- Restrict code fence stripping to fences at start-of-line (prevents inline "~~~markdown" from being treated as a fence and merging lines)
- Only strip fences for full JSON templates in Agent.read_prompt() and files.parse_file()
- Preserve fenced examples in markdown/tool prompts while keeping json-only prompts parseable
- Add regression tests for fence stripping and prompt fence policy
- Add Z.AI provider with base URL https://api.z.ai/api/paas/v4
- Add Z.AI Coding provider with base URL https://api.z.ai/api/coding/paas/v4
- Both use OpenAI-compatible protocol via LiteLLM
Available models: glm-4.6, glm-4.6v (vision), glm-4.5, glm-4.5-air
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
* **Mechanism**: `EventLoopThread.terminate()` now correctly stops the asyncio loop and joins the thread, removing it from the global registry.
* **Cleanup**: `DeferredTask.kill(terminate_thread=True)` now invokes `_drain_event_loop_tasks()`, which runs **inside** the target thread to explicitly cancel and await all pending tasks (including monologue loops) before killing the thread. This prevents "Task was destroyed but it is pending" warnings and ensures clean exits.
* **Tracking**: The scheduler now maintains a live registry (`_running_deferred_tasks`) of active `DeferredTask` objects, protected by a reentrant lock.
* **State Management**: The `run_task` wrapper uses `asyncio.shield` to ensure that even when a task is cancelled (e.g., by user action), the task state is reliably reset to `IDLE` in the database, preventing tasks from getting stuck in `RUNNING` state.
The fix is correctly propagated to all relevant destruction points using `terminate_thread=True`:
* **Dedicated Context**: `scheduler_task_delete.py` cancels the specific running task and terminates its thread.
* **Shared/Dedicated Context**: Both `chat_remove.py` (Delete Chat) and `chat_reset.py` (Reset Chat) now call `scheduler.cancel_tasks_by_context(...)`. This ensures that if a scheduler task is running in a chat window (monologue), resetting that chat immediately kills the background thread and stops the agent loop.
