codebee 0.1.5 → 0.1.7

package/app/core/automation.py

@@ -299,11 +299,36 @@ def _launch_run(t):
                                     time.strftime("%m-%d %H:%M"))).strip()[:40],
                "goal": t.get("prompt") or "",
                "workdir": (t.get("workdir") or "").strip()}
-    task = store.create_task(payload)
-    run = store.create_run("orchestration", task["title"], task_id=task["id"])
-    store.update_task_status(task["id"], "queued")
-    jobs.enqueue({"kind": "orchestration", "run_id": run["id"], "task_id": task["id"]})
-    return run["id"]
+    task = None
+    run = None
+    try:
+        task = store.create_task(payload)
+        run = store.create_run("orchestration", task["title"], task_id=task["id"])
+        store.update_task_status(task["id"], "queued")
+        jobs.enqueue({"kind": "orchestration", "run_id": run["id"],
+                      "task_id": task["id"]})
+        return run["id"]
+    except Exception:
+        # A scheduler failure must not leave a task/run that looks queued forever.
+        # Keep the public error generic; the detailed traceback stays in the
+        # service log and the run record remains useful for diagnostics.
+        log.exception("automation: 运行入队失败 task=%s run=%s",
+                      (task or {}).get("id"), (run or {}).get("id"))
+        if run:
+            try:
+                store.update_run(run["id"], status="failed",
+                                 error="任务入队失败，请稍后重试",
+                                 ended_at=time.strftime("%Y-%m-%d %H:%M:%S"))
+            except Exception:
+                log.exception("automation: 运行失败收口失败 run=%s", run.get("id"))
+        elif task:
+            # create_run may fail before a run is available. The task was
+            # already persisted by create_task, so mark it failed as well.
+            try:
+                store.update_task_status(task["id"], "failed")
+            except Exception:
+                log.exception("automation: 任务失败收口失败 task=%s", task.get("id"))
+        raise
 
 
 def _fire(snapshot, now):

package/app/core/catalog.py

@@ -10,6 +10,7 @@ from __future__ import annotations
 import copy
 import json
 import re
+import sys
 import threading
 
 from . import paths
@@ -216,6 +217,18 @@ CONFIG_PATCH = {
 }
 
 
+# 安装命令平台修正（{id: (install, upgrade)}）：仅非 Windows 套用。winget 是
+# Windows 包管理器、py 启动器 Windows 独有，这两类命令在 macOS/Linux 上必失败
+# ——load() 幂等改写为等价的 npm/pip 渠道（与 CONFIG_PATCH 同策略：覆盖无风险，
+# 旧数据不修正的话每次点安装/升级都注定失败）。
+INSTALL_PATCH_NONWIN = {
+    "claude-code": ("npm install -g @anthropic-ai/claude-code",
+                    "npm install -g @anthropic-ai/claude-code@latest"),
+    "aider": ("python3 -m pip install -U aider-chat",
+              "python3 -m pip install -U aider-chat"),
+}
+
+
 def npm_pkg_name(cmd):
     """从 npm 安装命令里取包名（支持 @scope/name@latest）。
 
@@ -289,6 +302,15 @@ def _apply_config_patch(entries):
             e["config"] = dict(patch)
 
 
+def _apply_install_patch(entries):
+    if sys.platform == "win32":
+        return
+    for e in entries:
+        patch = INSTALL_PATCH_NONWIN.get(e.get("id"))
+        if patch:
+            e["install"], e["upgrade"] = patch
+
+
 def _merge_new_defaults(entries):
     """把内置默认里「新增的」条目补进已加载清单（同 id 已存在则原样保留）。
 
@@ -321,6 +343,7 @@ def load(force=False):
         _apply_resume_patch(entries)
         _apply_launch_patch(entries)
         _apply_config_patch(entries)
+        _apply_install_patch(entries)
         _CACHE["entries"] = entries
         return entries
 

package/app/core/health.py

@@ -59,6 +59,13 @@ def init(data_dir=None):
                     _PROVIDERS[name] = st
         except (OSError, json.JSONDecodeError):
             pass
+        # 「绑定链·」静态告警胶囊已从产品移除（2026-09-18：普通用户误伤，
+        # 用户拍板）——老版本持久化的条目启动即清，别让 0.1.6 的状态阴魂不散。
+        stale = [k for k in _PROVIDERS if k.startswith("绑定链·")]
+        if stale:
+            for k in stale:
+                _PROVIDERS.pop(k, None)
+            _persist()
         if not _STARTED:
             _STARTED = True
             t = threading.Thread(target=_probe_loop, name="health-probe", daemon=True)
@@ -157,6 +164,8 @@ def report_failure(provider: str, error: str = "", *, model: str = "",
         _persist()
 
 
+# ------------------------------------------------- 静态死链告警（绑定层）
+
 # ---------------------------------------------------------------- 手动操作
 
 def silence(provider: str, minutes: int = 0):
@@ -215,6 +224,7 @@ def snapshot():
                 "last_error": st.get("last_error") or "",
                 "alerting": alerting,
                 "silenced": silenced,
+                "static": bool(st.get("static")),
             })
         # 稳定排序：告警中 > 故障中 > 恢复 > 正常
         rank = {"down": 0, "failing": 1, "recovered": 2, "ok": 3}
@@ -271,7 +281,8 @@ def _probe_loop():
         try:
             with _LOCK:
                 targets = [(name, st) for name, st in _PROVIDERS.items()
-                           if st.get("status") in ("failing", "down")
+                           if not st.get("static")  # 静态死链无端点可探，绑定恢复时自动解除
+                           and st.get("status") in ("failing", "down")
                            and (st.get("probe_next_at") or 0) <= _now()]
                 for name, st in targets:
                     idx = int(st.get("probe_backoff_idx") or 0)

package/app/core/jobs.py

@@ -134,7 +134,9 @@ def cancel_event_for(run_id):
     return ev
 
 
-AUTO_RESUME_MAX = 2      # 连载任务自动续跑上限（超时/中断后自动接着写，无需人工）
+AUTO_RESUME_MAX = 3      # 连载任务自动续跑上限（超时/中断后自动接着写，无需人工）
+AUTO_RESUME_DELAY_S = 300  # 自动续跑延迟入队秒数：网关限流/欠费窗口通常分钟级，
+                           # 立即重排会撞在同一堵墙上把续跑次数烧光（2026-09-17 七猫实测）
 
 
 def _maybe_auto_resume(run_id):
@@ -160,7 +162,13 @@ def _maybe_auto_resume(run_id):
             return False
         store.update_run(new_run["id"], auto_resumes=int(run.get("auto_resumes") or 0) + 1,
                          auto_resumed_from=run_id)
-        _QUEUE.put({"kind": "orchestration", "run_id": new_run["id"], "task_id": task["id"]})
+
+        def _enqueue():
+            _QUEUE.put({"kind": "orchestration",
+                        "run_id": new_run["id"], "task_id": task["id"]})
+        t = threading.Timer(AUTO_RESUME_DELAY_S, _enqueue)
+        t.daemon = True
+        t.start()
         return True
     except Exception:
         return False

package/app/core/manager.py

@@ -9,7 +9,9 @@ from __future__ import annotations
 import json
 import os
 import re
+import shlex
 import shutil
+import signal
 import socket
 import subprocess
 import sys
@@ -21,7 +23,8 @@ from pathlib import Path
 
 from . import catalog, paths, runner
 
-CREATE_NO_WINDOW = 0x08000000
+# 非 Windows 置 0：POSIX 的 Popen 对非零 creationflags 抛 ValueError（runner 同款守卫）
+CREATE_NO_WINDOW = 0x08000000 if os.name == "nt" else 0
 VERSION_TTL = 300  # 版本缓存 5 分钟
 
 _LOCK = threading.RLock()
@@ -69,6 +72,43 @@ def detect_entry(entry):
     return {"installed": False, "detail": ""}
 
 
+def _orphan_signature(cl):
+    """「Tutti 专属调用签名」判定（两个平台的清扫共用）。"""
+    return (("opencode" in cl and "--model" in cl)
+            or ("codex" in cl and "--skip-git-repo-check" in cl)
+            or ("kimi-code" in cl and "main.mjs" in cl))
+
+
+def _sweep_orphans_posix():
+    """POSIX 版清扫：ps 一次拉全量，签名同 Windows。孤儿判据不同——POSIX 的
+    孤儿进程会被内核过继给 1 号进程（launchd/init），Windows 则保留死掉的
+    ppid，所以这里认 ppid==1；用户手动在终端里跑的同名 CLI 父进程是活着的
+    shell（ppid!=1），不会被误杀。"""
+    try:
+        r = subprocess.run(["ps", "-eo", "pid=,ppid=,command="],
+                           capture_output=True, timeout=60)
+        killed = 0
+        for ln in r.stdout.decode("utf-8", "replace").splitlines():
+            parts = ln.strip().split(None, 2)
+            if len(parts) != 3 or not parts[0].isdigit() or not parts[1].isdigit():
+                continue
+            pid, ppid, cmd = int(parts[0]), int(parts[1]), parts[2]
+            if pid == 1 or ppid != 1 or not _orphan_signature(cmd.lower()):
+                continue
+            try:
+                os.killpg(pid, signal.SIGKILL)  # 服务 spawn 用了 start_new_session，pgid==pid
+                killed += 1
+            except Exception:
+                try:
+                    os.kill(pid, signal.SIGKILL)
+                    killed += 1
+                except Exception:
+                    pass
+        return killed
+    except Exception:
+        return 0
+
+
 def sweep_orphan_cli_processes():
     """启动清扫：服务重启会孤儿化正在跑的 CLI 孙进程（外部只杀服务 PID，不带
     /T），僵尸 opencode 更会劫持后续会话——opencode 是客户端-服务端架构，新
@@ -78,7 +118,11 @@ def sweep_orphan_cli_processes():
     按「Tutti 调用签名 + 父进程已死」双条件匹配，不误杀用户自己在用的 CLI：
       opencode：命令行含 opencode + --model（同步写入的 provider 固定 orch）
       codex：命令行含 codex + --skip-git-repo-check（Tutti 专属 flag 组合）
+      kimi：命令行含 kimi-code/dist/main.mjs（node 直启路径，2026-09-17 实测
+      僵尸 kimi 会占住讯飞网关同钥请求队列，堵死后续所有 kimi 调用）
     claude 不扫（签名与用户手动使用难区分）。返回清扫数量。"""
+    if os.name != "nt":
+        return _sweep_orphans_posix()
     ps_exe = os.path.join(os.environ.get("SystemRoot", r"C:\Windows"),
                           "System32", "WindowsPowerShell", "v1.0", "powershell.exe")
     if not os.path.isfile(ps_exe):
@@ -99,9 +143,7 @@ def sweep_orphan_cli_processes():
         for i in items:
             cl = str(i.get("CommandLine") or "").lower()
             ppid = i.get("ParentProcessId")
-            is_ours = (("opencode" in cl and "--model" in cl)
-                       or ("codex" in cl and "--skip-git-repo-check" in cl))
-            if not is_ours or ppid in live or not i.get("ProcessId"):
+            if not _orphan_signature(cl) or ppid in live or not i.get("ProcessId"):
                 continue
             try:
                 subprocess.run(["taskkill", "/F", "/T", "/PID", str(i["ProcessId"])],
@@ -165,6 +207,8 @@ def _uwp_version(package_dir):
 
 
 def _exe_version(path):
+    if os.name != "nt":
+        return None  # exe 版本探测是 Windows 专属功能（PowerShell 读 PE 资源）
     try:
         r = subprocess.run(
             ["powershell", "-NoProfile", "-Command",
@@ -188,7 +232,11 @@ def version_of(entry):
     cli = (entry.get("detect") or {}).get("cli")
     if cli and det.get("installed"):
         try:
-            r = subprocess.run(["cmd", "/c", cli, "--version"], capture_output=True,
+            # Windows 下 CLI 可能是 npm .cmd 垫片，须经 cmd /c 才能直接点名跑；
+            # POSIX 没有垫片，符号链接直接跑即可
+            probe = ["cmd", "/c", cli, "--version"] if os.name == "nt" \
+                else [cli, "--version"]
+            r = subprocess.run(probe, capture_output=True,
                                creationflags=CREATE_NO_WINDOW, timeout=20)
             out = (r.stdout or b"").decode("utf-8", "replace").strip()
             if not out:
@@ -832,6 +880,11 @@ def _sync_codex_settings(entry, model, cp):
     name = cp.get("name", "orch")
     def q(v):
         return '"%s"' % str(v).replace("\\", "\\\\").replace('"', '\\"')
+    if (cp.get("wire_api") or "responses") == "chat":
+        # codex 0.154+ 起 chat wire 被官方移除，写进 config.toml 会让 CLI 连配置
+        # 都载入不了（Error loading config.toml）——宁可明确拒绝也不落坏配置。
+        return ("供应商只有 chat completions wire，codex 0.154+ 已移除支持，未写入"
+                " config.toml——请为 codex 绑定 responses 兼容的供应商")
     pairs = [("name", q(cp.get("name", name))),
              ("base_url", q(cp.get("base_url", ""))),
              ("env_key", q(cp.get("env_key", "ORCH_API_KEY"))),
@@ -1432,9 +1485,13 @@ def launch(entry, open_browser=True):
         spawn_cmd = "%s > %s 2>&1" % (cmd, ('"%s"' % ls) if " " in ls else ls)
         try:
             log_path.parent.mkdir(parents=True, exist_ok=True)
-            subprocess.Popen(["cmd", "/c", spawn_cmd], cwd=str(paths.ROOT), env=env,
+            # 重定向语法 sh 与 cmd 通用，只换解释器；start_new_session 让子服务
+            # 脱离本服务进程组（杀树/退出互不牵连）
+            shell = ["cmd", "/c"] if os.name == "nt" else ["/bin/sh", "-c"]
+            subprocess.Popen(shell + [spawn_cmd], cwd=str(paths.ROOT), env=env,
                              stdin=subprocess.DEVNULL, stdout=subprocess.DEVNULL,
-                             stderr=subprocess.DEVNULL, creationflags=CREATE_NO_WINDOW)
+                             stderr=subprocess.DEVNULL, creationflags=CREATE_NO_WINDOW,
+                             start_new_session=(os.name != "nt"))
         except Exception as e:
             return {"ok": False, "error": "无法启动服务: %r" % e}
         if open_browser:
@@ -1444,8 +1501,25 @@ def launch(entry, open_browser=True):
                 "message": "%s 正在启动，就绪后浏览器会自动打开（%s）%s"
                            % (name, bare, ("；" + extra) if extra else "")}
 
+    if sys.platform == "darwin":
+        # Terminal.app 新开窗口跑交互 TUI；do script 的命令串常驻窗口，等价
+        # Windows 的 cmd /k（CLI 退出后窗口保留，报错不至于一闪而过）。
+        # 两层转义各管各的：shlex.quote 管 shell 层（cd 路径的引号），
+        # 反斜杠/双引号替换管 AppleScript 字符串层。
+        shell_cmd = "cd %s && %s" % (shlex.quote(str(paths.ROOT)), cmd)
+        asc = 'tell application "Terminal" to do script ' + \
+              '"' + shell_cmd.replace("\\", "\\\\").replace('"', '\\"') + '"'
+        try:
+            subprocess.Popen(["osascript", "-e", asc], cwd=str(paths.ROOT), env=env,
+                             stdin=subprocess.DEVNULL, stdout=subprocess.DEVNULL,
+                             stderr=subprocess.DEVNULL)
+        except Exception as e:
+            return {"ok": False, "error": "无法打开终端窗口: %r" % e}
+        return {"ok": True, "kind": "console", "message": "已在新的终端窗口打开 %s%s"
+                % (name, ("（" + extra + "）") if extra else "")}
+
     if sys.platform != "win32":
-        return {"ok": False, "error": "终端窗口拉起暂仅支持 Windows"}
+        return {"ok": False, "error": "终端窗口拉起暂仅支持 Windows/macOS"}
     # start 为目标命令新开一个可见终端窗口；cmd /k 让 CLI 退出后窗口保留，
     # 报错不至于一闪而过。外层 cmd 用 CREATE_NO_WINDOW 隐藏。
     argv = ["cmd", "/c", "start", "CodeBee %s" % name, "/D", str(paths.ROOT),
@@ -1536,7 +1610,10 @@ def check_update(entry, force=False):
     cmd = entry.get("install") or entry.get("upgrade") or ""
     pkg = _npm_pkg_name(cmd)
     if pkg:
-        r = runner.run_process(argv=["cmd", "/c", "npm", "view", pkg, "version"], timeout=90)
+        # Windows 的 npm 是 .cmd 垫片须经 cmd /c；POSIX 直接跑
+        npm_view = ["cmd", "/c", "npm", "view", pkg, "version"] if os.name == "nt" \
+            else ["npm", "view", pkg, "version"]
+        r = runner.run_process(argv=npm_view, timeout=90)
         latest = ""
         if r["ok"]:
             for line in (r["stdout"] or "").splitlines():
@@ -1550,7 +1627,7 @@ def check_update(entry, force=False):
             result["updatable"] = bool(_ver_tuple(latest) > _ver_tuple(cur_num))
             if not result["updatable"]:
                 result["note"] = "已是最新版本"
-    elif "winget" in cmd:
+    elif os.name == "nt" and "winget" in cmd:
         m = re.search(r"--id\s+([A-Za-z0-9._-]+)", cmd)
         wid = m.group(1) if m else None
         if not wid: