refactor(sdk): drop data-jcid, switch to semantic anchors + ClawX testids

- Introduce centralized SELECTOR constant table at top of client.py
- Replace all [data-jcid=...] selectors with:
  * Semantic anchors (data-role / data-message-id / data-streaming /
    data-sending / data-message-count) guarded by jiangchang main repo
  * ClawX upstream data-testids (chat-page / chat-composer-* / etc.)
- _latest_assistant_after: has_body now judged by inner_text non-empty
  (no longer depends on message-body sub-element)
- _gateway_state: degrade to window.__jc_gateway_state__ readback,
  returns 'unknown' until host exposes it
- new_task: rely solely on data-message-count==0 (welcome-screen no
  longer has a testid)
- wait_gateway_ready: degrade to chat-page selector wait
- send_file: raise NotImplementedError (Electron IPC hook missing)

Public API signatures unchanged. window.__jc_sending__ remains the
core sending signal.

Compatibility: jiangchang v2.0.17+ only.
Refs: jiangchang/docs/sdk-migration-discovery.md
      jiangchang/docs/JIANGCHANG_CUSTOM_ANCHORS.md
Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
2026-05-18 16:05:04 +08:00
parent 089f00c3ff
commit 4fc4cfb802
2 changed files with 190 additions and 122 deletions

View File

@@ -63,3 +63,57 @@ git push origin v0.1.0
```
仓库内还保留 `examples/``tools/``python-runtime/` 目录及既有 Skill / 前端相关的 GitHub Actions 工作流,与本 PyPI 包的发布彼此独立。
## v0.2.0 — data-jcid migration (internal refactor)
### TL;DR
本次内部重构:SDK 不再依赖匠厂主仓库的 `data-jcid` 属性。
所有公开 API(`ask` / `read` / `new_task` / `wait_gateway_ready` 等)
**签名与行为保持不变**,只有内部选择器实现切换到新的语义锚点 + ClawX testid。
### 改变了什么
- `client.py` 内所有 `[data-jcid="..."]` 选择器已移除
- 改用匠厂主仓库为 SDK 承诺的语义锚点:`data-role` / `data-message-id` /
`data-streaming`(在 ChatMessage 上)、`data-sending` / `data-message-count`
(在 chat-page 根上)
- `window.__jc_sending__` 仍是核心 sending 信号,保持不变
- 这些锚点由匠厂主仓库的以下机制守护:
- `docs/JIANGCHANG_CUSTOM_ANCHORS.md` 中央清单
- 源码内 `JIANGCHANG CUSTOM ANCHOR` 边界注释
- `harness/specs/jiangchang-custom-anchors.spec.ts` E2E 守护测试
- Gitea CI grep 检查 step
### 已知限制
- `send_file()` 暂未实现,抛 `NotImplementedError`
原因:附件上传走 Electron IPC,SDK 需要主仓库提供新 hook。
- `_gateway_state()` 永远返回 `'unknown'`
原因:gateway 状态不再暴露 DOM data-state;本方法已预留未来通过
`window.__jc_gateway_state__` 读取的实现路径。
这两个限制对常规 ask/read 工作流**无影响**,sending 完成判定依靠
`window.__jc_sending__` + `data-sending` + 文本稳定性合流即可。
### 升级影响
- 在 jiangchang v2.0.17(含)以上版本上跑:应正常工作
- 在 jiangchang v2.0.16(及更早含 data-jcid 的版本)上跑:不再兼容
如果需要在旧版本上跑,固定 SDK 版本到 v0.1.x
### 验证清单(集成方建议跑一次)
打开匠厂客户端 → 用 SDK 跑下面 3 步,确认基础流程通:
```python
from jiangchang_desktop_sdk import JiangchangDesktopClient
with JiangchangDesktopClient() as c:
c.ensure_app_running()
c.wait_gateway_ready()
c.new_task()
answer = c.ask("你好,请回复确认 SDK 可用")
assert answer, f"Empty answer: {answer!r}"
print("OK:", answer[:200])
```

View File

@@ -54,6 +54,42 @@ from .exceptions import (
)
from .types import AskOptions, AssertOptions, JiangchangMessage, LaunchOptions
# ============================================================================
# JIANGCHANG DOM SELECTORS
# ============================================================================
# 本节集中维护所有 Playwright 选择器,作为对匠厂前端的契约。
# 选择器分两类:
#
# (A) 主仓库语义锚点(由 docs/JIANGCHANG_CUSTOM_ANCHORS.md 守护,稳定)
# 这些是匠厂为 SDK 暴露的承诺契约,被 E2E 守护测试 + Gitea CI 双重保护。
#
# (B) 主仓库 ClawX 上游 data-testid(随上游同步可能漂移)
# 这些是 ClawX 上游或我们扩展的测试钩子,本身用于内部 E2E,
# SDK 借用但不保证不变。
#
# 修改前请确认 docs/sdk-migration-discovery.md 中对应元素仍存在。
# ============================================================================
# --- (A) 匠厂语义锚点 ---
SEL_CHAT_PAGE = '[data-testid="chat-page"]'
SEL_CHAT_PAGE_SENDING = '[data-testid="chat-page"][data-sending="true"]'
SEL_MESSAGE_ROW = '[data-role]' # ChatMessage 行容器
SEL_MESSAGE_ROW_USER = '[data-role="user"]'
SEL_MESSAGE_ROW_ASSISTANT = '[data-role="assistant"]'
SEL_MESSAGE_ROW_STREAMING = '[data-streaming="true"]'
# --- (B) ClawX/上游 data-testid ---
SEL_SIDEBAR_NEW_CHAT = '[data-testid="sidebar-new-chat"]'
SEL_CHAT_COMPOSER_INPUT = '[data-testid="chat-composer-input"]'
SEL_CHAT_COMPOSER_SEND = '[data-testid="chat-composer-send"]'
SEL_CHAT_MESSAGE_INDEXED = '[data-testid^="chat-message-"]' # chat-message-${idx}
SEL_EXECUTION_GRAPH = '[data-testid="chat-execution-graph"]'
SEL_EXECUTION_GRAPH_EXPANDED = '[data-testid="chat-execution-graph"][data-collapsed="false"]'
SEL_MAIN_LAYOUT = '[data-testid="main-layout"]'
# --- (C) 通用兜底(无 testid 时使用) ---
SEL_FALLBACK_TEXTAREA = 'textarea'
# ── 调试日志 ───────────────────────────────────────────────────────
_logger = logging.getLogger("jiangchang_desktop_sdk")
_handler_added = False
@@ -375,51 +411,46 @@ class JiangchangDesktopClient:
def _get_chat_input_locator(self):
page = self.get_page()
locator = page.locator('[data-jcid="chat-input"]')
locator = page.locator(SEL_CHAT_COMPOSER_INPUT)
if locator.count() > 0:
return locator
locator = page.locator("textarea").last
locator = page.locator(SEL_FALLBACK_TEXTAREA).last
if locator.count() > 0:
return locator
raise ConnectionError("找不到聊天输入框textarea请确认匠厂已加 data-jcid=\"chat-input\"")
raise ConnectionError(
'找不到聊天输入框,请确认匠厂已加 data-testid="chat-composer-input" '
"或页面中存在 textarea"
)
def _get_send_button_locator(self):
page = self.get_page()
locator = page.locator('[data-jcid="send-button"]')
locator = page.locator(SEL_CHAT_COMPOSER_SEND)
if locator.count() > 0:
return locator
locator = page.locator('button[type="submit"]')
if locator.count() > 0:
return locator
raise ConnectionError("找不到发送按钮,请确认匠厂已加 data-jcid=\"send-button\"")
raise ConnectionError(
'找不到发送按钮,请确认匠厂已加 data-testid="chat-composer-send" '
'或 button[type="submit"]'
)
def _get_new_task_button_locator(self):
page = self.get_page()
locator = page.locator('[data-jcid="new-task-button"]')
locator = page.locator(SEL_SIDEBAR_NEW_CHAT)
if locator.count() > 0:
return locator
# 回退:旧版本 data-testid
locator = page.locator('[data-testid="sidebar-new-chat"]')
if locator.count() > 0:
return locator
# 最后兜底:文本"新建任务"
locator = page.get_by_text("新建任务", exact=False).first
return locator
def _get_message_list_locator(self):
page = self.get_page()
locator = page.locator('[data-jcid="message-list"]')
if locator.count() > 0:
return locator
locator = page.locator('[class*="space-y-3"]').first
if locator.count() > 0:
return locator
raise ConnectionError("找不到消息列表容器 data-jcid=\"message-list\"")
return page.locator(SEL_CHAT_PAGE)
# ─── 新建任务 ─────────────────────────────────────────────────
def new_task(self) -> None:
"""点击"新建任务"按钮,等待消息列表清空或欢迎屏出现"""
"""点击"新建任务"按钮,等待消息列表清空chat-page data-message-count=0"""
if not self._connected:
raise ConnectionError("未连接到应用")
page = self.get_page()
@@ -431,16 +462,12 @@ class JiangchangDesktopClient:
_logger.warning("[new_task] 点击失败:%s", exc)
return
# 等待 chat-root.message-count=0 或 welcome-screen 出现,最多 5s
deadline = time.time() + 5
while time.time() < deadline:
try:
count = page.locator('[data-jcid="chat-root"]').first.get_attribute("data-jcid-message-count")
count = page.locator(SEL_CHAT_PAGE).first.get_attribute("data-message-count")
if count == "0":
_logger.debug("[new_task] 已进入新任务message-count=0")
return
if page.locator('[data-jcid="welcome-screen"]').count() > 0:
_logger.debug("[new_task] 欢迎屏已出现")
_logger.debug("[new_task] 已进入新任务(data-message-count=0")
return
except Exception:
pass
@@ -450,25 +477,35 @@ class JiangchangDesktopClient:
# ─── 网关健康检查 ─────────────────────────────────────────────
def _gateway_state(self) -> str:
"""Read gateway state via Electron IPC bridge (jiangchang exposes no DOM data-state).
Strategy: try to read window.__jc_gateway_state__ if exposed;
otherwise return 'unknown' and let upper logic continue.
"""
# TODO: 等匠厂在 src/stores/gateway/* 或 main 进程暴露
# window.__jc_gateway_state__ 之后,本方法将自动开始返回真实状态。
# 当前永远返回 'unknown',行为退化为"假定 gateway 健康"。
try:
page = self.get_page()
el = page.locator('[data-jcid="gateway-status"]').first
if el.count() == 0:
return "unknown"
return el.get_attribute("data-state") or "unknown"
state = self.get_page().evaluate(
"() => window.__jc_gateway_state__"
)
if isinstance(state, str) and state:
return state
except Exception:
return "unknown"
pass
return 'unknown'
def wait_gateway_ready(self, timeout_ms: int = 30000) -> None:
"""等待匠厂前端就绪。
由于 v2.0.17 起 gateway 状态不再暴露 DOM data-state,本方法降级为
等待 chat-page 根元素可见(隐含 React + gateway store 已初始化)。
Gateway 实际就绪与否,由后续 ask() 内的 _gateway_state 检查兜底
(当前永远返回 'unknown')。
"""
page = self.get_page()
try:
page.wait_for_function(
"""() => {
const el = document.querySelector('[data-jcid="gateway-status"]');
return el && el.dataset.state === 'running';
}""",
timeout=timeout_ms,
)
page.wait_for_selector(SEL_CHAT_PAGE, timeout=timeout_ms)
except Exception as exc:
_logger.warning("[wait_gateway_ready] 超时/异常:%s(继续,但后续可能报 GatewayDown", exc)
@@ -481,7 +518,7 @@ class JiangchangDesktopClient:
def _message_node_count(self) -> int:
try:
return self.get_page().locator('[data-jcid="message"]').count()
return self.get_page().locator(SEL_MESSAGE_ROW).count()
except Exception:
return 0
@@ -489,10 +526,10 @@ class JiangchangDesktopClient:
"""返回 `[data-jcid="message"]` 节点列表中最后一条 role=user 的 index找不到返回 -1。"""
try:
page = self.get_page()
total = page.locator('[data-jcid="message"]').count()
total = page.locator(SEL_MESSAGE_ROW).count()
for i in range(total - 1, -1, -1):
node = page.locator('[data-jcid="message"]').nth(i)
role = node.get_attribute("data-jcid-role") or ""
node = page.locator(SEL_MESSAGE_ROW).nth(i)
role = node.get_attribute("data-role") or ""
if role.lower() == "user":
return i
return -1
@@ -507,7 +544,7 @@ class JiangchangDesktopClient:
return False
try:
page = self.get_page()
node = page.locator('[data-jcid="message"]').nth(user_idx)
node = page.locator(SEL_MESSAGE_ROW).nth(user_idx)
text = node.inner_text(timeout=2000) or ""
t = re.sub(r"\s+", "", text)
q = re.sub(r"\s+", "", question)
@@ -529,22 +566,19 @@ class JiangchangDesktopClient:
return None
try:
page = self.get_page()
nodes = page.locator('[data-jcid="message"]')
nodes = page.locator(SEL_MESSAGE_ROW)
total = nodes.count()
for i in range(total - 1, user_idx, -1):
node = nodes.nth(i)
role = (node.get_attribute("data-jcid-role") or "").lower()
role = (node.get_attribute("data-role") or "").lower()
if role != "assistant":
continue
body = node.locator('[data-jcid="message-body"]')
has_body = body.count() > 0
body_text = ""
if has_body:
try:
body_text = body.first.inner_text(timeout=2000) or ""
body_text = node.inner_text(timeout=2000) or ""
except Exception:
body_text = ""
streaming = (node.get_attribute("data-jcid-streaming") or "false").lower() == "true"
has_body = len(body_text.strip()) > 0
streaming = (node.get_attribute("data-streaming") or "false").lower() == "true"
return {
"index": i,
"has_body": has_body,
@@ -575,8 +609,8 @@ class JiangchangDesktopClient:
def _chat_root_sending(self) -> Optional[bool]:
"""chat-root[data-jcid-sending] 的 bool 化;读不到返回 None。"""
try:
val = self.get_page().locator('[data-jcid="chat-root"]').first.get_attribute(
"data-jcid-sending"
val = self.get_page().locator(SEL_CHAT_PAGE).first.get_attribute(
"data-sending"
)
if val is None:
return None
@@ -586,8 +620,8 @@ class JiangchangDesktopClient:
def _chat_root_message_count(self) -> Optional[int]:
try:
val = self.get_page().locator('[data-jcid="chat-root"]').first.get_attribute(
"data-jcid-message-count"
val = self.get_page().locator(SEL_CHAT_PAGE).first.get_attribute(
"data-message-count"
)
if val is None:
return None
@@ -626,16 +660,16 @@ class JiangchangDesktopClient:
"""
等待 agent 整轮(可能包含多次 thinking / tool_use 往返)完成。
核心判据 —— 只有 agent 最终答复才有 `message-body`(文本气泡)
核心判据 —— agent 最终答复在 data-role=assistant 行上有非空 inner_text
源码依据D:\\AI\\jiangchang/src/stores/chat/helpers.ts
- isToolOnlyMessage 显式允许 thinking 与 tool_use 共存仍视为中间轮
- 中间工具轮的 ChatMessage 无 hasText → 不渲染 MessageBubble →
也就不会有 `data-jcid="message-body"`
- 只有最终含 text 的助手消息才会渲染 message-body
- 中间工具轮可能无可见正文气泡 → data-role 行 inner_text 为空
- 只有最终含 text 的助手消息行 inner_text 非空has_body
因此可靠的完成条件为:
A最后一条 assistant message 存在 [data-jcid="message-body"]
B其文本连续 `stable_seconds` 秒无变化
Auser_msg_index 之后最后一条 assistant 的 has_body 为 True
(由 _latest_assistant_after 根据 inner_text 判定)
B其 body_text 连续 `stable_seconds` 秒无变化;
C同时 window.__jc_sending__ 不是 true允许 None / False以容忍
早期 React 未挂载注入钩子的情况)。
@@ -675,10 +709,10 @@ class JiangchangDesktopClient:
assistant = self._latest_assistant_after(user_msg_index)
# 完成判据需同时满足:
# (1) 三路 sending 信号全部静默chat-root / window / 未收起的 execution graph
# (2) user_msg_index 之后存在 assistant 消息
# (3) 该 assistant 存在 message-body
# (4) 该 assistant 的 data-jcid-streaming != 'true'
# (1) 三路 sending 信号全部静默chat-page data-sending / window / 未收起的 execution graph
# (2) user_msg_index 之后存在 assistant 消息data-role 行)
# (3) 该 assistant has_body 为 True_latest_assistant_afterinner_text 非空)
# (4) 该 assistant 的 data-streaming != 'true'
# (5) body 文本连续 stable_seconds 秒无变化
can_check_stable = (
not sending_any
@@ -740,7 +774,7 @@ class JiangchangDesktopClient:
raise JcTimeoutError(
f"等待 AI 回复超时({timeout_ms}ms。user_msg_index={user_msg_index} 最后阶段 phase={phase}"
"可能原因:模型思考时间过长 / 工具链循环未结束 / UI 未挂载 message-body / Gateway 卡住。"
"可能原因:模型思考时间过长 / 工具链循环未结束 / assistant 正文未出现data-role 行 inner_text 为空)/ Gateway 卡住。"
)
# ─── 核心交互ask / read / assert ────────────────────────────
@@ -844,70 +878,56 @@ class JiangchangDesktopClient:
poll_interval=opts.poll_interval,
)
# 6. 选答:只在 my_user_idx 之后的 assistant 中找带 message-body 的最后一条
# 6. 选答:只在 my_user_idx 之后的 assistant 中找 has_body 的最后一条
picked = self._latest_assistant_after(my_user_idx)
if picked is None or not picked["has_body"]:
_logger.warning(
"[ask] user_idx=%d 之后未找到带 message-body 的 assistantpicked=%s",
"[ask] user_idx=%d 之后未找到 has_body 的 assistantpicked=%s",
my_user_idx, picked,
)
# 兜底:取我们 user 之后的任意最后一条 assistant inner_text
try:
page = self.get_page()
nodes = page.locator('[data-jcid="message"]')
nodes = page.locator(SEL_MESSAGE_ROW)
total = nodes.count()
for i in range(total - 1, my_user_idx, -1):
node = nodes.nth(i)
if (node.get_attribute("data-jcid-role") or "").lower() == "assistant":
if (node.get_attribute("data-role") or "").lower() == "assistant":
return node.inner_text(timeout=2000) or ""
except Exception:
pass
return ""
_logger.debug(
"[ask] 选定 assistant_idx=%duser_idx=%d 之后第一条带 message-bodylen=%d",
"[ask] 选定 assistant_idx=%duser_idx=%d 之后 has_bodylen=%d",
picked["index"], my_user_idx, len(picked["body_text"]),
)
return picked["body_text"]
def send_file(self, file_path: str, message: Optional[str] = None) -> None:
if not self._connected:
raise ConnectionError("未连接到应用")
page = self.get_page()
"""[暂未实现] 发送文件附件。
attach_btn = page.locator('[data-jcid="attach-button"]')
if attach_btn.count() == 0:
attach_btn = page.locator("button").filter(
has_text=re.compile(r"附件|上传|paperclip|attach", re.I)
匠厂 v2.0.17(ClawX v0.4.3 合并后)的附件上传通道由原生 Electron
showOpenDialog + IPC stage-paths 实现,不再暴露 HTML <input type="file">。
Playwright 无法直接触发 Electron 原生对话框。需要匠厂主仓库新增:
- window.__jc_stage_paths__(paths: string[]) 调用桥,或
- 一个隐藏 <input type="file"> 备用通道,带 data-testid
在匠厂提供其中之一之前,本方法抛 NotImplementedError 而非静默成功。
Tracking: 见 docs/sdk-migration-discovery.md §7
"""
raise NotImplementedError(
"send_file is temporarily unavailable on jiangchang v2.0.17+. "
"Reason: attachments now go through Electron IPC; SDK requires "
"a new DOM-or-window hook from the host application. "
"See docs/sdk-migration-discovery.md §7."
)
file_input = page.locator('input[type="file"]')
if file_input.count() == 0 and attach_btn.count() > 0:
attach_btn.first.click()
time.sleep(0.5)
file_input = page.locator('input[type="file"]')
if file_input.count() == 0:
raise AssertError(
"找不到文件上传 input[type='file']。请确认匠厂已加 data-jcid=\"attach-button\""
)
abs_path = os.path.abspath(file_path)
if not os.path.exists(abs_path):
raise AppNotFoundError(f"上传文件不存在: {abs_path}")
file_input.set_input_files(abs_path)
time.sleep(1)
if message:
input_locator = self._get_chat_input_locator()
input_locator.click()
input_locator.press_sequentially(message, delay=25)
input_locator.press("Enter")
def read(self) -> List[JiangchangMessage]:
"""
严格读取:仅接受带 data-jcid="message" + data-jcid-role 节点。
严格读取:仅接受带 data-role 语义锚点的 ChatMessage 节点。
无 role 的节点返回 role='unknown'(不再默认为 assistant
"""
if not self._connected:
@@ -917,41 +937,35 @@ class JiangchangDesktopClient:
messages: List[JiangchangMessage] = []
try:
msg_list = self._get_message_list_locator()
msg_list.wait_for(state="visible", timeout=5000)
page.locator(SEL_CHAT_PAGE).wait_for(state="visible", timeout=5000)
except Exception as exc:
_logger.debug("[read] 消息列表不可见:%s", exc)
_logger.debug("[read] 聊天页不可见:%s", exc)
return messages
items = page.locator('[data-jcid="message"]')
items = page.locator(SEL_MESSAGE_ROW)
total = items.count()
_logger.debug("[read] data-jcid=\"message\" 节点数=%d", total)
_logger.debug("[read] data-role 消息行数=%d", total)
for i in range(total):
el = items.nth(i)
try:
role_attr = el.get_attribute("data-jcid-role")
streaming_attr = el.get_attribute("data-jcid-streaming") or "false"
# 优先从 message-body正文气泡取干净文本避免带入 Thinking / ToolCards
body_locator = el.locator('[data-jcid="message-body"]')
if body_locator.count() > 0:
content = body_locator.first.inner_text(timeout=2000)
else:
role_attr = el.get_attribute("data-role")
streaming_attr = el.get_attribute("data-streaming") or "false"
content = el.inner_text(timeout=2000)
role = (role_attr or "unknown").lower()
tool_status = "streaming" if streaming_attr == "true" else None
messages.append(
JiangchangMessage(
id=el.get_attribute("data-jcid-message-id") or f"msg-{i}",
id=el.get_attribute("data-message-id") or f"msg-{i}",
role=role,
content=content,
timestamp=time.time(),
is_error=(el.get_attribute("data-jcid-error") == "true"),
tool_name=el.get_attribute("data-jcid-tool"),
is_error=False,
tool_name=None,
tool_status=tool_status,
)
)
if streaming_attr == "true":
messages[-1].tool_status = "streaming"
except Exception as exc:
_logger.debug("[read] 读取第%d条异常:%s", i, exc)
continue