Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Tactic: Decompose a main question into a hierarchy of independently answerable sub-questions
.claude/skills/yogsoth-ai-sub-question-decomposition/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-21 | ✗→✓ | ▲ Improved | -8% | 0% |
| case-22 | ✗→✓ | ▲ Improved | 18% | 0% |
| case-01 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 2% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 2% | 0% |
Decompose a main question into a hierarchy of independently answerable sub-questions — when the question is too complex to answer in one pass.
Structure "decomposing a complex question" as: generate sub-questions → map dependencies → design answering order. The three steps run sequentially, each step's output being the next step's input.
| SOP | Responsibility | When to call | |-----|------|---------| | sub-question-generation | Decompose the main question into sub-questions | First step, mandatory | | dependency-mapping | Map dependency relationships among sub-questions | After sub-questions are generated | | answering-sequence-design | Design the optimal answering order | After the dependency graph is complete |
Serial mode (the only mode):
Each step strictly depends on the previous step's output; no parallelization.
After execution, report:
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | answering-sequence-design | SOP: Design the optimal answering order for sub-questions | | dependency-mapping | SOP: map dependency relationships among sub-questions | | sub-question-generation | SOP: Decompose a main research question into independently answerable sub-questions |
<!-- 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-21 | fail→pass | 27,112 | 15,726 | -42% | 1 | 1 | 0% | 3,505 | 3,218 | -8% | 0 | 0 | — |
case-22 | fail→pass | 27,604 | 16,526 | -40% | 1 | 1 | 0% | 3,291 | 3,870 | +18% | 0 | 0 | — |
case-01 | fail→pass | 21,325 | 17,319 | -19% | 1 | 1 | 0% | 2,679 | 2,650 | -1% | 0 | 0 | — |
case-02 | fail→pass | 26,212 | 24,266 | -7% | 1 | 1 | 0% | 3,506 | 3,583 | +2% | 0 | 0 | — |
case-03 | fail→pass | 19,573 | 26,696 | +36% | 1 | 1 | 0% | 3,470 | 3,527 | +2% | 0 | 0 | — |
case-04 | pass→fail | 27,375 | 60,773 | +122% | 1 | 1 | 0% | 3,607 | 8,395 | +133% | 0 | 0 | — |
case-05 | pass→pass | 14,142 | 9,967 | -30% | 1 | 1 | 0% | 1,487 | 2,096 | +41% | 0 | 0 | — |
case-06 | pass→pass | 13,559 | 18,252 | +35% | 1 | 1 | 0% | 1,180 | 2,619 | +122% | 0 | 0 | — |
case-07 | fail→pass | 11,924 | 19,204 | +61% | 1 | 1 | 0% | 1,105 | 3,020 | +173% | 0 | 0 | — |
case-08 | fail→fail | 22,396 | 15,474 | -31% | 1 | 1 | 0% | 2,700 | 2,613 | -3% | 0 | 0 | — |
case-09 | fail→fail | 24,182 | 21,294 | -12% | 1 | 1 | 0% | 3,205 | 3,042 | -5% | 0 | 0 | — |
case-10 | fail→pass | 3,721 | 17,840 | +379% | 1 | 1 | 0% | 591 | 2,624 | +344% | 0 | 0 | — |
case-11 | fail→fail | 15,230 | 16,956 | +11% | 1 | 1 | 0% | 2,535 | 3,395 | +34% | 0 | 0 | — |
case-12 | fail→pass | 23,521 | 22,249 | -5% | 1 | 1 | 0% | 4,451 | 3,442 | -23% | 0 | 0 | — |
case-13 | fail→pass | 50,324 | 17,101 | -66% | 1 | 1 | 0% | 4,356 | 3,334 | -23% | 0 | 0 | — |
case-14 | fail→pass | 30,455 | 22,665 | -26% | 1 | 1 | 0% | 4,237 | 3,824 | -10% | 0 | 0 | — |
case-15 | fail→pass | 36,433 | 19,154 | -47% | 1 | 1 | 0% | 5,788 | 3,616 | -38% | 0 | 0 | — |
case-16 | fail→pass | 30,668 | 28,822 | -6% | 1 | 1 | 0% | 4,098 | 4,787 | +17% | 0 | 0 | — |
case-17 | fail→pass | 51,548 | 45,356 | -12% | 1 | 1 | 0% | 7,778 | 4,481 | -42% | 0 | 0 | — |
case-18 | fail→pass | 22,107 | 19,781 | -11% | 1 | 1 | 0% | 3,500 | 3,801 | +9% | 0 | 0 | — |
case-19 | fail→pass | 26,066 | 22,859 | -12% | 1 | 1 | 0% | 3,435 | 3,391 | -1% | 0 | 0 | — |
case-20 | fail→pass | 46,734 | 28,523 | -39% | 1 | 1 | 0% | 6,487 | 4,184 | -36% | 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 +68 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.