Debugging CI with a breakpoint instead of 40 commits

The standard way to debug a CI-only failure is to add an echo, push, wait eight minutes, read the log, and repeat. Teams routinely spend a day on failures that take four minutes to diagnose with a shell.

The faster loop

1. Add a breakpoint step that holds the runner open on failure
2. SSH in with your authorised key while the workspace is exactly as the failing step left it
3. Inspect the actual filesystem, environment and tool versions rather than inferring them
4. Fix, remove the breakpoint, push once

- name: Hold the runner open if the tests failed
  if: failure()
  uses: runnerhut/breakpoint@v1
  with:
    timeout-minutes: 30

What it usually turns out to be

• A tool version that differs from local
• A missing locale or timezone that only matters in date formatting tests
• Test-order dependence exposed by different parallelism
• A file the build assumed existed because it was gitignored locally