Reading team runs

A team run is one execution of the whole team against an instruction — the lead's plan, every member's contribution, and the consolidated result. Most of the time you read the inbox summary and click through; this article is for when you want the detail.

Where to find runs

Open the team. The conversation sidebar that lists your team chats also lists past runs — they live side by side. Use the filter pills above the list — All / Chats / Runs — to narrow to just runs. The list is grouped by Today, Yesterday, Previous 7 days, and Previous 30 days, with a search box for finding a specific one. Click any entry to open it.

How runs get created

Two paths:

A task you've assigned to the team. When the AI Schedule on the task fires — Run now, Run once, or any recurring cadence — a run is created. The lead reads the task's title and description as the instruction.
A manual Run Team. From the team's header, click Run Team to open the Run Team dialog. Write an instruction and start the run.

The dialog's instruction limit is generous but finite — long briefs are fine; pasting a whole document is not.

Reading a successful run

Click a run to open the detail page. From the top:

Summary. The lead's wrap-up — the headline of what got done.
Metadata. Four cards at a glance: who led the run, how many steps, total duration, and tokens used.
Step timeline. The lead's working steps in order — each step is a card showing what the lead did, what tool they called, and what came back. Handoffs to roster members show up in this timeline as steps too.
Child runs. When the lead delegated work to roster members, each member's contribution appears as a separate child run section. Click through to see that member's own steps in detail.

Reading a failed run

Same layout, with an Error card at the top. Common patterns:

No lead or only one member. The team needs both a lead and at least one other member to run — saving a one-member team is allowed, but running it fails until you add a second agent.
Tool error inside a member. A roster member tried to call a tool they don't have access to. Add the tool in that agent's settings. Integration tool calls show by connector and action (gmail.send_email, jira.create_issue); permission-denied on one usually means the connector needs reconnecting in Settings → Integrations.
Agent paused. A member tried a sensitive integration action and the run stopped waiting for approval. The team run pauses with the held action surfaced in chat; see Approvals and sensitive actions.
Model error or rate limit. Transient. Try again.

Re-running

There's no "retry" button on a failed or completed run. To run the same work again, click Run Team from the team header and write the instruction again. If the work is scheduled on a task, the next scheduled run will use the same task instruction automatically.

Tips and gotchas

Run history is per-team. To see what a single agent on the team did across all of their runs, open the agent directly and use the Runs filter on its sidebar.
The instruction you send through Run Team is read by the lead, not by every member. The lead plans the work and brings members in — you don't address members directly from this dialog.
Long runs take time. If a run is still in progress, the detail page shows partial steps as they stream in; you don't need to wait on a final page load.