🚀 CodeBuddy Code v2.111.0 Release

📦 Version Information

Component	Version
CodeBuddy Code CLI	v2.111.0
Agent SDK JS	v0.3.191
Agent SDK Python	v0.3.190

✨ New Features

Message Queue Editing

When the Agent is responding, messages typed by the user enter a queue. You can now press the ↑ arrow key to pull a queued message back into the input box for further editing, and an empty input box shows a "Press ↑ to edit queued messages" hint. System-injected queue items (async notifications, channel messages, etc.) remain in the queue and continue to be processed automatically.

Message Queue Mode

• Press Tab while a run is in progress to queue a message, sending them one by one after the current turn ends
• Enter still performs real-time steer (instant guidance)
• When pressing Esc to interrupt the current turn, if only queue messages exist they are filled back into the input box for continued editing
• New Option+↑ (Alt+↑) shortcut to pull the last queued message out for editing
• The thinking mode toggle key migrates from Tab to Option+T (Alt+T) to avoid conflicts

Channel Auto Connect

• Added CODEBUDDY_CHANNEL_AUTO_CONNECT / CODEBUDDY_WECOM_AUTO_CONNECT / CODEBUDDY_WECHAT_AUTO_CONNECT environment variables to automatically connect configured WeCom/WeChat bots at startup
• WeCom bots configured via the CODEBUDDY_WECOM_BOT_ID + CODEBUDDY_WECOM_BOT_SECRET environment variables now automatically appear in the /remote-control list

🔧 Improvements

• Mode Switch Shortcut: Windows now supports Shift+Tab to switch modes (consistent with macOS/Linux), while keeping Alt+M as an alias, allowing users to customize overrides in keybindings.json
• Background Task Auto-cleanup: In long sessions, terminated background tasks are automatically cleaned up after notification, preventing continuous memory growth; the task panel only shows in-progress tasks
• ACP Tool Call Failure Handling: Distinguishes three failure types (tool_not_found / deferred_tool_direct_call / invalid_tool_input), so the frontend can provide more precise hints based on the failure type
• Auto+Plan Composite Permission Mode: When entering Plan mode from Auto mode, the Auto classifier stays active and automatically restores after exiting Plan
• WeCom/WeChat Multi-instance Kick: Ensures the most recently connected instance stays active; when a new instance connects, the old one proactively yields
• WeChat Session Timeout Credential Retention: Credentials are no longer cleared after being kicked out, so users can manually reconnect without re-scanning the QR code

🐛 Bug Fixes

• Session AI Title Generation: Fixed the issue where title generation could not capture the real user task name in scenarios involving system prompts, command wrapping, and short noisy input; fixed the issue where a session permanently had no title after the first title generation failed
• Task Panel Shortcut: Unified the x key to terminate tasks, and restored j/k up/down navigation
• Rewind Revert: Fixed the issue where stale version numbers caused files to not be correctly restored when reverting with /rewind
• Prewarm Process First Handshake: Fixed the issue where the first initialize handshake occasionally hung for about 10s after the prewarm process was activated, stabilizing initialize at the millisecond level
• MCP Tool Injection: Fixed the issue where non-deferred built-in tools of connector-proxy could not be injected into the LLM tool list in prewarm + serve mode
• QR Scan Panel: Fixed the issue where pressing ESC could not exit during a busy state
• Configuration Wizard Input Box: Fixed the issue where the Bot ID was carried into the Secret input box in the WeCom bot configuration wizard

📝 Documentation

• Version Troubleshooting Guide: Added a troubleshooting guide for "still on the old version after updating", covering common scenarios such as Homebrew/nvm/residual binaries/shell cache