Troubleshooting¶
Read the task's result and events using the same database or server as the original request:
landing action view act_example --json
landing action logs act_example
Replace the ID with the affected action. Start with the observed failure, then check its owning layer:
| Symptom | Next step |
|---|---|
| Exit code 2 | Check arguments, inputs, workspace, and configuration. |
| Remote connection failure or timeout | Read stderr; check the server URL, availability, and proxy. |
| Model failure or empty completion | Inspect the action error and saved model diagnostics; verify provider settings before retrying. |
| Fix validation failed | Read validation events and inspect the workspace diff to distinguish repair and environment failures. |
| Review returned nonzero | Read status and decision separately: completed block or inconclusive is a recommendation. |
| Another worker owns the database | Use --server for that worker or select a separate database. |
| Work continues after Ctrl-C | Remote waiting stops independently; use action cancel to stop the task. |
| GitHub task skipped | Check caller permission, trust policy, explicit command prefix, and upstream workflow source. |
| Review cancelled | Inspect the current PR head and github.review_stopped; an unverifiable head also stops tools. |
| Duty timed out after a push | Inspect the last tool and native timeout; finish with candidate and pending-check links rather than waiting for the current feedback job. |
| Required publication failed | Inspect gh authorization and the requested destination; reuse the same database and delivery key when retrying delivery. |
| Advice is wrong despite passing checks | Record the concrete mistake in the work item and improve guidance, tools, or code where it belongs. |
| Restored action is interrupted | Inspect its history and workspace before retrying. |
action retry creates a new task from the original request and does not undo edits. Completed tasks do not replay automatically. See CLI exit codes.
For reviews, distinguish the candidate, actual CI checkout, and deployed revision. For backup problems, inspect Litestream logs and the replica; readiness does not measure replication health.
When asking for help, include the smallest reproduction, observed impact, relevant revisions, and useful logs with credentials removed. Share missing evidence explicitly so another person can help resolve it.