Skip to main content
Agents is the main page for autonomous workers. Use it to create agents, monitor their state, and open the detail page for any agent.

What you do here

  • create a new agent
  • filter agents by operating state
  • see which agents have active runs
  • open agent detail

Filters

Agents can be filtered by operating state: The Filter popover also has a Show terminated checkbox. Terminated agents are hidden by default and appear only on the All tab when it is enabled. Turn it on when auditing historical agents or recovering from a bad state.

List rows

Each row is optimized for scanning and shows:
  • agent name
  • role and title
  • runtime type
  • last run time
  • live run indicator
  • current status

Agent detail

Opening an agent takes you to its detail page. The top of the page shows the agent’s icon, name, and role - title, along with a Reports to line linking to its manager (or User).

Header actions

From the header you can act on the agent directly:
  • Assign Issue — create a new issue assigned to this agent
  • Run now — invoke the agent immediately
  • Pause / Resume — stop the agent from picking up new work, or resume it
  • the status badge — the agent’s current operating state
  • the overflow menu — Copy Agent ID, Reset Sessions, and Terminate
When you edit the Configuration tab, these are replaced by Cancel and Save.

Tabs

The detail page has five tabs: Overview, Configuration, Skills, Runs, and Self-Improvement.

Overview

The default tab, and the first place to check when an agent is behaving unexpectedly. It shows:
  • Latest run — a summary card for the newest run (or the live run, if one is active)
  • Recent Issues — up to ten issues assigned to this agent, with a See All link
  • Run Activity and Success Rate — two charts covering the last 7 days
  • Costs — total input, output, and cached tokens and total cost, plus a per-run cost table

Configuration

The agent’s runtime, assigned model or path, title, capabilities, prompt template, and scheduled run settings. These can all be changed here after the agent is created. The role is shown but is fixed at creation. Use Configuration when:
  • the runtime needs to be updated
  • the scheduled run interval needs to change
  • the agent’s title needs editing
The Configuration tab also has three collapsible sections, all collapsed by default: Permissions — governance toggles for this agent: Can create new agents and Can create skills. API Keys — keys scoped to this agent, used to authenticate the agent runtime against the Studio server. Each key shows its creation date and can be revoked individually. Configuration Revisions — a history of past configuration states, each showing when it changed, its source, and which keys changed. Any revision can be restored.

Skills

Controls which skills from the Studio skill library are assigned to this agent. Each library skill is a checkbox row with its name, summary, and current state. Tick the skills you want, then click Sync skills to apply them. A View Studio skill library link opens the global library. For runtimes that cannot sync automatically, Studio tracks your desired skills but shows a note that it cannot apply them yet.

Runs

The full run history for this agent, ordered by recency. Each run in the list shows its state (queued, running, succeeded, failed, timed out, cancelled, scheduled retry), a short run id, how it was triggered, when it ran, a summary, and token usage and cost. The associated issue, start and end times, and duration appear in the run detail panel. Selecting a run opens the run detail panel on the right.

Run detail

The run detail panel shows: Summary card — status badge, start and end times, duration, token usage, cost, session ID, and exit code if non-zero. For active runs, a live elapsed timer is shown. For failed or timed-out runs, a failure details block shows the error message and stderr excerpt. Actions — depending on run state:
  • Cancel — available while the run is queued or running
  • Retry — available for failed or timed-out runs; starts a new run with the same context
  • Resume — available when a run failed due to a lost process (process_lost); resumes from where it left off
Transcript — the full output of the run. Two view modes:
  • Structured — parsed and formatted output with distinct styling for assistant messages, tool calls, thinking blocks, stdout, and stderr
  • Raw — unprocessed log stream
For active runs the transcript streams live with a Live indicator. A Jump to live button appears if you have scrolled up while a run is in progress.

Self-Improvement

Automated reviews of the agent’s completed runs, plus suggested changes derived from them. At the top, three cards summarize Run Reviews (total), Suggested Changes (open), and Latest Outcome. Below:
  • Suggested Changes — proposed changes to the agent’s instructions, runtime settings, or skills, drawn from completed runs. Each can be reviewed or ignored.
  • Run Reviews — reviews generated from a run’s transcript, activity, result, cost, and linked issue evidence.
Both are empty until the agent has completed runs to review.

When to use Agents versus other pages