOT_FOUND")
-        print(_runtime_paths_debug_text(), file=sys.stderr)
-        return
-    print(json.dumps(acc, ensure_ascii=False))
-
-
 def cmd_pick_web(platform_input: str):
     """
     供 llm-manager 等：取该平台用于网页自动化的账号候选。
-    优先 login_status=1（与 pick-logged-in 一致）；若无，则取该平台 updated_at 最新的一条，
-    便于「已 add 未标登录」时仍打开 profile，在浏览器内登录后继续任务。
+    按 updated_at、created_at 倒序选一条；是否已登录由调用方（如网页自动化）在同一会话内处理。
     成功：stdout 仅一行 JSON（与 get 一致）；失败：首行 ERROR:。
     """
     get_skill_logger().info("pick_web platform_input=%r", platform_input)
@@ -352,29 +316,3 @@ def cmd_delete_by_platform_phone(platform_input: str, phone: str) -> None:
         f"✅ 已删除账号：ID {rid} | {name} | {_PLATFORM_PRIMARY_CN.get(key, key)} | 手机 {phone_norm}"
     )
     log.info("delete_by_platform_phone_done id=%s platform=%s", rid, key)
-
-
-def cmd_set_login_status(account_id_str: str, status_str: str) -> None:
-    """
-    跨技能机器接口：回写 accounts.login_status。
-    成功：stdout 首行 OK:SET_LOGIN_STATUS 或 OK:CLEARED_LOGIN_STATUS，进程退出码 0。
-    失败：stdout 首行 ERROR:...，退出码 1。
-    """
-    aid_s = (account_id_str or "").strip()
-    st_s = (status_str or "").strip()
-    if not aid_s.isdigit():
-        print("ERROR:INVALID_ACCOUNT_ID")
-        sys.exit(1)
-    if st_s not in ("0", "1"):
-        print("ERROR:INVALID_STATUS 须为 0 或 1")
-        sys.exit(1)
-    aid = int(aid_s)
-    init_db()
-    if get_account_by_id(aid) is None:
-        get_skill_logger().warning("set_login_status_not_found account_id=%r", aid)
-        print("ERROR:ACCOUNT_NOT_FOUND")
-        sys.exit(1)
-    ok = st_s == "1"
-    mark_login_status(aid, ok)
-    get_skill_logger().info("set_login_status account_id=%s success=%s", aid, ok)
-    print("OK:SET_LOGIN_STATUS" if ok else "OK:CLEARED_LOGIN_STATUS")

scripts/service/browser_service.py

@@ -7,7 +7,7 @@ import tempfile
 import time
 from urllib.parse import urlparse
 
-from db.accounts_repo import get_account_by_id, mark_login_status
+from db.accounts_repo import get_account_by_id, touch_account_updated_at
 from util.logging_config import (
     get_skill_logger,
     get_skill_log_file_path,
@@ -21,6 +21,9 @@ from util.platforms import (
 
 # 子进程入口脚本与本模块同目录（独立解释器执行，非 import）
 _SERVICE_DIR = os.path.dirname(os.path.abspath(__file__))
+# PyArmor 在模块首行注入对 pyarmor_runtime_* 的 import；若以「脚本路径」启动子进程，
+# sys.path[0] 仅为 service/，找不到 scripts/ 下的运行时包。必须用 -m 且 cwd=scripts/（标准做法）。
+_SCRIPTS_ROOT = os.path.dirname(_SERVICE_DIR)
 
 
 def _win_find_exe(candidates):
@@ -96,14 +99,6 @@ def _login_dom_grace_seconds() -> float:
         return 4.0
 
 
-def _wait_login_timeout_seconds() -> int:
-    """供 wait-login / llm 编排：默认 180 秒，JIANGCHANG_WAIT_LOGIN_TIMEOUT_SECONDS。"""
-    try:
-        return max(30, min(600, int((os.getenv("JIANGCHANG_WAIT_LOGIN_TIMEOUT_SECONDS") or "180").strip())))
-    except ValueError:
-        return 180
-
-
 def _login_out_bundle_for(platform_key: str) -> dict:
     """传给子进程：anchor + 未登录可见的选择器列表。"""
     spec = PLATFORMS.get(platform_key, {}) if platform_key else {}
@@ -132,7 +127,7 @@ def _run_login_browser_session(
     """
     启动 Playwright 登录检测子进程；**是否已登录以子进程 DOM 判定为准**。
 
-    成功后尽力 `mark_login_status` 供 list/排序展示，**跨技能编排请勿再依赖该字段**（可能与真实 Cookie 会话脱节）。
+    若 DOM 判定成功则刷新该账号 `updated_at`，便于 `pick-web` 等按最近活跃选号；**是否已登录仅以本次检测为准**。
 
     返回 (是否已登录成功, end_reason)：end_reason 为 login_child 写入的 reason，失败时可能为
     timeout / user_closed / subprocess_error。
@@ -173,6 +168,13 @@ def _run_login_browser_session(
     if verbose_ui:
         print(f"正在为账号 [{target['name']}] 打开 {browser_name} …")
         print(f"访问地址：{url}")
+        if platform == "gemini":
+            print(
+                "Gemini / Google 登录：若提示「浏览器不安全」或「无法登录」，请关闭本窗口后，"
+                "用普通 Chrome 启动参数 --user-data-dir= 指向下面 profile 目录，在普通浏览器里完成"
+                " Google 登录后完全退出 Chrome，再执行本 login；详见 llm-manager Gemini 引擎说明。"
+            )
+            print(f"  （本账号 profile 目录）{profile_dir}")
         print(
             "请在窗口内完成登录。系统根据页面是否仍出现「登录」等未登录控件判断；"
             "成功后自动关闭窗口并写入状态（无需手动关浏览器）。"
@@ -196,7 +198,6 @@ def _run_login_browser_session(
         url,
         log_file,
     )
-    login_runner_path = os.path.join(_SERVICE_DIR, "login_child_runner.py")
     cfg = {
         "channel": channel,
         "profile_dir": profile_dir,
@@ -218,21 +219,49 @@ def _run_login_browser_session(
         json.dump(cfg, jf, ensure_ascii=False)
         cfg_path = jf.name
     proc_rc = None
+    child_hard_timeout = False
+    child_stderr = ""
+    child_stdout = ""
     try:
         r = subprocess.run(
-            [sys.executable, login_runner_path, cfg_path],
+            [sys.executable, "-m", "service.login_child_runner", cfg_path],
+            cwd=_SCRIPTS_ROOT,
             timeout=timeout_sec + 180,
             env=subprocess_env_with_trace(),
+            capture_output=True,
+            text=True,
+            encoding="utf-8",
+            errors="replace",
         )
         proc_rc = r.returncode
+        child_stderr = (r.stderr or "").strip()
+        child_stdout = (r.stdout or "").strip()
+        if child_stderr:
+            log.warning(
+                "login_child_stderr account_id=%s len=%s tail=%s",
+                account_id,
+                len(child_stderr),
+                child_stderr[-6000:],
+            )
+        if child_stdout and proc_rc != 0:
+            log.warning(
+                "login_child_stdout account_id=%s len=%s tail=%s",
+                account_id,
+                len(child_stdout),
+                child_stdout[-2000:],
+            )
         if proc_rc != 0:
             if verbose_ui:
-                print("⚠️ 浏览器进程异常退出，将仅根据已写入的检测结果更新状态")
-            log.warning("login_subprocess_nonzero_return code=%s", proc_rc)
+                print("⚠️ 浏览器子进程非零退出，以结果 JSON 为准（若已写入）")
+            log.info(
+                "login_subprocess_nonzero_return code=%s (exit code 不单独作为成败依据，见 result JSON)",
+                proc_rc,
+            )
     except subprocess.TimeoutExpired as ex:
+        child_hard_timeout = True
         if verbose_ui:
-            print("⚠️ 等待浏览器超时，将仅根据已写入的检测结果更新状态")
-        log.warning("login_subprocess_timeout err=%s", ex)
+            print("⚠️ 浏览器子进程等待超时，按「限时内未完成登录」处理")
+        log.warning("login_subprocess_hard_timeout err=%s", ex)
     finally:
         try:
             os.unlink(cfg_path)
@@ -240,16 +269,37 @@ def _run_login_browser_session(
             pass
 
     interactive_ok = False
-    end_reason = "subprocess_error"
+    end_reason = "timeout" if child_hard_timeout else "subprocess_error"
+    child_detail = ""
     try:
         with open(result_path, encoding="utf-8") as rf:
             data = json.load(rf)
             interactive_ok = bool(data.get("interactive_ok"))
-            end_reason = str(data.get("end_reason") or "").strip() or (
-                "login_ok" if interactive_ok else "timeout"
-            )
-    except Exception:
-        end_reason = "subprocess_error"
+            er = str(data.get("end_reason") or "").strip()
+            if er:
+                end_reason = er
+            elif interactive_ok:
+                end_reason = "login_ok"
+            elif not child_hard_timeout:
+                end_reason = "timeout"
+            child_detail = str(data.get("detail") or "").strip()
+        if child_detail:
+            log.warning("login_child_result_detail account_id=%s detail=%s", account_id, child_detail[:8000])
+    except Exception as read_ex:
+        log.error(
+            "login_child_result_read_failed account_id=%s err=%s proc_rc=%s stderr_tail=%s",
+            account_id,
+            read_ex,
+            proc_rc,
+            (child_stderr or "")[-3000:],
+        )
+        if child_hard_timeout:
+            end_reason = "timeout"
+            interactive_ok = False
+        elif proc_rc is not None and proc_rc != 0:
+            end_reason = "child_crash"
+        else:
+            end_reason = "subprocess_error"
     try:
         os.unlink(result_path)
     except OSError:
@@ -265,7 +315,8 @@ def _run_login_browser_session(
         proc_rc,
         ok,
     )
-    mark_login_status(account_id, ok)
+    if ok:
+        touch_account_updated_at(account_id)
     return ok, end_reason if not ok else "login_ok"
 
 
@@ -282,7 +333,7 @@ def cmd_login(account_id):
     ok, _end_reason = _run_login_browser_session(target, timeout_sec, verbose_ui=True)
     log_file = get_skill_log_file_path()
     if ok:
-        print("✅ 已判定登录成功，状态已写入数据库")
+        print("✅ 已判定页面登录成功")
     else:
         log.warning(
             "login_not_detected account_id=%s platform=%s see_log=%s",
@@ -290,70 +341,10 @@ def cmd_login(account_id):
             (target.get("platform") or "").strip().lower(),
             log_file,
         )
-        print("⚠️ 未检测到有效登录，状态为未登录。请关闭其他占用该用户目录的浏览器后重试，或延长 JIANGCHANG_LOGIN_TIMEOUT_SECONDS 后再登录。")
+        print("⚠️ 未检测到有效登录。请关闭其他占用该用户目录的浏览器后重试，或延长 JIANGCHANG_LOGIN_TIMEOUT_SECONDS 后再登录。")
         print(f"ℹ️ 详细轮询日志见：{log_file}（可将 JIANGCHANG_LOG_LEVEL=DEBUG 打开更细粒度）")
 
 
-def cmd_wait_login(account_id, force: bool = False) -> int:
-    """
-    跨技能编排：若未登录则打开浏览器并在限时内等待登录。
-    返回 0：已登录（原本已登录或本次登录成功）；1：失败（超时 / 用户关浏览器 / 其它）。
-    stdout 含 OK:WAIT_LOGIN 或 ERROR:WAIT_LOGIN_*。
-
-    force=True：即使库内 login_status=1 也仍打开浏览器并限时检测（供搜狐发布等「页面仍要求登录」场景）。
-    """
-    log = get_skill_logger()
-    log.info("wait_login_command account_id=%r force=%s", account_id, force)
-    target = get_account_by_id(account_id)
-    if not target:
-        log.warning("wait_login_aborted account_not_found account_id=%r", account_id)
-        print("ERROR:ACCOUNT_NOT_FOUND")
-        return 1
-    # 库内 login_status=1 时跳过开窗（与真实 Cookie 可能不同步时用 force=True）
-    if int(target.get("login_status") or 0) == 1 and not force:
-        print("OK:WAIT_LOGIN")
-        return 0
-
-    if not resolve_chromium_channel():
-        _print_browser_install_hint()
-        return 1
-    try:
-        import playwright  # noqa: F401
-    except ImportError:
-        print("ERROR:需要 playwright：pip install playwright && playwright install chromium")
-        return 1
-    if not (target.get("profile_dir") or "").strip():
-        print("ERROR:PROFILE_DIR_MISSING 库中 profile_dir 为空。")
-        return 1
-
-    sec = _wait_login_timeout_seconds()
-    print(
-        f"⏳ 账号「{target.get('name', target['id'])}」尚未登录，将打开浏览器；请在 {sec} 秒内完成登录。"
-        "关闭浏览器将视为放弃。"
-    )
-    ok, end_reason = _run_login_browser_session(target, sec, verbose_ui=False)
-    if ok:
-        print("OK:WAIT_LOGIN")
-        return 0
-    if end_reason == "user_closed":
-        print("ERROR:WAIT_LOGIN_ABORTED 已关闭浏览器，未完成登录。")
-        return 1
-    if end_reason == "timeout":
-        print("ERROR:WAIT_LOGIN_TIMEOUT 限时内未完成登录。")
-        return 1
-    if end_reason == "playwright_missing":
-        print("ERROR:需要 playwright：pip install playwright && playwright install chromium")
-        return 1
-    if end_reason == "profile_missing":
-        print("ERROR:PROFILE_DIR_MISSING 库中 profile_dir 为空。")
-        return 1
-    if end_reason == "no_channel":
-        _print_browser_install_hint()
-        return 1
-    print(f"ERROR:WAIT_LOGIN_FAILED end_reason={end_reason}")
-    return 1
-
-
 def cmd_open(account_id):
     """打开该账号的持久化浏览器，仅用于肉眼确认是否已登录；不写数据库。"""
     get_skill_logger().info("open account_id=%r", account_id)
@@ -386,9 +377,8 @@ def cmd_open(account_id):
     print(f"正在打开 [{target['name']}] 的 {browser_name}（仅查看，不写入数据库）")
     print(f"地址：{url}")
     print("请在本窗口中自行确认登录态。关闭浏览器后命令结束。")
-    print("需要自动检测并写入数据库时，请执行：python main.py login <id>")
+    print("需要自动检测登录时，请执行：python main.py login <id>")
 
-    open_runner_path = os.path.join(_SERVICE_DIR, "open_child_runner.py")
     cfg = {"channel": channel, "profile_dir": profile_dir, "url": url}
     with tempfile.NamedTemporaryFile(
         mode="w", suffix=".json", delete=False, encoding="utf-8"
@@ -397,7 +387,8 @@ def cmd_open(account_id):
         cfg_path = jf.name
     try:
         subprocess.run(
-            [sys.executable, open_runner_path, cfg_path],
+            [sys.executable, "-m", "service.open_child_runner", cfg_path],
+            cwd=_SCRIPTS_ROOT,
             env=subprocess_env_with_trace(),
         )
     finally:

scripts/service/login_child_runner.py

@@ -1,36 +1,57 @@
-"""Playwright 登录检测子进程：由 browser_service.cmd_login 以独立解释器启动，argv[1] 为 JSON 配置路径。"""
+"""Playwright 登录检测子进程：由 browser_service 以独立解释器启动，argv[1] 为 JSON 配置路径。
+
+注意：不得在模块顶层 import playwright / jiangchang_skill_core —— 若导入失败，进程会在 main() 之前以退出码 1 退出，
+且写不了 result_path，父进程只能得到 subprocess_error。所有重依赖均在 _run_login_child 内延迟加载。
+"""
+from __future__ import annotations
+
 import json
 import os
 import sys
 import time
+import traceback
 from urllib.parse import urlparse
 
 _scripts_dir = os.path.dirname(os.path.dirname(os.path.abspath(__file__)))
 if _scripts_dir not in sys.path:
     sys.path.insert(0, _scripts_dir)
 
-from jiangchang_skill_core.unified_logging import attach_unified_file_handler
-from playwright.sync_api import sync_playwright
-
-try:
-    from playwright.sync_api import Error as PlaywrightError
-except ImportError:
-    PlaywrightError = Exception  # type: ignore[misc,assignment]
+from util.playwright_stealth import (
+    STEALTH_INIT_SCRIPT,
+    persistent_context_launch_parts,
+    stealth_enabled,
+)
 
 
 def _is_context_closed_error(ex: BaseException) -> bool:
     s = str(ex).lower()
     if "target closed" in s or "browser has been closed" in s or "context was destroyed" in s:
         return True
-    if isinstance(ex, PlaywrightError) and (
-        "closed" in s or "destroyed" in s
-    ):
+    if "closed" in s or "destroyed" in s:
         return True
     return False
 
 
+def _write_result_file(
+    result_path: str,
+    interactive_ok: bool,
+    end_reason: str,
+    detail: str = "",
+) -> None:
+    if not (result_path or "").strip():
+        return
+    payload: dict = {"interactive_ok": interactive_ok, "end_reason": end_reason}
+    d = (detail or "").strip()
+    if d:
+        payload["detail"] = d[:8000]
+    try:
+        with open(result_path, "w", encoding="utf-8") as rf:
+            json.dump(payload, rf, ensure_ascii=False)
+    except Exception as write_ex:
+        print(f"login_child: failed to write result file: {write_ex}", file=sys.stderr)
+
+
 def page_location_href(p):
-    # Prefer location.href over page.url for SPA (e.g. Sohu shell path lag).
     try:
         href = p.evaluate("() => location.href")
         if href and str(href).strip():
@@ -124,142 +145,246 @@ def evaluate_logged_out_dom(ctx, main_page, bundle):
     return False, ""
 
 
-def main():
-    with open(sys.argv[1], encoding="utf-8") as f:
-        c = json.load(f)
-    bundle = c.get("login_detect_bundle") or {}
-    poll = float(c.get("poll_interval", 1.5))
+def _run_login_child(c: dict) -> None:
+    """单次登录检测；结果一律写入 c['result_path']（由 finally 保证）。"""
+    result_path = (c.get("result_path") or "").strip()
     try:
-        dom_grace = float(c.get("dom_grace_sec", 4.0))
-    except (TypeError, ValueError):
-        dom_grace = 4.0
-    if dom_grace < 0:
-        dom_grace = 0.0
-    t0 = time.time()
-    deadline = t0 + float(c["timeout_sec"])
+        from playwright.sync_api import sync_playwright
+    except ImportError:
+        _write_result_file(
+            result_path,
+            False,
+            "playwright_missing",
+            traceback.format_exc(),
+        )
+        return
+
     interactive_ok = False
     end_reason = "timeout"
-    result_path = c.get("result_path") or ""
-    log_path = (c.get("log_file") or "").strip()
+    ctx = None
     lg = None
-    if log_path:
-        lg = attach_unified_file_handler(
-            log_path,
-            skill_slug="account-manager",
-            logger_name="openclaw.skill.account_manager.login_child",
-        )
     last_eval_detail = ""
-    with sync_playwright() as p:
-        ctx = p.chromium.launch_persistent_context(
-            user_data_dir=c["profile_dir"],
-            headless=False,
-            channel=c["channel"],
-            no_viewport=True,
-            args=["--start-maximized"],
-        )
+    detail_msg = ""
+
+    try:
         try:
-            page = ctx.pages[0] if ctx.pages else ctx.new_page()
-            page.goto(c["url"], wait_until="domcontentloaded", timeout=60000)
-            if dom_grace > 0:
-                time.sleep(dom_grace)
-            if lg is not None:
-                lg.info(
-                    "goto_done initial_pages=%s poll_interval_sec=%s dom_grace_sec=%s",
-                    len(ctx.pages or []),
-                    poll,
-                    dom_grace,
+            from jiangchang_skill_core.unified_logging import attach_unified_file_handler
+        except Exception as e:
+            end_reason = "logging_import_failed"
+            detail_msg = traceback.format_exc()
+            print(detail_msg, file=sys.stderr)
+            return
+
+        try:
+            profile_dir = (c.get("profile_dir") or "").strip()
+            if not profile_dir:
+                end_reason = "profile_missing"
+                detail_msg = "profile_dir empty in config"
+                return
+
+            os.makedirs(profile_dir, exist_ok=True)
+            try:
+                n_entries = len(os.listdir(profile_dir))
+            except OSError as e:
+                end_reason = "profile_dir_unusable"
+                detail_msg = f"{type(e).__name__}: {e}"
+                return
+        except Exception as e:
+            end_reason = "profile_dir_unusable"
+            detail_msg = traceback.format_exc()
+            print(detail_msg, file=sys.stderr)
+            return
+
+        log_path = (c.get("log_file") or "").strip()
+        if log_path:
+            try:
+                lg = attach_unified_file_handler(
+                    log_path,
+                    skill_slug="account-manager",
+                    logger_name="openclaw.skill.account_manager.login_child",
                 )
-            while time.time() < deadline:
-                iter_start = time.time()
-                try:
-                    still_logged_out, detail = evaluate_logged_out_dom(ctx, page, bundle)
-                    ok = not still_logged_out
-                    last_eval_detail = detail
+            except Exception as e:
+                lg = None
+                print(f"login_child: attach_unified_file_handler failed: {e}", file=sys.stderr)
+
+        if lg is not None:
+            lg.info(
+                "login_child_start profile_dir=%r profile_entry_count=%s channel=%r url=%r timeout_sec=%r",
+                profile_dir,
+                n_entries,
+                c.get("channel"),
+                c.get("url"),
+                c.get("timeout_sec"),
+            )
+
+        bundle = c.get("login_detect_bundle") or {}
+        poll = float(c.get("poll_interval", 1.5))
+        try:
+            dom_grace = float(c.get("dom_grace_sec", 4.0))
+        except (TypeError, ValueError):
+            dom_grace = 4.0
+        if dom_grace < 0:
+            dom_grace = 0.0
+        t0 = time.time()
+        deadline = t0 + float(c["timeout_sec"])
+
+        with sync_playwright() as p:
+            try:
+                args, ignore_automation = persistent_context_launch_parts()
+                lc_kw = dict(
+                    user_data_dir=profile_dir,
+                    headless=False,
+                    channel=c["channel"],
+                    no_viewport=True,
+                    args=args,
+                )
+                if ignore_automation is not None:
+                    lc_kw["ignore_default_args"] = ignore_automation
+                ctx = p.chromium.launch_persistent_context(**lc_kw)
+                if stealth_enabled():
+                    ctx.add_init_script(STEALTH_INIT_SCRIPT)
+            except Exception as e:
+                if _is_context_closed_error(e):
+                    end_reason = "user_closed"
+                else:
+                    end_reason = "launch_failed"
+                    detail_msg = f"{type(e).__name__}: {e}"
                     if lg is not None:
-                        parts = []
-                        for tab in list(ctx.pages or []):
-                            try:
-                                href = page_location_href(tab)
-                                pu = (tab.url or "").strip()
-                                parts.append("href=%r playwright_url=%r" % (href, pu))
-                            except Exception:
-                                parts.append("(tab_read_error)")
-                        lg.debug(
-                            "poll iter_start_elapsed=%.1fs deadline_in=%.1fs tabs=%s poll_interval_sec=%s logged_out=%s ok=%s detail=%s | %s",
-                            iter_start - t0,
-                            deadline - iter_start,
-                            len(ctx.pages or []),
-                            poll,
-                            still_logged_out,
-                            ok,
-                            detail,
-                            " ; ".join(parts) if parts else "no_tabs",
-                        )
-                    if ok:
-                        interactive_ok = True
-                        end_reason = "login_ok"
-                        if lg is not None:
-                            lg.info("login_detected_ok %s", detail)
-                        break
-                except Exception as ex:
-                    if _is_context_closed_error(ex):
-                        end_reason = "user_closed"
-                        if lg is not None:
-                            lg.warning("login_poll_browser_closed err=%s", ex)
-                        break
-                    if lg is not None:
-                        lg.warning("poll_exception err=%s", ex, exc_info=True)
-                spent = time.time() - iter_start
-                time.sleep(max(0.0, poll - spent))
-            if not interactive_ok and end_reason != "user_closed":
+                        lg.exception("launch_persistent_context failed: %s", e)
+                    print(detail_msg, file=sys.stderr)
+                return
+
+            try:
+                page = ctx.pages[0] if ctx.pages else ctx.new_page()
+                page.goto(c["url"], wait_until="domcontentloaded", timeout=60000)
+                if dom_grace > 0:
+                    time.sleep(dom_grace)
                 if lg is not None:
-                    lg.warning("login_poll_exhausted detail=%s", last_eval_detail)
-                rem = max(0.0, deadline - time.time())
-                if rem > 0:
+                    lg.info(
+                        "goto_done initial_pages=%s poll_interval_sec=%s dom_grace_sec=%s",
+                        len(ctx.pages or []),
+                        poll,
+                        dom_grace,
+                    )
+                while time.time() < deadline:
+                    iter_start = time.time()
                     try:
-                        ctx.wait_for_event("close", timeout=rem * 1000)
-                    except Exception:
-                        pass
-            else:
-                if lg is not None:
-                    lg.info("login_dom_ok_preparing_to_release_profile_dir")
-                if interactive_ok:
-                    # 给 Cookie/会话落盘一点时间；随后必须关闭上下文，否则 llm-manager 无法再对同一 profile 启动持久化浏览器
+                        still_logged_out, dom_detail = evaluate_logged_out_dom(ctx, page, bundle)
+                        ok = not still_logged_out
+                        last_eval_detail = dom_detail
+                        if lg is not None:
+                            parts = []
+                            for tab in list(ctx.pages or []):
+                                try:
+                                    href = page_location_href(tab)
+                                    pu = (tab.url or "").strip()
+                                    parts.append("href=%r playwright_url=%r" % (href, pu))
+                                except Exception:
+                                    parts.append("(tab_read_error)")
+                            lg.debug(
+                                "poll elapsed=%.1fs left=%.1fs tabs=%s logged_out=%s ok=%s detail=%s | %s",
+                                iter_start - t0,
+                                deadline - iter_start,
+                                len(ctx.pages or []),
+                                still_logged_out,
+                                ok,
+                                dom_detail,
+                                " ; ".join(parts) if parts else "no_tabs",
+                            )
+                        if ok:
+                            interactive_ok = True
+                            end_reason = "login_ok"
+                            if lg is not None:
+                                lg.info("login_detected_ok %s", dom_detail)
+                            break
+                    except Exception as ex:
+                        if _is_context_closed_error(ex):
+                            end_reason = "user_closed"
+                            if lg is not None:
+                                lg.warning("login_poll_browser_closed err=%s", ex)
+                            break
+                        if lg is not None:
+                            lg.warning("poll_exception err=%s", ex, exc_info=True)
+                    spent = time.time() - iter_start
+                    time.sleep(max(0.0, poll - spent))
+
+                if not interactive_ok and end_reason != "user_closed":
+                    if lg is not None:
+                        lg.warning("login_poll_exhausted detail=%s", last_eval_detail)
+                    rem = max(0.0, deadline - time.time())
+                    if rem > 0:
+                        try:
+                            ctx.wait_for_event("close", timeout=rem * 1000)
+                        except Exception:
+                            pass
+                elif interactive_ok:
+                    if lg is not None:
+                        lg.info("login_dom_ok_preparing_to_release_profile_dir")
                     print(
                         "INFO:LOGIN_DOM_OK 页面检测已判定登录成功；约 2 秒后关闭本窗口以释放用户目录，供后续大模型网页会话使用。",
                         flush=True,
                     )
                     time.sleep(2.0)
-        except Exception as e:
-            if _is_context_closed_error(e):
-                end_reason = "user_closed"
+            except Exception as e:
                 interactive_ok = False
-                if lg is not None:
-                    lg.warning("login_runner_browser_closed err=%s", e)
-            else:
-                if lg is not None:
-                    lg.exception("login_runner_fatal err=%s", e)
-                print(e, file=sys.stderr)
-                raise
-        finally:
-            if result_path:
+                if _is_context_closed_error(e):
+                    end_reason = "user_closed"
+                    if lg is not None:
+                        lg.warning("login_runner_browser_closed err=%s", e)
+                else:
+                    end_reason = "page_error"
+                    detail_msg = traceback.format_exc()
+                    if lg is not None:
+                        lg.exception("login_runner_page_err %s", e)
+                    print(detail_msg, file=sys.stderr)
+            finally:
                 try:
-                    with open(result_path, "w", encoding="utf-8") as rf:
-                        json.dump(
-                            {
-                                "interactive_ok": interactive_ok,
-                                "end_reason": end_reason,
-                            },
-                            rf,
-                            ensure_ascii=False,
-                        )
+                    ctx.close()
                 except Exception:
                     pass
+                ctx = None
+
+    except Exception as e:
+        interactive_ok = False
+        end_reason = "child_fatal"
+        detail_msg = traceback.format_exc()
+        if lg is not None:
+            lg.exception("login_child_unhandled %s", e)
+        print(detail_msg, file=sys.stderr)
+    finally:
+        if ctx is not None:
             try:
                 ctx.close()
             except Exception:
                 pass
+        _write_result_file(result_path, interactive_ok, end_reason, detail_msg)
+
+
+def _entrypoint() -> None:
+    """读取配置 → 运行检测；任意异常都写入 result JSON，进程以 0 退出（由父进程读 JSON 判定成败）。"""
+    result_path = ""
+    if len(sys.argv) < 2:
+        print("login_child: missing cfg path argv", file=sys.stderr)
+        return
+
+    cfg_path = sys.argv[1]
+    try:
+        with open(cfg_path, encoding="utf-8") as f:
+            c = json.load(f)
+    except Exception:
+        tb = traceback.format_exc()
+        print(tb, file=sys.stderr)
+        return
+
+    result_path = (c.get("result_path") or "").strip()
+
+    try:
+        _run_login_child(c)
+    except Exception:
+        tb = traceback.format_exc()
+        print(tb, file=sys.stderr)
+        _write_result_file(result_path, False, "child_fatal", tb)
 
 
 if __name__ == "__main__":
-    main()
+    _entrypoint()

scripts/service/open_child_runner.py

@@ -4,18 +4,30 @@ import sys
 
 from playwright.sync_api import sync_playwright
 
+from util.playwright_stealth import (
+    STEALTH_INIT_SCRIPT,
+    persistent_context_launch_parts,
+    stealth_enabled,
+)
+
 
 def main():
     with open(sys.argv[1], encoding="utf-8") as f:
         c = json.load(f)
     with sync_playwright() as p:
-        ctx = p.chromium.launch_persistent_context(
+        args, ignore_automation = persistent_context_launch_parts()
+        lc_kw = dict(
             user_data_dir=c["profile_dir"],
             headless=False,
             channel=c["channel"],
             no_viewport=True,
-            args=["--start-maximized"],
+            args=args,
         )
+        if ignore_automation is not None:
+            lc_kw["ignore_default_args"] = ignore_automation
+        ctx = p.chromium.launch_persistent_context(**lc_kw)
+        if stealth_enabled():
+            ctx.add_init_script(STEALTH_INIT_SCRIPT)
         try:
             page = ctx.pages[0] if ctx.pages else ctx.new_page()
             page.goto(c["url"], wait_until="domcontentloaded", timeout=60000)

scripts/util/platforms.py

@@ -62,6 +62,12 @@ PLATFORMS = {
         "prefix": "元宝",
         "aliases": ["元宝"],
     },
+    "gemini": {
+        "url": "https://gemini.google.com",
+        "label": "Gemini",
+        "prefix": "Gemini",
+        "aliases": ["谷歌Gemini", "Google Gemini", "google gemini", "Bard"],
+    },
 }
 
 # 未登录页共性：主行动点含「登录」——submit 按钮、Element 主按钮、Semi 文案、标题、通栏主按钮等，Playwright :has-text 可覆盖多数站点。

scripts/util/playwright_stealth.py

@@ -0,0 +1,40 @@
+"""
+淡化 Playwright 启动时的自动化指纹，减轻 Google 账号页「此浏览器或应用可能不安全」类拦截。
+
+关闭：环境变量 OPENCLAW_PLAYWRIGHT_STEALTH=0（或 false/off/no）。
+"""
+
+from __future__ import annotations
+
+import os
+
+STEALTH_INIT_SCRIPT = """
+(() => {
+  try {
+    Object.defineProperty(navigator, "webdriver", { get: () => undefined });
+  } catch (e) {}
+})();
+"""
+
+
+def stealth_enabled() -> bool:
+    v = (os.environ.get("OPENCLAW_PLAYWRIGHT_STEALTH") or "1").strip().lower()
+    return v not in ("0", "false", "no", "off")
+
+
+def persistent_context_launch_parts(
+    *,
+    extra_args: list[str] | None = None,
+) -> tuple[list[str], list[str] | None]:
+    """
+    返回 (args, ignore_default_args)。
+    ignore_default_args 为 None 时不应传给 launch_persistent_context。
+    """
+    args = ["--start-maximized"]
+    if extra_args:
+        args = args + list(extra_args)
+    if not stealth_enabled():
+        return args, None
+    if "--disable-blink-features=AutomationControlled" not in args:
+        args.append("--disable-blink-features=AutomationControlled")
+    return args, ["--enable-automation"]