Approvals, Runtime Files, and Gmail Triggers

What’s New

Human-in-the-Loop Approvals Gate any agent tool call behind a human decision. Mark a tool as requiring approval and the run pauses when that tool is about to fire, so a person can approve or reject the action before it happens. This is built for tools with real-world side effects, like sending an email or updating a record. See Approvals.

Runtime Files Between Tools Tools within a single agent run can now pass files to one another. A tool that produces a file, such as an attachment saved from an incoming email, makes it available to a later tool that consumes it, so multi-step document workflows hold together from start to finish.

Gmail Inbox Triggers Start an agent automatically when an email arrives. Connect a Gmail inbox as a trigger and incoming messages kick off a run, turning your agents into always-on inbox responders. See Triggers and runs.

Scheduled Runs Tasks now have a Schedule tab for running them on a recurring schedule directly from the dashboard, with no external scheduler required.

Improvements

  • Agent overview now shows a seven-day run summary and indicates which agents are shared with the project
  • A new Trigger column on run listings shows what started each run
  • Clickable resource links in skill instructions
  • Trigger event errors appear in a dedicated column with one-click copy
  • API keys now show their expiry date in the dashboard and via the API
  • Broader image support, including HEIF, with correct handling of image orientation

Approvals are configured on the agent revision, so every change to an approval policy is versioned. Rolling an agent back restores its previous policy alongside the rest of its wiring.