Qwen Code v0.22.2 Changelog

This update is all about making the AI a better, safer teammate for developers. It focuses heavily on improving how the AI reviews code (giving clearer reasons for its suggestions and avoiding endless back-and-forth arguments), while also making the interface cleaner and ensuring the AI asks for permission before running complex tasks. It matters most to developers who use this tool to help them write, check, and manage code.

What's new

• Smarter code reviews: The AI now gives clearer feedback on your code. It provides specific rules on how to fix issues and adds a "do-not-refute" list so it won't accidentally reject good suggestions.
• New AI models: You can now easily use Kimi (by Moonshot AI) as a built-in AI provider.
• Output styles: You can choose how the AI talks to you with four new styles: Concise, Proactive, Explanatory, and Learning.
• Safer workflows: The AI will now show you a detailed pop-up box (with the script name and arguments) and require your explicit permission before running any complex, multi-step tasks.
• Dead code finder: A new tool scans your project to find unused "dead" code and orphaned files, then suggests ways to clean them up.
• DingTalk integration: You can now connect the tool directly to DingTalk (a popular messaging app) to get direct messages, mentions, and to-do tracking.
• Web Shell security: You can now open a local web interface with a secure password token to keep your work safe.
• Scheduled tasks: You can now set up automated tasks that reuse an existing, live session instead of starting a brand-new one.

Improvements

• Better web interface: The "compact mode" in the web shell is much cleaner. It groups the AI's thinking, tool usage, and parallel tasks into one single line that you can click to expand, instead of cluttering the screen.
• Remembering settings: If the background system (daemon) restarts, it now remembers which AI model you were using instead of resetting to the default.
• Tracking AI usage: The system now tracks exactly how many "tokens" (pieces of words the AI processes) a specific goal used, right alongside how many turns it took.
• Cleaner VS Code: The VS Code extension now uses the exact same chat display as the web version, making the experience consistent and replacing the old, clunky layout.
• Safer code execution: When the AI reviews code, it runs the project's own test commands inside a secure, isolated container instead of running them directly on your computer.
• Faster critical warnings: If the AI spots a serious, critical bug, it will warn you immediately rather than waiting for a scheduled round 6 of review.

Fixes

• Web Shell skill bug: Fixed an issue where turning off a "skill" in the web interface wouldn't actually refresh the list, causing disabled skills to pop back up. It now shows a clear error if a refresh fails instead of showing old, stale data.
• Invalid comment anchors: Fixed a problem when leaving comments on certain code targets. If the exact line reference (anchor) is invalid, it now safely moves critical warnings to the main summary and deletes minor suggestions instead of failing or posting broken comments.

Breaking changes

• The old Electron desktop app has been completely removed (the code was moved to a different project). If you relied on the desktop app, you will need to switch to the web shell or VS Code extension.
• The computer-control feature (Computer Use) was rebuilt. The old built-in tools were replaced with a new setup that automatically configures an external server (Node REPL MCP) the first time you use it.