Troubleshooting
When a routine stops, the answer is almost always already written down. This is how to find it and get moving again.
Overview
What this covers
Failed steps, unavailable sources, retries, what happens after a restart, and how to re-run a routine without producing a second half-finished result.
How it works
How it works
Harness is built to fail visibly rather than quietly:
- Transient failures are retried. A source that is briefly unavailable does not end the run.
- Scheduled routines resume after a restart. You do not get a silently half-finished job.
- A failed step records where it stopped — the last successful step, what it was attempting and why it could not continue.
That record is the first place to look, before changing any settings.
Step by step
Step by step
- Open the run in the audit trail.
- Find the last successful step.
- Read the error recorded on the step after it.
- Fix the cause — a permission, a moved file, an expired key.
- Re-run the routine and watch that step pass.
- Confirm the result and the trail agree before scheduling it again.
Common problems
Common problems
- Source unavailable. A share was offline or a file locked. Re-run once the source is back.
- Provider outage or rate limit. The reasoning call failed — see AI providers & keys.
- A capability was revoked. Someone tidied up permissions. Re-grant the specific target the run names.
- The machine ran out of disk. Audit history and working files live on that PC — clear space and re-run.
- Nothing failed, but nothing finished. Check the approval queue; the run is waiting for a person.
Still stuck
Start again from a known-good setup
Re-reading the setup guide with the audit trail open resolves most stubborn cases.