Gemini CLI repairs a failed Cloud Run deployment

Richard Seroter used Gemini CLI to build and deploy a family vacation calendar application. The deployment request specified Cloud Run, a region and a Google Cloud project, and asked the agent to deploy source files rather than make a container manually. He had connected a Cloud Run MCP server, which supplied a structured deployment tool.

The initial deployment completed, but the service did not start correctly. Gemini inspected the service logs and identified a missing reference to the listening port. It changed the application and redeployed. This is the concrete agent behaviour in the account: the task continued from tool execution into observing a failure and correcting it.

Seroter kept command approvals enabled during his experiments. In a later CI/CD trial, he rejected an incorrect deployment request, supplied current Cloud Deploy documentation and asked to inspect the corrected files before proceeding. That later workflow also needed several failed-build iterations. The account illustrates why an operator should provide an explicit target and retain approval of infrastructure changes. It does not establish that every generated configuration is correct or that the calendar was production hardened; the reported result is a repaired deployment in his own experiment.

Prompt
Original prompt

Source: Blog · Richard Seroter