What problem does it solve? A .NET test that fails in the full suite but passes alone is usually a parallelism flake, not a regression — but without a systematic process, developers waste time blaming their branch or chasing the wrong root cause. This Skill provides a classification ladder that separates real bugs from intra-test races, inter-test contention, and environmental flakes before any fix is attempted. ## Core Features & Use Cases - Classification ladder: Reproduce in the full suite, re-run in isolation, then run a clean-main baseline in a throwaway git worktree to determine whether the flake pre-exists your branch. - Root-cause classes with fixes: Class A intra-test races (lock-guard fake collections instead of serializing production concurrency), Class B inter-test contention (xunit v3 DisableParallelization collections for wall-clock I/O tests), and Class C environmental flakes (hermetic child PATH/env, cold-worktree asset provisioning). - Gate discipline: Rules for per-PR gates with known-flake tolerance, one-PR-per-flake-fix scope, and handling concurrent-session half-fixes that break origin/main. - Use Case: A watch-integration test fails once in three full-suite runs but passes every isolated re-run; the ladder identifies it as Class B inter-test contention, and the fix serializes the test class with a documented xunit collection definition. ## Quick Start Diagnose why my .NET test passes alone but fails in the full test suite and tell me how to fix it.
