Home Help Zone 3 · Execution Execution Runs

Execution Runs

An execution run is created each time a backlog item is dispatched to an AI Pod. It tracks the full lifecycle of the AI agent’s work — from queueing through planning, execution, code commit, and pull request — including logs, cost tracking, and approval checkpoints.

Viewing Runs

There are two ways to access execution runs:

  • Global runs list — click Runs in the sidebar to see all runs across all projects
  • Project runs list — inside a project, click the Runs tab to see runs for that project only

Both views show the same information, filtered by scope.

Run Statuses

StatusMeaning
QueuedDispatched to SQS; waiting for an agent to pick it up
RunningAgent is actively executing the task
Awaiting approvalSupervised mode — agent has reached a checkpoint and is waiting for human approval before continuing
CompletedAgent finished successfully; artefacts and PR link available
DoneRun reviewed and closed by a human
FailedAgent encountered an unrecoverable error
AbandonedA supervised-mode checkpoint was cancelled by a human
PR rejectedThe pull request raised by the agent was closed without merging
Dead letterThe SQS message was moved to the dead-letter queue after repeated failures

Run Details

Click any run row to expand its details panel (or open it in a log viewer modal):

  • Status badge — current lifecycle status with colour coding
  • Phase — the agent’s last reported phase: STARTED → PLANNING → EXECUTING → COMMITTING → PR_RAISED → COMPLETED / FAILED
  • Duration — time from start to completion or failure
  • Cost — AI token usage (input/output tokens) and estimated Fargate compute cost in USD
  • Started / Completed timestamps

Log Viewer

Click View logs (the scroll icon) on any run to open a real-time log panel. This shows the chronological CloudWatch Logs stream for the run:

  • Plan summaries and reasoning steps
  • Files modified
  • Test results
  • Error messages and stack traces

Logs are paginated; click Load more to fetch older entries. The panel can be closed with the ✕ button.

Supervised Mode

When an item is dispatched with supervised mode, the agent pauses at key checkpoints and waits for human approval before continuing:

  1. The run status changes to Awaiting approval
  2. A notification (or email, if configured) is sent to the project Admin
  3. The run details panel shows Approve and Cancel buttons
  4. Click Approve to allow the agent to continue from where it paused
  5. Click Cancel to stop the run — the run status moves to Abandoned

Supervised mode is recommended when you are building trust with AI Pods on a codebase and want to review the agent’s plan before it writes any code.

Autopilot Mode

In autopilot mode, the agent runs end-to-end without human intervention. It plans, implements, commits, and opens a pull request automatically. The run status moves from Queued → Running → Completed in one uninterrupted sequence.

Dispatching a Run

Runs are created from the backlog item detail page. Open a Ready item and click Dispatch to Pod. Choose the pod and the mode (autopilot or supervised), then confirm. The run is created immediately and its status updates in real time.

Filtering Runs

Use the status filter chips at the top of the runs list to narrow the view to specific statuses (e.g. show only Awaiting approval to find items needing your review, or Failed to investigate errors).