Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Format task plan as bite-sized executable tasks following superpowers:writing-plans conventions
.claude/skills/yogsoth-ai-plan-formatting/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -42% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -48% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -49% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -59% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -61% | 0% |
把任务信息格式化为 bite-sized 可执行 plan 这件事，直接 Skill load superpowers:writing-plans，不再自造格式规则。writing-plans 自带 HARD-GATE （零 TBD/TODO/占位符）。本 SOP 仅作为"调用 writing-plans"的路由点；如需子代理 承载，仍经 spawn-agent。
<!-- BEGIN available-tables (generated) --> <!-- external rows hand-maintained; do not regenerate this file -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | spawn-agent | Spawn a customized CC subagent with full MCP tool access. Used by SOPs that declare execution: subagent. | | superpowers:writing-plans | Produce a bite-sized, TDD-structured implementation plan from the spec |
<!-- END available-tables (generated) -->
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 18,888 | 14,695 | -22% | 1 | 1 | 0% | 3,256 | 1,887 | -42% | 0 | 0 | — |
case-02 | fail→fail | 20,730 | 30,667 | +48% | 1 | 1 | 0% | 2,366 | 2,678 | +13% | 0 | 0 | — |
case-03 | fail→fail | 17,462 | 17,422 | -0% | 1 | 1 | 0% | 2,883 | 3,386 | +17% | 0 | 0 | — |
case-04 | pass→pass | 24,783 | 32,496 | +31% | 1 | 1 | 0% | 4,453 | 5,917 | +33% | 0 | 0 | — |
case-05 | pass→fail | 28,813 | 16,499 | -43% | 1 | 1 | 0% | 2,243 | 2,272 | +1% | 0 | 0 | — |
case-06 | pass→pass | 6,751 | 9,752 | +44% | 1 | 1 | 0% | 1,295 | 1,044 | -19% | 0 | 0 | — |
case-07 | fail→pass | 16,511 | 4,254 | -74% | 1 | 1 | 0% | 1,820 | 943 | -48% | 0 | 0 | — |
case-08 | fail→pass | 14,914 | 4,268 | -71% | 1 | 1 | 0% | 1,700 | 874 | -49% | 0 | 0 | — |
case-09 | fail→pass | 14,947 | 7,378 | -51% | 1 | 1 | 0% | 1,643 | 678 | -59% | 0 | 0 | — |
case-10 | fail→pass | 19,295 | 8,476 | -56% | 1 | 1 | 0% | 1,785 | 698 | -61% | 0 | 0 | — |
case-11 | fail→pass | 12,230 | 5,021 | -59% | 1 | 1 | 0% | 1,123 | 1,143 | +2% | 0 | 0 | — |
case-12 | fail→pass | 19,822 | 4,474 | -77% | 1 | 1 | 0% | 2,128 | 861 | -60% | 0 | 0 | — |
case-13 | fail→pass | 16,781 | 5,009 | -70% | 1 | 1 | 0% | 1,852 | 1,091 | -41% | 0 | 0 | — |
case-14 | pass→pass | 14,739 | 9,085 | -38% | 1 | 1 | 0% | 1,535 | 858 | -44% | 0 | 0 | — |
case-15 | fail→pass | 14,827 | 8,007 | -46% | 1 | 1 | 0% | 1,393 | 616 | -56% | 0 | 0 | — |
case-16 | fail→fail | 15,003 | 16,694 | +11% | 1 | 1 | 0% | 1,675 | 2,083 | +24% | 0 | 0 | — |
case-17 | fail→pass | 12,985 | 7,928 | -39% | 1 | 1 | 0% | 1,258 | 854 | -32% | 0 | 0 | — |
case-18 | pass→pass | 13,679 | 3,224 | -76% | 1 | 1 | 0% | 1,349 | 775 | -43% | 0 | 0 | — |
case-19 | pass→pass | 4,875 | 11,540 | +137% | 1 | 1 | 0% | 672 | 1,190 | +77% | 0 | 0 | — |
case-20 | fail→pass | 14,837 | 11,636 | -22% | 1 | 1 | 0% | 1,586 | 1,288 | -19% | 0 | 0 | — |
case-21 | fail→pass | 15,425 | 4,903 | -68% | 1 | 1 | 0% | 1,463 | 942 | -36% | 0 | 0 | — |
case-22 | fail→pass | 10,035 | 8,475 | -16% | 1 | 1 | 0% | 1,416 | 715 | -50% | 0 | 0 | — |
case-23 | pass→pass | 12,975 | 8,510 | -34% | 1 | 1 | 0% | 1,465 | 660 | -55% | 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 +52 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.