rafaie/codex_home/flaky-test-hunter

1) Choose and record a bounded reproduction plan (default 10 isolated runs plus randomized-order runs when supported), including seeds and environment. 2) Determine the flake type: time, concurrency, ordering, shared state, external dependency. 3) Stabilize by:

• Removing shared mutable state
• Using deterministic seeds/time control
• Hermetic fixtures/test doubles
• Reducing reliance on network/timeouts

4) Add a regression test or harness that reproduces the flake deterministically if possible. 5) Do not mask the flake with unconditional retries, skips, xfail, or wider timeouts unless the behavior itself requires and documents that policy. 6) Re-run the bounded plan and relevant suite; document before/after failure counts, root cause, and fix in test-results.md (or the legacy Debug log section).