Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Socket command threading and focus policy for cmux CLI/socket work. Use when adding or changing socket commands, CLI commands, telemetry commands, focus/select/open/close/send-key behavior, or automation that could steal app focus.
.claude/skills/manaflow-ai-cmux-socket-policy/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -35% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -27% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -24% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -9% | 0% |
| case-03 | ✓→✗ | ▼ Worse | 22% | 0% |
DispatchQueue.main.sync for high-frequency socket telemetry commands such as report_*, ports_kick, status/progress updates, or log metadata updates.DispatchQueue.main.async only when needed.window.focus, workspace.select/next/previous/last, surface.focus, pane.focus/last, browser focus commands, and v1 focus equivalents.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 12,753 | 6,833 | -46% | 1 | 1 | 0% | 2,528 | 1,649 | -35% | 0 | 0 | — |
case-02 | pass→pass | 14,856 | 11,040 | -26% | 1 | 1 | 0% | 2,817 | 2,585 | -8% | 0 | 0 | — |
case-03 | pass→fail | 6,863 | 6,533 | -5% | 1 | 1 | 0% | 1,235 | 1,506 | +22% | 0 | 0 | — |
case-04 | pass→pass | 12,256 | 3,205 | -74% | 1 | 1 | 0% | 1,714 | 889 | -48% | 0 | 0 | — |
case-05 | pass→pass | 14,711 | 11,198 | -24% | 1 | 1 | 0% | 3,201 | 2,781 | -13% | 0 | 0 | — |
case-06 | pass→pass | 15,512 | 8,288 | -47% | 1 | 1 | 0% | 2,923 | 1,868 | -36% | 0 | 0 | — |
case-07 | pass→pass | 9,484 | 3,073 | -68% | 1 | 1 | 0% | 1,622 | 933 | -42% | 0 | 0 | — |
case-08 | fail→pass | 17,336 | 9,743 | -44% | 1 | 1 | 0% | 3,118 | 2,266 | -27% | 0 | 0 | — |
case-09 | pass→pass | 8,926 | 3,498 | -61% | 1 | 1 | 0% | 1,343 | 971 | -28% | 0 | 0 | — |
case-23 | fail→fail | 15,512 | 12,474 | -20% | 1 | 1 | 0% | 3,184 | 2,712 | -15% | 0 | 0 | — |
case-10 | pass→pass | 10,178 | 3,590 | -65% | 1 | 1 | 0% | 1,591 | 842 | -47% | 0 | 0 | — |
case-11 | fail→pass | 16,965 | 11,202 | -34% | 1 | 1 | 0% | 3,417 | 2,580 | -24% | 0 | 0 | — |
case-12 | pass→pass | 17,998 | 10,544 | -41% | 1 | 1 | 0% | 3,635 | 2,437 | -33% | 0 | 0 | — |
case-13 | pass→pass | 16,120 | 4,781 | -70% | 1 | 1 | 0% | 2,807 | 1,208 | -57% | 0 | 0 | — |
case-14 | pass→pass | 17,010 | 10,294 | -39% | 1 | 1 | 0% | 3,343 | 2,191 | -34% | 0 | 0 | — |
case-15 | pass→pass | 13,683 | 18,527 | +35% | 1 | 1 | 0% | 2,903 | 3,427 | +18% | 0 | 0 | — |
case-16 | fail→pass | 18,531 | 13,827 | -25% | 1 | 1 | 0% | 3,455 | 3,144 | -9% | 0 | 0 | — |
case-17 | pass→pass | 8,480 | 3,868 | -54% | 1 | 1 | 0% | 1,340 | 848 | -37% | 0 | 0 | — |
case-18 | pass→pass | 14,008 | 10,495 | -25% | 1 | 1 | 0% | 3,006 | 2,371 | -21% | 0 | 0 | — |
case-19 | pass→pass | 14,834 | 3,677 | -75% | 1 | 1 | 0% | 1,763 | 1,032 | -41% | 0 | 0 | — |
case-20 | pass→pass | 15,763 | 11,337 | -28% | 1 | 1 | 0% | 3,093 | 2,714 | -12% | 0 | 0 | — |
case-21 | pass→pass | 14,529 | 12,161 | -16% | 1 | 1 | 0% | 2,910 | 2,960 | +2% | 0 | 0 | — |
case-22 | fail→fail | 16,191 | 16,269 | +0% | 1 | 1 | 0% | 3,138 | 2,842 | -9% | 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. 23 cases were attempted. The headline lift of +13 percentage points is the difference between those two pass rates over the 23 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.