Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Self-correction via critique loop - 18.5 percentage point improvement
.claude/skills/miosa-osa-reflection-loop/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -19% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-21 | ✗→✓ | ▲ Improved | -10% | 0% |
| case-01 | ✓→✓ | ＝ Same ✓ | -58% | 0% |
| case-12 | ✓→✓ | ＝ Same ✓ | 40% | 0% |
Self-correction mechanism that achieves 18.5 percentage point improvement (78.6% → 97.1%).
This skill auto-activates on:
Generate first-pass response to the task.
Ask yourself:
List specific issues found:
For each issue:
The reflection loop integrates with:
Track reflection effectiveness:
Based on research showing 18.5pp improvement with reflection patterns
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | fail→pass | 41,355 | 30,649 | -26% | 1 | 1 | 0% | 8,220 | 6,659 | -19% | 0 | 0 | — |
case-01 | pass→pass | 34,735 | 18,675 | -46% | 1 | 1 | 0% | 6,624 | 2,763 | -58% | 0 | 0 | — |
case-02 | fail→fail | 19,143 | 17,311 | -10% | 1 | 1 | 0% | 3,596 | 3,454 | -4% | 0 | 0 | — |
case-12 | pass→pass | 14,897 | 19,262 | +29% | 1 | 1 | 0% | 2,919 | 4,091 | +40% | 0 | 0 | — |
case-03 | pass→pass | 5,239 | 6,490 | +24% | 1 | 1 | 0% | 1,072 | 1,628 | +52% | 0 | 0 | — |
case-04 | pass→pass | 13,739 | 12,407 | -10% | 1 | 1 | 0% | 2,558 | 2,761 | +8% | 0 | 0 | — |
case-05 | pass→pass | 5,787 | 8,600 | +49% | 1 | 1 | 0% | 1,118 | 1,757 | +57% | 0 | 0 | — |
case-06 | pass→pass | 23,424 | 23,846 | +2% | 1 | 1 | 0% | 4,736 | 5,403 | +14% | 0 | 0 | — |
case-08 | fail→fail | 18,263 | 22,547 | +23% | 1 | 1 | 0% | 3,493 | 4,693 | +34% | 0 | 0 | — |
case-09 | pass→pass | 22,587 | 20,281 | -10% | 1 | 1 | 0% | 4,085 | 4,028 | -1% | 0 | 0 | — |
case-10 | pass→pass | 23,601 | 19,943 | -15% | 1 | 1 | 0% | 4,487 | 4,056 | -10% | 0 | 0 | — |
case-11 | pass→pass | 39,023 | 29,539 | -24% | 1 | 1 | 0% | 8,035 | 6,242 | -22% | 0 | 0 | — |
case-13 | pass→pass | 26,769 | 20,158 | -25% | 1 | 1 | 0% | 5,096 | 4,095 | -20% | 0 | 0 | — |
case-14 | fail→fail | 34,064 | 27,778 | -18% | 1 | 1 | 0% | 6,550 | 5,800 | -11% | 0 | 0 | — |
case-15 | pass→pass | 27,294 | 20,913 | -23% | 1 | 1 | 0% | 5,440 | 4,315 | -21% | 0 | 0 | — |
case-16 | pass→pass | 35,357 | 35,141 | -1% | 1 | 1 | 0% | 7,698 | 7,582 | -2% | 0 | 0 | — |
case-17 | pass→pass | 24,275 | 24,441 | +1% | 1 | 1 | 0% | 4,551 | 5,264 | +16% | 0 | 0 | — |
case-18 | fail→pass | 26,054 | 27,395 | +5% | 1 | 1 | 0% | 5,084 | 5,551 | +9% | 0 | 0 | — |
case-19 | pass→pass | 29,132 | 21,968 | -25% | 1 | 1 | 0% | 4,052 | 4,625 | +14% | 0 | 0 | — |
case-20 | pass→pass | 23,024 | 19,518 | -15% | 1 | 1 | 0% | 4,248 | 4,029 | -5% | 0 | 0 | — |
case-21 | fail→pass | 25,200 | 19,780 | -22% | 1 | 1 | 0% | 4,993 | 4,508 | -10% | 0 | 0 | — |
case-22 | pass→pass | 31,348 | 27,327 | -13% | 1 | 1 | 0% | 6,734 | 5,852 | -13% | 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 +14 percentage points is the difference between those two pass rates over the 22 comparable cases.
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.