Task clarity · Creative previews · Skill marketplace

Task timelines now show cleaner localized progress summaries, with less raw tool noise and fewer intermediate process groups while work is still streaming.
Generated artifacts tuck away more naturally while a task is active, and deleted attachments are pushed to the end of attachment grids so current files stay easier to scan.
Browser-based continuations are more reliable during auth handoffs and callbacks, so long-running generation flows are less likely to lose their place.

Image and video generation flows preserve reference images more consistently and avoid provider watermarks by default.
Markdown and bullet rendering in task replies is stricter, so generated plans and notes keep their intended structure.
Research-rendered pages and design-preview skills are easier for the agent to discover and use without extra setup.

Skill cards and detail dialogs now show clearer author and source information, including GitHub-backed attribution for refreshed market skills.
Tool progress groups reveal more useful action details, making it easier to understand what the agent is doing without exposing raw commands.

A new public prompt page brings paired GPT Image 2 and Seedance 2 examples together for image-to-video creative workflows.