Qwen Code v0.21.1 Changelog

This release is all about making the AI a smarter, more independent coder and giving you a much better dashboard to control it from. It matters most to developers who use Qwen Code to automatically review their work, catch bugs, and manage GitHub pull requests, because the AI can now do a lot more of the heavy lifting without needing constant human hand-holding.

What's new

• Active Code Verification: When checking your code, the AI no longer just reads it—it can actually run small tests (called "mutation tests") to prove whether the code actually works.
• Web Dashboard Upgrades: You can now do full Git tasks in the web interface, like picking branches, writing commits, and opening pull requests without leaving the app. You can also widen the sidebar to see more and use a native folder picker.
• GitHub Integration: The system can now read GitHub notifications, respond when someone mentions it, and show you a read-only panel of pull requests.
• Performance Stats: The /stats command now shows you exactly how fast the AI is generating text, including metrics like "time to first token" (how long it takes to start typing) and "tokens per second" (how fast it types).
• Smarter "Skills" Management: You can now set AI tools (skills) to be "soft disabled," meaning they stay off by default but can be easily turned on if you specifically ask for them.
• Chinese Language Support: The AI's deep-verification reports can now be generated in Chinese.
• Image Zooming: Added a tool to zoom into images at full resolution so the AI can see the details better.

Improvements

• Better Rate-Limit Handling: If the AI talks to an outside service that asks it to slow down (rate limiting), you can now customize exactly how long it waits before trying again.
• Instant Trust Updates: If an administrator changes what folders or tools the AI is allowed to use, the AI applies those rules instantly without needing to be restarted.
• Safer Code Reviews: The system now strictly enforces how code reviews are submitted. It even has a "tripwire" to catch and flag anyone trying to sneak around the rules using terminal commands.
• Fewer Duplicate Errors: If the automated testing system finds the same error twice, it just adds a comment to the existing bug report instead of spamming you with new ones.
• Persistent Terminal History: If the web terminal crashes or hits an error while loading your past commands, it now saves that error info so it doesn't get lost.

Fixes

• Timezone Bug: The AI was calculating "insight days and hours" using the wrong timezone in some places. It now correctly uses your local time everywhere.
• Crash on Empty Cleanup: The system used to crash when trying to clean up temporary files if there were no files to clean. It now handles empty folders safely.
• Confused Background Shells: The AI sometimes misread what was happening in quiet, background terminal windows. A "status sidecar" (a small file that just reports status) was added so the AI knows exactly what the terminal is doing.
• Worktree Leaks: Temporary coding spaces (worktrees) used during reviews were sticking around and cluttering the system. They are now properly cleaned up at the end of every prompt.

Breaking changes

No breaking changes in this release.