Read plain-language summaries of each tool step
Agent activity steps now describe what a tool searched for or acted on, and saved activity files carry a readable name.
You expand an agent activity to see what a step did, and the detail line hands you a wall of JSON. Saving that step to your library gave you the same string as a file name.
Readable activity detail lines
A knowledge base search now reads Search: "a", "b". A Linear issue search narrowed by filters reads Search: started. Any other tool step, including Slack, Gmail and Linear updates, shows its inputs as readable pairs such as channel: "#eng", text: "hello team".
Steps that carry a title or a set of links show those instead, so you see the issue you created or the pages that were fetched.
Named saves and downloads
Expanding an activity that read a Slack channel, a Notion page, a Drive file, a Gmail thread, a calendar, a memory or an action list used to name the save and download after the payload behind the card. Click "Add to project" now and the PDF lands in your library under the activity's name, the downloaded file matches, and both tooltips read back the same thing.
Open any agent run and expand a step to see what it was working on at a glance.