Reply

Your answer to the agent: one of its question's buttons (its id), your words, or both; words alone continue its work.

OS3's job agent on a run that failed or stalled: its state, what it waits for, the runs and dataset versions it made, and its steps as you see them. events, oldest first, one of: trigger (a run failed or stalled), note (its plan, or why it takes its next step), read (what it read of your runs, data or workspace), web_search and web_page, command (code it ran on a sample of your data, with its output and files), data_change (a new dataset version, with what it dropped and why; or, how script, one it made with its own script on its computer, with the script, what changed and the files it left out), job_started (a restart or another run, with its changed settings and estimate), question and reply, data_ready and job_ended (what it waited for), step_failed (a step OS3 refused, and why), outcome (the summary that ends each of its turns) and stopped; on its computer: command (language shell: a command with its output, the files it wrote and kept and the hosts it reached), file (a file it wrote, with its text), browser (a page it opened or worked, with what it typed and its screenshot: GET /workspaces/{id}/agent-computer/steps/{step_id}/files/screenshot.png) and data_copied (a dataset version's files copied in). after pages them: the events from that index on.