[Bug] One failed response (e.g. 502 Bad Gateway) permanently blocks an automation — queue pauses with "The queue was paused because the response failed" #1259

提交前确认 · Pre-submission checklist

• 我已搜索过现有 issue，确认这不是重复 / I searched existing issues and confirmed this isn't a duplicate.
• 我已阅读 CONTRIBUTING.md / I've read CONTRIBUTING.md.

问题类别 · Category

稳定性 / 崩溃 · Stability / Crash

涉及的 Agent 框架 · Agent framework

ZCode Agent（自研）

严重程度 · Severity

阻塞使用 · Blocking (无法使用核心功能 / core function unusable)

复现频率 · Reproducibility

偶现 · Sometimes

问题描述 · Description

A single failed model response (e.g. 502 Bad Gateway, timeout, ECONNRESET — currently frequent per the api.z.ai incident #1214) permanently pauses the session's input queue with the blocking banner:

"The queue was paused because the response failed"

Once the queue is paused, all subsequent scheduled-automation firings for that session are admitted but never dispatched. The automation is effectively dead until a human notices and manually resumes the session — which defeats the purpose of unattended scheduled tasks: exactly during provider instability, when retries matter most, every automation silently stops overnight.

复现步骤 · Steps to reproduce

1. Create a scheduled automation (cron, e.g. every N minutes) on a session in ZCode desktop.
2. Let one firing hit a transient provider error (502 / timeout / dropped stream; simulate a failing response if the provider is healthy).
3. The turn fails; the input queue shows the blocking banner "The queue was paused because the response failed".
4. Observe: every subsequent automation firing is queued but never promoted; the queue stays paused indefinitely. No auto-recovery happens.

期望表现 · Expected behavior

A single transient failure should not permanently disable a scheduled automation. The queue should auto-resume — retry the failed response with backoff (per #1179 / #1226) or at minimum resume on the next scheduled firing. If pausing is intentional as a safety measure, it should be time-bounded and surfaced as a notification/badge, not a silent dead state.

实际表现 · Actual behavior

One failed response → the queue is paused with the blocking banner; later automation runs pile up unprocessed; the automation stays dead until manual intervention (opening the session and resuming manually). During multi-hour provider incidents this kills all affected automations for the whole incident window.

ZCode 版本 · ZCode version

3.16.3.8992

设备 / 系统 / 浏览器 · Device / OS / Browser

Desktop, Windows 11

截图 / 录屏 / 日志 · Screenshots / Recordings / Logs

The banner text quoted above is shown in place of the input queue. No matching lines were found in ~/.zcode/v2/logs by grepping for "queue was paused" — happy to provide full logs and the affected session/automation IDs on request.

Related queue-family reports (same "queued but never promoted" symptom, different triggers): #583, #724, #1112, #1160. Related resilience feature requests: #1179, #1226, #1227. Provider incident making this frequent right now: #1214.