Troubleshooting Staffless Workflows
Diagnose failed runs, missing output, integration issues, and high-cost behavior with a practical incident checklist.
Start With Output Events
The Output stream is the fastest way to pinpoint failure stage: prompt interpretation, tool call, external API, or delivery layer.
Always capture run timestamp, workspace, skill, and model before making changes.
Most Common Failure Patterns
Repeated tool loops usually indicate unclear skill constraints or too broad task goals.
Silent completion issues often come from delivery channel mismatch or missing destination identifiers.
- Tool call iteration limit reached.
- Unknown or disabled delivery channel.
- Credential expired or missing scope.
- Model mismatch between chat and cron task.
Stabilize Before Scaling
Once a task works reliably three times in a row, then schedule it. Production reliability comes from controlled scope, explicit outputs, and repeatable run settings.
Treat every high-value automation as a small system with monitoring and rollback options.
FAQ
What should I do if costs spike unexpectedly?
Check model selection on scheduled tasks, inspect run count, and review tool loop behavior. Cost spikes are often caused by repeated retries or broad prompts.
How do I reduce run failures quickly?
Narrow task scope, pin skill and model, validate integrations, and force a manual test run before re-enabling recurrence.