Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Reframe research questions using abstraction laddering, HMW formulation, and Socratic probing. Find the most productive level and framing for investigation.
.claude/skills/yogsoth-ai-question-reformulation/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-15 | ✗→✓ | ▲ Improved | 173% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 247% | 0% |
| case-08 | ✓→✗ | ▼ Worse | -5% | 0% |
| case-09 | ✓→✗ | ▼ Worse | -21% | 0% |
| case-13 | ✓→✗ | ▼ Worse | -25% | 0% |
Reframe research questions to find the most productive framing.
The current question framing may be limiting — need to explore alternative framings at different abstraction levels.
| Base SOP | Target | ±10% Range | |----------|--------|------------| | web-search | 20 | 18–22 | | web-research | 5 | 4–6 | | paper-overview | 20 | 18–22 | | paper-search | 15 | 13–17 | | paper-research | 5 | 4–6 |
<HARD-GATE>
| SOP | Done | Target | % |
|-----|------|--------|---|
| web-search | ? | 20 | ? |
| web-research | ? | 5 | ? |
| paper-overview | ? | 20 | ? |
| paper-search | ? | 15 | ? |
| paper-research | ? | 5 | ? |
Budget Gate: OPEN/CLOSED (>=80% required to exit)
</HARD-GATE>Import: web-search, web-research, paper-overview, paper-search, paper-research Subagent: abstraction-laddering, hmw-formulation, socratic-probing
Use abstraction laddering to move between concrete and abstract framings. Generate HMW questions from tensions/insights. Apply Socratic probing to test question quality.
Reformulated Questions — abstraction ladder, HMW set, Socratic challenges, recommended framing with reasoning.
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | abstraction-laddering | Move between concrete and abstract framings — 3 levels up (Why?) and 3 levels down (How?) to find the most productive research level. | | hmw-formulation | Generate "How Might We" questions at different scope levels (narrow, medium, broad). Ensures each is actionable without being prescriptive. | | socratic-probing | Apply 6 types of Socratic questions to test claims and assumptions. Exposes weaknesses and strengthens reasoning. |
<!-- 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-15 | fail→pass | 20,537 | 37,166 | +81% | 1 | 1 | 0% | 2,322 | 6,348 | +173% | 0 | 0 | — |
case-01 | fail→fail | 17,822 | 56,995 | +220% | 1 | 1 | 0% | 2,896 | 8,802 | +204% | 0 | 0 | — |
case-02 | fail→fail | 27,332 | 40,715 | +49% | 1 | 1 | 0% | 3,145 | 6,392 | +103% | 0 | 0 | — |
case-03 | fail→fail | 20,335 | 3,780 | -81% | 1 | 1 | 0% | 3,108 | 1,160 | -63% | 0 | 0 | — |
case-14 | pass→pass | 30,101 | 63,071 | +110% | 1 | 1 | 0% | 3,448 | 8,758 | +154% | 0 | 0 | — |
case-04 | fail→fail | 19,482 | 21,111 | +8% | 1 | 1 | 0% | 2,278 | 1,858 | -18% | 0 | 0 | — |
case-05 | fail→fail | 18,429 | 16,379 | -11% | 1 | 1 | 0% | 2,855 | 1,462 | -49% | 0 | 0 | — |
case-06 | pass→pass | 14,284 | 24,937 | +75% | 1 | 1 | 0% | 1,563 | 4,612 | +195% | 0 | 0 | — |
case-07 | fail→pass | 16,774 | 41,405 | +147% | 1 | 1 | 0% | 1,823 | 6,323 | +247% | 0 | 0 | — |
case-08 | pass→fail | 17,220 | 26,598 | +54% | 1 | 1 | 0% | 1,911 | 1,822 | -5% | 0 | 0 | — |
case-09 | pass→fail | 21,268 | 29,077 | +37% | 1 | 1 | 0% | 2,241 | 1,778 | -21% | 0 | 0 | — |
case-10 | pass→pass | 20,514 | 80,506 | +292% | 1 | 1 | 0% | 2,370 | 8,762 | +270% | 0 | 0 | — |
case-11 | fail→fail | 20,493 | 29,502 | +44% | 1 | 1 | 0% | 2,089 | 2,007 | -4% | 0 | 0 | — |
case-12 | pass→pass | 23,877 | 63,589 | +166% | 1 | 1 | 0% | 2,699 | 8,749 | +224% | 0 | 0 | — |
case-13 | pass→fail | 19,610 | 64,232 | +228% | 1 | 1 | 0% | 2,132 | 1,605 | -25% | 0 | 0 | — |
case-16 | fail→fail | 12,985 | 30,892 | +138% | 1 | 1 | 0% | 1,238 | 5,736 | +363% | 0 | 0 | — |
case-17 | pass→pass | 18,459 | 51,149 | +177% | 1 | 1 | 0% | 2,212 | 6,157 | +178% | 0 | 0 | — |
case-18 | pass→fail | 20,758 | 20,590 | -1% | 1 | 1 | 0% | 2,253 | 1,546 | -31% | 0 | 0 | — |
case-19 | fail→fail | 36,011 | 21,237 | -41% | 1 | 1 | 0% | 4,784 | 1,239 | -74% | 0 | 0 | — |
case-20 | pass→fail | 21,252 | 18,434 | -13% | 1 | 1 | 0% | 2,435 | 1,384 | -43% | 0 | 0 | — |
case-21 | pass→fail | 16,085 | 36,062 | +124% | 1 | 1 | 0% | 2,173 | 7,773 | +258% | 0 | 0 | — |
case-22 | pass→fail | 8,161 | 63,039 | +672% | 1 | 1 | 0% | 1,496 | 8,767 | +486% | 0 | 0 | — |
case-23 | pass→fail | 15,597 | 50,665 | +225% | 1 | 1 | 0% | 2,052 | 8,762 | +327% | 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, and 14 counted toward the lift figure. The other 9 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of -26 percentage points is the difference between those two pass rates over the 14 comparable cases. 8 cases got worse with the skill loaded, and they are 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.