Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Strategy: decompose a complex research question into a hierarchy of independently answerable sub-questions
.claude/skills/yogsoth-ai-decomposition-formulation/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 29% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 73% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 115% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 58% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 109% | 0% |
Decompose a complex research question into independently answerable sub-questions — when a single experiment cannot answer the whole question.
Core logic: a complex problem = a combination of several simpler problems. Find the right decomposition such that each sub-question can be answered independently, and the answers to the sub-questions combine to answer the main question.
| Tier | Sub-questions | Dependency analysis | Sequence design | |------|---------|---------|---------| | S | ≥2 sub-questions | basic dependency identification | suggested order | | M | ≥3 sub-questions | dependency graph + critical path | optimal sequence + parallelization opportunities | | L | ≥5 sub-questions | full dependency graph + cycle detection | multi-path plan + resource allocation |
After the Strategy completes, you must call context-checkpoint and record:
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| Tactic | When to use | | --- | --- | | sub-question-decomposition | Tactic: Decompose a main question into a hierarchy of independently answerable sub-questions |
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | finer-criteria-check | SOP: check research-question quality against each of the 5 FINER criteria |
<!-- 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→fail | 17,389 | 17,892 | +3% | 1 | 1 | 0% | 3,032 | 3,854 | +27% | 0 | 0 | — |
case-02 | fail→fail | 17,044 | 19,839 | +16% | 1 | 1 | 0% | 2,876 | 3,983 | +38% | 0 | 0 | — |
case-03 | fail→fail | 19,824 | 23,045 | +16% | 1 | 1 | 0% | 3,315 | 4,567 | +38% | 0 | 0 | — |
case-04 | fail→pass | 13,885 | 14,886 | +7% | 1 | 1 | 0% | 2,359 | 3,044 | +29% | 0 | 0 | — |
case-05 | fail→pass | 13,422 | 17,658 | +32% | 1 | 1 | 0% | 2,068 | 3,586 | +73% | 0 | 0 | — |
case-06 | fail→pass | 10,864 | 19,117 | +76% | 1 | 1 | 0% | 1,806 | 3,889 | +115% | 0 | 0 | — |
case-07 | pass→pass | 22,185 | 17,462 | -21% | 1 | 1 | 0% | 2,136 | 3,505 | +64% | 0 | 0 | — |
case-08 | pass→pass | 12,198 | 15,024 | +23% | 1 | 1 | 0% | 1,919 | 3,179 | +66% | 0 | 0 | — |
case-09 | pass→pass | 17,352 | 14,106 | -19% | 1 | 1 | 0% | 2,747 | 3,032 | +10% | 0 | 0 | — |
case-10 | pass→pass | 15,536 | 12,251 | -21% | 1 | 1 | 0% | 2,571 | 2,795 | +9% | 0 | 0 | — |
case-11 | fail→pass | 10,946 | 14,785 | +35% | 1 | 1 | 0% | 1,896 | 3,001 | +58% | 0 | 0 | — |
case-12 | fail→pass | 11,335 | 17,503 | +54% | 1 | 1 | 0% | 1,720 | 3,588 | +109% | 0 | 0 | — |
case-13 | pass→pass | 9,059 | 10,561 | +17% | 1 | 1 | 0% | 1,404 | 2,520 | +79% | 0 | 0 | — |
case-14 | pass→pass | 9,900 | 14,292 | +44% | 1 | 1 | 0% | 1,606 | 3,023 | +88% | 0 | 0 | — |
case-15 | pass→pass | 13,320 | 12,603 | -5% | 1 | 1 | 0% | 2,130 | 2,581 | +21% | 0 | 0 | — |
case-16 | pass→pass | 11,842 | 11,103 | -6% | 1 | 1 | 0% | 1,936 | 2,406 | +24% | 0 | 0 | — |
case-17 | pass→pass | 20,036 | 16,097 | -20% | 1 | 1 | 0% | 2,011 | 3,237 | +61% | 0 | 0 | — |
case-18 | fail→pass | 12,714 | 21,347 | +68% | 1 | 1 | 0% | 2,008 | 4,252 | +112% | 0 | 0 | — |
case-19 | pass→fail | 22,745 | 32,801 | +44% | 1 | 1 | 0% | 3,207 | 5,105 | +59% | 0 | 0 | — |
case-20 | pass→pass | 5,198 | 6,702 | +29% | 1 | 1 | 0% | 815 | 1,952 | +140% | 0 | 0 | — |
case-21 | fail→pass | 5,551 | 12,901 | +132% | 1 | 1 | 0% | 1,071 | 3,236 | +202% | 0 | 0 | — |
case-22 | fail→fail | 4,446 | 12,740 | +187% | 1 | 1 | 0% | 656 | 2,668 | +307% | 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 +27 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.