July 22, 2026

This release makes the references and outputs in an agent's workbench far easier to organize and act on — clearer lists, consistent actions everywhere, and folders to group things your way.

What's New

• A clearer Generated Outputs list — outputs now show as a sortable, newest-first list with the time each was created and which agent produced it, so two similarly-named outputs are easy to tell apart. Sort by newest, name, or type and filter by type — your choice sticks per workbench
• Consistent actions on every item — hover or right-click any reference or output, in list or grid, to rename, star, download, remove, or save it to your library. Actions work the same way everywhere
• Save an output to your library — turn a great output into a reusable reference in a couple of clicks, instead of asking the agent to recreate it in chat
• Folders for references and outputs — group a task's items into folders, Finder-style: create a folder, drag items in, or right-click to add and remove; open a folder to focus on its contents and use the breadcrumb to step back. Empty folders are fine — make one first and fill it later
• Download on the web, and in bulk — the output viewer now downloads on the web too, not just desktop. Select several items and download them together as a single ZIP, and zipped exports (like metadata bundles) download intact
• Find within a document (⌘F) — search inside a plan, reference, or output to jump between highlighted matches with a running count — handy for QA-ing a long plan. Esc closes, Enter cycles

What's Improved

• A cleaner references picker — the add-references picker now shows each reference's description and how many tasks use it, so duplicates are easy to distinguish. Searching with a type filter active no longer hides real matches, and your filters persist between opens
• Capture details when you save a reference — the new-reference dialog now lets you add a description, type, and tags up front, so your library stays organized from the start
• Fewer near-duplicate outputs — when an agent refines work it already produced, it now updates that output instead of filing a new "v2" each time, keeping your Outputs manageable
• Plans appear in one place — a plan now shows once, under Generated Outputs, instead of also appearing under Resources
• High-severity dependency security patches and behind-the-scenes hardening