Qwen Code v0.21.12 Changelog

This release is all about making the app more reliable and user-friendly, especially for developers using it to review code. It gives you better ways to upload files, stops the AI from making wild or unverified claims when checking code, and fixes a bunch of annoying visual glitches on the desktop app.

What's new

• Drag-and-drop file uploads: You can now drag files right into the web composer (or use the @ file panel) to share them with the app, and you'll see a progress bar while they upload.
• Smarter code reviews: When the AI reviews code, it now groups similar bugs together instead of listing them one by one. It also evaluates feedback based on whether it's actually correct, not just on who wrote it.
• "Thinking" summaries: If you use Compact Mode (Ctrl+O), you can now see the AI's reasoning as a foldable "Thinking..." summary, rather than having it hidden completely.
• Better web shell management: The interface for managing Web Shell channels got a redesign and now supports full policy configuration and workspace binding.
• More user control: You can now use your personal settings to control things like default review effort levels and comments, stopping repository files from overriding what you want.

Improvements

• AI auto-fix limits: The AI now has a "growth brake" when automatically fixing code. If it tries to add too many lines of code or tests, it will stop and ask a human maintainer for help instead of patching endlessly.
• Proof required for critical bugs: If the AI flags a "Critical" bug, it now has to run a test to prove it. If it can't prove the bug is real, the claim gets downgraded to "low confidence" and hidden from public pull request posts.
• Safer background tasks: The app now tracks when background terminal shells are still working, which prevents it from accidentally cleaning up or deleting tasks that are still running.
• Better logging for developers: For the engineers behind the scenes, the app now adds special tracking IDs (OpenTelemetry traces) to logs. This makes it much easier to debug what went wrong if a request fails.
• Accurate review tracking: Review ledgers (the app's memory of what it has checked) now use a secure anchor ID to ensure it never accidentally certifies code it hasn't actually reviewed.

Fixes

• Jittery desktop sidebar: Fixed an annoying visual glitch where the project list in the desktop app's sidebar would jump around. It now reserves space for the scrollbar.
• Links opening in the wrong place: Fixed an issue where clicking web links in the desktop app wouldn't work. They now reliably open in your computer's default web browser.
• First message failing: Fixed a bug in the standalone Web Shell where your very first prompt might fail to send due to a synchronization error.
• Duplicate buttons: Fixed restricted secondary workspaces in the web shell showing two identical "archive" buttons. Now only one shows up.
• Windows startup flash: Prevented an extra terminal window from flashing on your screen when starting the app on Windows, and fixed the startup logo's centering.
• Security gap in quotes: Fixed a security loophole where tricky inline text quotes could bypass approval limits. Now, only properly formatted block quotes can trigger those limits.
• Cleaner test runs: Fixed an issue where your computer's local Git settings could "poison" the AI's test runs. The tests are now isolated (hermetic) so they aren't affected by your personal setup.
• Windows installer fixes: The Windows installer now uses built-in Windows tools to verify downloads instead of using PowerShell, which prevents failures when checking file hashes.

Breaking changes

No breaking changes in this release.