# Copilot troubleshooting

Diagnose missing context, incomplete evidence, and failed Compute Copilot actions.

## Diagnostic order

Use the session context, cited evidence, and work request in that order. This separates a prompt problem from a data problem and an operation problem.

| Symptom               | Check                                 | Next action                                   |
| --------------------- | ------------------------------------- | --------------------------------------------- |
| No instance context   | Space, region, and selected instance  | Re-select the instance                        |
| Generic answer        | Time range and question scope         | Add a concrete time range and expected signal |
| Missing evidence      | Metrics or events availability        | Open instance metrics and events directly     |
| Action rejected       | Permission and instance state         | Resolve the dashboard error before retrying   |
| Request still running | Work request state and latest message | Wait, then refresh the timeline               |

## Capture diagnostics

1. Copy the session ID and selected instance ID.
2. Record the question, time range, and cited evidence.
3. Copy the work request ID for any confirmed action.
4. Capture the latest error message without secrets.
5. Re-run the smallest request that reproduces the issue.

## Escalation boundary

Do not repeat a state-changing action while the original work request is still running. Use the request ID and timeline entries when contacting the platform team.

## Related documentation

- [Compute product overview](/compute/)
- [Create an instance](/compute/instances/create/)
- [Instance details](/compute/instances/detail/)
- [Work requests](/compute/operations/work-requests/)