happier-ci-stabilize

Run the collect-actions-failures.mjs script with the repository, run ID, and optional attempt number. It downloads complete logs for every failed, cancelled, or timed-out job and writes a summary.json with failed steps and error excerpts under /tmp.

Dispatch the nightly workflow with resume_run_id set to the completed origin run ID, but only when product source, packaging, and candidate bytes are unchanged. The resume path reuses individually verified immutable candidates while rerunning downstream validation gates.

Use a native failed-job rerun for transient runner, network, or external service failures on the same SHA. Start a fresh run when product source, build scripts, dependencies, or signing inputs changed, since old artifacts no longer prove the new code.

Yes, the failure collection script invokes the gh api command to fetch run metadata, job lists, and job logs. You must have gh installed and authenticated with access to the target repository before running the collector.

A failed mutation such as a release PATCH or upload may have succeeded remotely, so blind retry can duplicate or overwrite publication. Reconcile the observed remote state first, and only retry operations proven idempotent.

Remove coverage only when evidence shows it duplicates the same observable contract, polices formatting or constants with no public contract, or duplicates a canonical testkit boundary. Strengthen the owner-level test first and preserve one discriminating test per real contract.