Errors — diagnosing failed runs
Errors gathers every run that didn't finish, with the trace and context you need to find the cause and get your agent working again.
Last updated July 14, 2026
Even well-built agents hit snags — a connected tool times out, a trigger sends unexpected data, or an instruction turns out to be ambiguous. Errors is where those failures surface, so nothing fails silently.
It's a focused view of just the runs that didn't finish, pulled out of the broader run history so you can spot and fix problems fast.
What you'll find here
Each entry in Errors represents a run that stopped before completing. For every one, you can see:
- When it happened — and how the run was started.
- Where it stopped — the step where the agent ran into trouble.
- What went wrong — the error detail and the surrounding context.
- The full trace — the same step-by-step record you'd see in Activity, up to the point of failure.
Common causes
Most failures fall into a few familiar buckets. Knowing the pattern usually points straight to the fix.
| Symptom | Likely cause | Where to fix it |
|---|---|---|
| A tool step fails | A disconnected integration or an expired connection | Reconnect the integration, then re-run |
| The agent lacks access | A required Tool, Skill, or Secret isn't configured | The agent's Tools, Skills, or Secrets |
| Wrong or empty result | Ambiguous instructions or missing Knowledge | The agent's instructions or Knowledge |
| A trigger run fails | Unexpected incoming data from the event | Review the trace to see what arrived |
How to diagnose a failure
Read the error detail and note which step stopped the run.
Look at the steps leading up to the failure. Often the real cause is an earlier step that returned something unexpected.
Reconnect a tool, tighten an instruction, add the missing Skill or Secret — then change one thing at a time so you know what fixed it.
Run the agent and confirm the new run completes in Activity.
Frequently asked
Does a failed run still count toward my usage?
An agent consumes credits for the work it actually performs, so a run that did some work before failing may use credits up to the point it stopped. Fixing the root cause prevents repeated failed attempts.
Will a failed automation retry on its own?
Errors surfaces the failure so you can address it. For runs started by a schedule or trigger, resolve the underlying cause and the next scheduled or triggered run will pick up cleanly.
My run failed but produced no error I understand — what now?
Open the full trace and read the last few steps. The step just before the failure usually shows the input or result that caused the problem, even when the final message is terse.