Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Manage MSBuild Change Waves: create new waves, condition features behind opt-out flags, write tests for wave-gated features, document change waves in ChangeWaves.md, and retire expired waves. Use when adding changes that need an opt-out or rotating out old change waves. Changes that introduce a user-visible behavior change should consider whether to use a changewave.
.claude/skills/dotnet-changewaves/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -32% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 0% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -4% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -62% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -36% | 0% |
A Change Wave is an opt-out flag that groups risky features together. Users disable features by setting the environment variable MSBUILDDISABLEFEATURESFROMVERSION to the wave version. This skill covers the how — the full lifecycle: creating a wave, conditioning code on it, testing, documenting, and retiring.
For the when — deciding whether a change is a breaking change and whether it needs a ChangeWave at all — see assessing-breaking-changes.
Use a change wave when a change is valuable but has meaningful compatibility risk for existing builds.
Good candidates:
Usually avoid a change wave for:
If uncertain, default to safety:
Document the decision in PR notes so reviewers can validate the call.
If a changewave is appropriate, consult ChangeWaves-Dev.md (developer-facing) and ChangeWaves.md (public-facing) for details on how to follow the below steps.
Look up the current MSBuild version. If a wave already exists for that version in src/Framework/ChangeWaves.cs, use it. Otherwise, create a new one.
C# and MSBuild examples are in ChangeWaves-Dev.md. Prefer to put checks inline rather than abstracting out a method to host the check.
Write normal tests for the new behavior. Then add at least one test that verifies the old behavior is preserved when the wave is opted out.
Add an entry to the Current Rotation of Change Waves section in ChangeWaves.md.
### {Major}.{Minor} heading at the top of the current rotation list.Each entry is a bullet with a link to the PR:
markdown### 18.6 - [Short description of the change.](https://github.com/dotnet/msbuild/pull/NNNNN)
Waves rotate out when a new major .NET version will be released. When retiring a wave, follow the detailed description in ChangeWaves-Dev.md.
eng/Versions.propsAllWaves entry added in ChangeWaves.cs (if new wave)AreFeaturesEnableddocumentation/wiki/ChangeWaves.md| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 18,978 | 12,253 | -35% | 1 | 1 | 0% | 3,201 | 2,167 | -32% | 0 | 0 | — |
case-02 | fail→pass | 16,288 | 9,679 | -41% | 1 | 1 | 0% | 2,514 | 2,502 | -0% | 0 | 0 | — |
case-03 | fail→pass | 15,246 | 8,684 | -43% | 1 | 1 | 0% | 2,384 | 2,297 | -4% | 0 | 0 | — |
case-04 | fail→fail | 20,886 | 14,795 | -29% | 1 | 1 | 0% | 2,921 | 2,693 | -8% | 0 | 0 | — |
case-05 | pass→pass | 13,799 | 15,056 | +9% | 1 | 1 | 0% | 2,486 | 3,550 | +43% | 0 | 0 | — |
case-06 | pass→fail | 9,943 | 5,829 | -41% | 1 | 1 | 0% | 1,713 | 1,765 | +3% | 0 | 0 | — |
case-07 | fail→pass | 20,243 | 3,105 | -85% | 1 | 1 | 0% | 3,380 | 1,293 | -62% | 0 | 0 | — |
case-08 | fail→pass | 12,098 | 3,316 | -73% | 1 | 1 | 0% | 1,910 | 1,215 | -36% | 0 | 0 | — |
case-09 | pass→pass | 7,058 | 2,371 | -66% | 1 | 1 | 0% | 1,279 | 1,222 | -4% | 0 | 0 | — |
case-20 | pass→pass | 11,734 | 3,869 | -67% | 1 | 1 | 0% | 1,693 | 1,510 | -11% | 0 | 0 | — |
case-10 | fail→pass | 13,001 | 3,710 | -71% | 1 | 1 | 0% | 2,045 | 1,557 | -24% | 0 | 0 | — |
case-11 | pass→pass | 11,111 | 4,375 | -61% | 1 | 1 | 0% | 1,594 | 1,592 | -0% | 0 | 0 | — |
case-12 | pass→pass | 11,863 | 2,942 | -75% | 1 | 1 | 0% | 1,695 | 1,303 | -23% | 0 | 0 | — |
case-13 | pass→pass | 14,711 | 5,614 | -62% | 1 | 1 | 0% | 2,279 | 1,670 | -27% | 0 | 0 | — |
case-14 | pass→pass | 12,895 | 6,285 | -51% | 1 | 1 | 0% | 1,924 | 1,917 | -0% | 0 | 0 | — |
case-15 | fail→pass | 18,953 | 2,308 | -88% | 1 | 1 | 0% | 3,013 | 1,210 | -60% | 0 | 0 | — |
case-16 | fail→pass | 8,578 | 2,723 | -68% | 1 | 1 | 0% | 1,404 | 1,203 | -14% | 0 | 0 | — |
case-17 | pass→pass | 12,725 | 4,233 | -67% | 1 | 1 | 0% | 1,662 | 1,331 | -20% | 0 | 0 | — |
case-18 | pass→pass | 9,419 | 3,088 | -67% | 1 | 1 | 0% | 1,592 | 1,370 | -14% | 0 | 0 | — |
case-19 | pass→pass | 6,161 | 2,170 | -65% | 1 | 1 | 0% | 1,123 | 1,199 | +7% | 0 | 0 | — |
case-21 | fail→pass | 17,814 | 4,041 | -77% | 1 | 1 | 0% | 3,038 | 1,439 | -53% | 0 | 0 | — |
case-22 | pass→pass | 10,341 | 3,474 | -66% | 1 | 1 | 0% | 1,855 | 1,374 | -26% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +36 percentage points is the difference between those two pass rates over the 22 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.