Troubleshooting · By Rivtor · Updated
Debug failed deployments from Cursor using structured errors
When a deployment triggered from Cursor fails, identify whether the failure belongs to the agent connection, source upload, build, migration or release check. Use the reported stage and cause to make a targeted fix, then verify the deployed application independently.
See how Rivtor connects to your agent →
Locate the failure stage
Cursor is the client initiating the workflow; it is not necessarily the process building or hosting the app. Start with the tool result, selected project and source input. A missing local upload and a failed production build require different fixes.
| Stage | Evidence to inspect |
|---|---|
| Connection | Enabled server, sign-in state and available tools |
| Source | Correct folder or repository revision |
| Build | First failing command, dependency and file |
| Migration | Database target, revision and schema error |
| Release | Listener, health path and application startup |
Rivtor’s error reference explains its structured failures. Use the actual code and stage instead of asking the agent to make broad speculative changes.
Give the agent a bounded investigation
An example prompt is: “Read the failed stage and identify the smallest relevant configuration or source change. Explain the target project, required variable names and verification step before retrying. Do not print secrets.” This is a review pattern, not an automated guarantee.
Check that a proposed dependency fix belongs in the failing service’s manifest. A package installed only in the local workspace may still be absent from the service image. A production port error is not normally fixed by changing unrelated frontend code.
Separate configuration from feature changes
If the failure names a missing secret, supply it through the supported secret workflow. Do not hardcode a value into source or weaken application authorization. If a migration fails, preserve its evidence and inspect schema state before retrying; deleting revision history can make recovery harder.
Use Cursor setup to confirm local versus remote source handling. The open editor folder and the platform’s source input must refer to the same intended changes.
Verify after the retry
Read the deployment result, visit the intended hostname and execute one synthetic user operation. Confirm the desired source change appears and that existing data persists. A successful agent response or a live URL is not sufficient to establish a complete application works. Keep skipped checks visible when writing release notes or a support reproduction.