Qwen Code 0.26.5 adds interactive card context and bounded PR autofix

Qwen Code 0.26.5 updates both the CLI and Desktop. The overlapping release notes are covered together rather than counted as separate launches.

Continue from an interactive card

Generated UI is enabled by default. Values from an interactive card can be attached as removable context above the chat composer, so users can continue a conversation from that state. A new Workflow switch changes saved session preferences for the next turn without interrupting the current response.

Put a boundary around PR autofix

The PR-watch skill adds an --in-place mode that can work in the current checkout, push the branch and reply on the pull request. That is a write-capable workflow, so teams should review its scope before using it on shared work.

A retry budget hands a failing check to a person after three fix pushes. Blocked rounds post what was attempted, what prevented progress and what human action is needed. The notes also say PR watch merges a stale base before attempting fixes when CI failures may come from a branch falling behind.

Managed deployment and history

An experimental Helm deployment combines Managed Agent, Hosted Harness, MySQL, local execution and an authenticated gateway. It is explicitly experimental, not a blanket production-readiness claim. Managed history gains bounded SQL queries and upward paging while live updates continue.

These are agent-interface and workflow changes, not new model weights or an included inference allowance. This summary follows the project’s release notes; it does not claim an independent test of the automation or its safety.

See the Qwen Code hub.