Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Shared behavior and mutation-path rules for cmux. Use when a behavior is exposed through multiple entrypoints such as keyboard shortcuts, command palette, context menu, CLI, settings, debug menu, optimistic UI, or tests that previously missed a bug.
.claude/skills/manaflow-ai-cmux-shared-behavior/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -5% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -41% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -32% | 0% |
| case-18 | ✗→✓ | ▲ Improved | -40% | 0% |
When a behavior is exposed through multiple surfaces (keyboard shortcut, command palette, context menu, CLI/socket command, settings UI, debug menu), implement one shared action/model path and verify every entrypoint that should invoke it. Do not patch one surface and leave the others with duplicated logic.
Keep one mutation path. Record pending state with a request id or a previous snapshot, reconcile from the authoritative result, and handle failure with an explicit rollback or error state. Never let each entrypoint maintain its own optimistic copy.
When a user says tests missed a bug, add or adjust behavior-level coverage around the exact repro path before claiming the fix is complete.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | pass→pass | 10,149 | 4,157 | -59% | 1 | 1 | 0% | 1,691 | 913 | -46% | 0 | 0 | — |
case-05 | pass→pass | 9,572 | 3,765 | -61% | 1 | 1 | 0% | 1,488 | 817 | -45% | 0 | 0 | — |
case-06 | pass→pass | 14,009 | 10,545 | -25% | 1 | 1 | 0% | 2,370 | 1,841 | -22% | 0 | 0 | — |
case-07 | fail→pass | 13,519 | 10,259 | -24% | 1 | 1 | 0% | 2,434 | 2,316 | -5% | 0 | 0 | — |
case-08 | pass→pass | 12,431 | 6,045 | -51% | 1 | 1 | 0% | 1,914 | 1,211 | -37% | 0 | 0 | — |
case-01 | pass→pass | 10,794 | 6,089 | -44% | 1 | 1 | 0% | 1,920 | 1,316 | -31% | 0 | 0 | — |
case-02 | pass→pass | 13,855 | 7,357 | -47% | 1 | 1 | 0% | 2,541 | 1,431 | -44% | 0 | 0 | — |
case-03 | pass→pass | 14,345 | 8,073 | -44% | 1 | 1 | 0% | 2,551 | 1,755 | -31% | 0 | 0 | — |
case-09 | fail→pass | 10,342 | 5,982 | -42% | 1 | 1 | 0% | 1,741 | 1,151 | -34% | 0 | 0 | — |
case-10 | pass→pass | 10,598 | 5,602 | -47% | 1 | 1 | 0% | 1,952 | 1,177 | -40% | 0 | 0 | — |
case-11 | pass→pass | 15,751 | 11,735 | -25% | 1 | 1 | 0% | 2,642 | 2,306 | -13% | 0 | 0 | — |
case-12 | pass→fail | 12,010 | 3,954 | -67% | 1 | 1 | 0% | 1,822 | 771 | -58% | 0 | 0 | — |
case-13 | pass→pass | 13,391 | 8,245 | -38% | 1 | 1 | 0% | 2,267 | 1,242 | -45% | 0 | 0 | — |
case-14 | fail→pass | 13,061 | 7,505 | -43% | 1 | 1 | 0% | 2,244 | 1,327 | -41% | 0 | 0 | — |
case-15 | pass→pass | 14,814 | 10,263 | -31% | 1 | 1 | 0% | 2,336 | 1,642 | -30% | 0 | 0 | — |
case-16 | pass→pass | 10,695 | 3,399 | -68% | 1 | 1 | 0% | 1,812 | 725 | -60% | 0 | 0 | — |
case-17 | fail→pass | 11,752 | 7,256 | -38% | 1 | 1 | 0% | 1,914 | 1,309 | -32% | 0 | 0 | — |
case-18 | fail→pass | 17,441 | 11,544 | -34% | 1 | 1 | 0% | 2,118 | 1,263 | -40% | 0 | 0 | — |
case-19 | pass→pass | 10,367 | 6,342 | -39% | 1 | 1 | 0% | 1,417 | 961 | -32% | 0 | 0 | — |
case-20 | pass→pass | 14,868 | 10,195 | -31% | 1 | 1 | 0% | 2,747 | 2,237 | -19% | 0 | 0 | — |
case-21 | pass→pass | 18,468 | 22,425 | +21% | 1 | 1 | 0% | 3,260 | 2,873 | -12% | 0 | 0 | — |
case-22 | pass→pass | 21,422 | 14,306 | -33% | 1 | 1 | 0% | 2,834 | 2,805 | -1% | 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 +18 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is 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.