Rhdh Prow Trigger

Trigger an RHDH nightly ProwJob

Execute a job that is already configured in openshift/release. This skill never edits CI configuration: changing which jobs exist is /rhdh-prow-jobs for test entries and pools, and /rhdh-prow-release-branch for a release branch's whole job set.

Route

Load workflows/trigger-nightly.md. It covers listing jobs, mapping a natural-language request to a full job name, the override options, and execution.

Authentication

scripts/trigger_nightly_job.py uses a dedicated kubeconfig at ~/.config/openshift-ci/kubeconfig so it never disturbs the user's current cluster context. It consumes an existing oc session and never performs a login.

When oc is missing, the dedicated kubeconfig is absent, or the session has expired, stop and tell the user to run /setup-rhdh-skills openshift-ci. Setup owns login; this skill does not.

The public script hands scripts/gangway_adapter.py only a kubeconfig path and the request payload. That adapter alone retrieves the transient credential and authenticates the request, and returns credential-free response data. Keep tokens out of arguments, output, and anything reported back.

Execution rules

Triggering a job is an external write; --dry-run and --list are not. Follow /mutation-gate, with the full job name as the target.

• Preview with --dry-run first. It prints the adapter request without executing.
• The preview carries the full command, the parameters, what the run will cost and touch, and how to abort it. Get explicit approval before running without --dry-run.
• GKE and OSD-GCP each share one cluster. Never start a second job on the same platform while one is running; warn the user before triggering either.
• Approval to inspect or dry-run is not approval to execute. Ask again.
• --image-repo requires --tag; --tag works alone. --playwright-version is overlay-only. Image, chart, and alert overrides are rejected for overlay jobs.

Completion

A trigger is complete when the executed command is shown exactly as run, the Gangway API response is reported, and the run URL or ID is stated prominently — or, when the API returned no identifier, that absence is stated rather than implied. A dry run is complete when the printed request is shown and it is clear that nothing was executed. A failure is complete when the error is reported with its likely cause, expired authentication being the common one, and the user is pointed at /setup-rhdh-skills openshift-ci when the session is the problem. No credential appears in any of it.