Terminal
culprit resume 77f60c9fccf8Every agent run and decision is saved to the ledger as it happens. resume picks up where the run
stopped and reuses everything already measured.
When you need it#
| The run ended with | Do this |
|---|---|
budget_exceeded |
Resume with a higher cap: culprit resume <run_id> --budget-usd 6 |
infra_error |
Fix the cause (key, credit, outage), then culprit resume <run_id> |
| You pressed Ctrl-C | culprit resume <run_id> |
Terminal
culprit resume 77f60c9fccf8 --budget-usd 6The run keeps its original settings, including the model, so every commit is still judged the same way.
localise, fix and audit --run-id resume the same way: run the same command again.
Options#
| Option | Default | Description |
|---|---|---|
--budget-usd X |
unchanged | Raise the run's cost cap. |
--workers N |
4 | Agent runs in parallel. |
--db PATH |
.culprit/ledger.sqlite |
The ledger holding the run. |
Exit code 0 when the run ends found or no_regression, 1 otherwise.