Workflow Audit scans a repository's GitHub Actions workflows with twelve small checkers: script injection, unpinned actions, missing timeouts and so on. Each runs as its own Python process, so a checker that crashes can't hide what the others found.
Timed one by one, every checker took about 30 ms, and nearly all of that was Python starting up. The work was tiny; the waiting was serial. Twelve start-ups in a row made each audit cost a third of a second, and the test file, which audits about two dozen fixtures, took 8.3 seconds.
The checkers don't depend on each other, so the fix is to start them all at once and collect the answers in the original order. Same processes, same isolation, same report: we compared the JSON output on this repository's own workflows before and after, and the files are byte-identical. A new test pins the parallel path against the old loop.
On a 4-core machine the test file now takes 2.5 seconds, and scanning this repository's workflows takes 0.25 s instead of 0.63 s. That second number is what anyone using the GitHub Action waits on.
The method and the ranked list of what to try next are in docs/history/ci-perf-2026-11-17-workflow-audit-parallel-checks.md.