Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Tactic: iteratively refine a research question until it passes all 5 FINER criteria
.claude/skills/yogsoth-ai-question-refinement-loop/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 75% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -41% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -19% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -7% | 0% |
Iteratively refine a research question — polish it repeatedly until it passes all 5 FINER criteria.
Structure "refining the question" as a loop: check → identify problems → revise → re-check. Each round focuses on the criteria that did not pass.
| SOP | Responsibility | When to call | |-----|------|---------| | finer-criteria-check | Check each of the 5 FINER criteria item by item | The check step of each loop | | scope-assessment | Assess whether the question's scope is appropriate | When F (Feasible) fails | | success-criteria-definition | Define measurable success criteria | After the loop ends, to confirm the RQ is measurable |
Standard loop:
Max iterations: 3 rounds. If it still fails after 3 rounds → report which criterion persistently fails + suggest a fundamental change of direction.
After execution, report:
<!-- BEGIN available-tables (generated) -->
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 | | scope-assessment | SOP: Assess whether a research question has appropriate scope (too broad / appropriate / too narrow) | | success-criteria-definition | SOP: Define measurable success criteria for a research question |
<!-- 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 | 30,888 | 31,953 | +3% | 1 | 1 | 0% | 4,767 | 5,392 | +13% | 0 | 0 | — |
case-02 | pass→pass | 17,537 | 27,751 | +58% | 1 | 1 | 0% | 2,491 | 5,066 | +103% | 0 | 0 | — |
case-03 | pass→pass | 13,379 | 26,103 | +95% | 1 | 1 | 0% | 2,452 | 4,450 | +81% | 0 | 0 | — |
case-04 | pass→pass | 25,066 | 37,387 | +49% | 1 | 1 | 0% | 3,176 | 6,287 | +98% | 0 | 0 | — |
case-05 | pass→pass | 17,014 | 18,696 | +10% | 1 | 1 | 0% | 2,121 | 2,901 | +37% | 0 | 0 | — |
case-06 | pass→pass | 23,312 | 20,647 | -11% | 1 | 1 | 0% | 2,598 | 3,331 | +28% | 0 | 0 | — |
case-07 | fail→pass | 17,300 | 19,454 | +12% | 1 | 1 | 0% | 1,747 | 3,057 | +75% | 0 | 0 | — |
case-08 | pass→pass | 19,725 | 13,795 | -30% | 1 | 1 | 0% | 2,225 | 2,928 | +32% | 0 | 0 | — |
case-09 | pass→pass | 18,309 | 22,984 | +26% | 1 | 1 | 0% | 2,264 | 3,744 | +65% | 0 | 0 | — |
case-10 | fail→pass | 18,061 | 13,011 | -28% | 1 | 1 | 0% | 1,928 | 1,143 | -41% | 0 | 0 | — |
case-11 | fail→pass | 14,671 | 8,978 | -39% | 1 | 1 | 0% | 1,442 | 1,164 | -19% | 0 | 0 | — |
case-12 | fail→fail | 34,213 | 9,004 | -74% | 1 | 1 | 0% | 2,454 | 1,211 | -51% | 0 | 0 | — |
case-13 | fail→pass | 13,115 | 9,053 | -31% | 1 | 1 | 0% | 2,266 | 2,103 | -7% | 0 | 0 | — |
case-14 | fail→pass | 31,729 | 14,943 | -53% | 1 | 1 | 0% | 2,216 | 2,224 | +0% | 0 | 0 | — |
case-15 | pass→pass | 25,538 | 8,309 | -67% | 1 | 1 | 0% | 1,138 | 892 | -22% | 0 | 0 | — |
case-16 | fail→pass | 9,565 | 7,663 | -20% | 1 | 1 | 0% | 1,306 | 910 | -30% | 0 | 0 | — |
case-17 | pass→pass | 14,038 | 4,482 | -68% | 1 | 1 | 0% | 1,397 | 776 | -44% | 0 | 0 | — |
case-18 | pass→pass | 13,973 | 9,918 | -29% | 1 | 1 | 0% | 1,356 | 759 | -44% | 0 | 0 | — |
case-19 | pass→pass | 8,642 | 7,441 | -14% | 1 | 1 | 0% | 1,364 | 778 | -43% | 0 | 0 | — |
case-20 | pass→pass | 16,085 | 6,952 | -57% | 1 | 1 | 0% | 1,502 | 1,677 | +12% | 0 | 0 | — |
case-21 | pass→pass | 14,355 | 6,752 | -53% | 1 | 1 | 0% | 2,227 | 1,622 | -27% | 0 | 0 | — |
case-22 | pass→pass | 11,179 | 14,342 | +28% | 1 | 1 | 0% | 1,815 | 1,783 | -2% | 0 | 0 | — |
case-23 | fail→fail | 10,969 | 3,291 | -70% | 1 | 1 | 0% | 1,640 | 879 | -46% | 0 | 0 | — |
case-24 | fail→pass | 17,013 | 18,671 | +10% | 1 | 1 | 0% | 2,406 | 2,646 | +10% | 0 | 0 | — |
case-25 | fail→pass | 6,028 | 4,918 | -18% | 1 | 1 | 0% | 740 | 980 | +32% | 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. 25 cases were attempted. The headline lift of +36 percentage points is the difference between those two pass rates over the 25 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.