Troubleshooting Agent Work
Start with the affected agent's timeline and the currently selected workspace. Record what you expected, what happened, and when. Resolve the specific blocker before submitting the same work again.
| Symptom | Check | Next step |
|---|---|---|
| An agent or control is missing | Workspace, Team, accepted invitation, role, and feature availability | Follow Workspaces And Navigation; ask an administrator to verify access |
| Message accepted, no work begins | Active state, usage, pending approvals, account availability | Resolve the displayed blocker; installation operators should check the interactive worker |
| Agent repeatedly asks for access | App assignment, credential scope, provider grant, requested permission | Reconnect through the protected app or secrets flow and verify a small read |
| File uploaded but unusable | Pending, rejected, or expired scanning state | Wait for a clean result or follow Files And Workspaces; do not bypass quarantine |
| Email is waiting rather than sent | Outbox review state, recipient permissions, stale version, expiry | Review the exact message in Outbox |
| Email is accepted but missing | Connected mailbox or Organization sender, spam/quarantine, provider delivery record | Verify receipt or provider trace before resending |
| Reminder did not run | Next run, zone, enabled schedule, agent activity, limits | Inspect Schedules And Reminders and scheduler health |
| Some broadcast recipients did not respond | Delivery summary, paused agents, changed access | Follow up only after distinguishing delivery from completed work |
| Tool failed after a settings change | Published revision, agent binding, credential, approval validity | Rebind or request a fresh approval where required |
| Agent repeats work or consumes unexpected usage | Duplicate prompts, multiple schedules, peer loops, repeated provider events | Stop current work, identify the trigger, and adjust the specific workflow |
| A retired agent is needed again | Owning workspace and archive state | Use the authorized restore path; retention may have removed old content |
Reconnect an app
Open the connection in Integrations or the agent's app settings. Confirm the intended provider account and workspace before reconnecting. Grant the scopes the workflow needs, then verify the agent's assignment and try a read against a known accessible record.
A provider login succeeding does not prove access to every file, channel, mailbox, or record. Check provider-side sharing and any per-agent subscriptions or file selections. Do not paste a replacement access token into chat.
After a timeout
A timeout can occur after a service accepted a write. Inspect the platform timeline, Outbox or approval status, and the destination service before repeating a send, record creation, or other change. Retrying an uncertain operation can duplicate its effect.
If chat lost its live connection, reload and inspect saved messages before resubmitting. Closing a browser tab does not stop background work. Use Stop or pause for that purpose.
Prepare a useful support report
Include the deployment, workspace or Team, agent name and ID, approximate time with zone, expected result, actual result, safe error text, and the relevant timeline event or job ID. State whether the operation reads or changes an external system and whether you already retried it.
Exclude passwords, API keys, recovery codes, webhook tokens, signed download URLs, and unrelated private content. On a customer-run deployment, include the affected service and revision plus redacted logs.