Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →CRITICAL. MUST activate when intent is unclear, you cannot follow original intent, you cannot easily or reliably solve the problem, something came as SURPRISE or UNEXPECTED, you cannot bet $100 on your solution, you detect unknowns or assumptions that critically affect the solution, you detect deviation NOT complying with original intent, you panic, or user asked to UNDO.
.claude/skills/griddynamics-deviation/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -46% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -41% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -7% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -54% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -30% | 0% |
<deviation>
<process>
post-mortem for full harness diagnosis; recommendation is required, NEVER run it yourself.</process>
<pitfalls>
</pitfalls>
</deviation>
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 11,053 | 4,668 | -58% | 1 | 1 | 0% | 1,742 | 941 | -46% | 0 | 0 | — |
case-02 | fail→pass | 10,103 | 4,516 | -55% | 1 | 1 | 0% | 1,638 | 969 | -41% | 0 | 0 | — |
case-03 | fail→fail | 11,660 | 5,045 | -57% | 1 | 1 | 0% | 1,790 | 1,075 | -40% | 0 | 0 | — |
case-04 | pass→fail | 19,040 | 5,494 | -71% | 1 | 1 | 0% | 2,764 | 1,152 | -58% | 0 | 0 | — |
case-05 | pass→fail | 8,598 | 8,169 | -5% | 1 | 1 | 0% | 1,405 | 1,778 | +27% | 0 | 0 | — |
case-06 | pass→pass | 9,367 | 7,823 | -16% | 1 | 1 | 0% | 1,615 | 1,438 | -11% | 0 | 0 | — |
case-07 | fail→pass | 8,426 | 4,923 | -42% | 1 | 1 | 0% | 1,280 | 1,184 | -8% | 0 | 0 | — |
case-08 | fail→pass | 13,451 | 5,195 | -61% | 1 | 1 | 0% | 2,235 | 1,033 | -54% | 0 | 0 | — |
case-09 | fail→pass | 7,459 | 3,785 | -49% | 1 | 1 | 0% | 1,124 | 786 | -30% | 0 | 0 | — |
case-10 | fail→pass | 7,848 | 1,865 | -76% | 1 | 1 | 0% | 1,333 | 415 | -69% | 0 | 0 | — |
case-11 | fail→pass | 5,985 | 1,802 | -70% | 1 | 1 | 0% | 1,052 | 394 | -63% | 0 | 0 | — |
case-12 | fail→pass | 10,364 | 5,315 | -49% | 1 | 1 | 0% | 1,416 | 1,074 | -24% | 0 | 0 | — |
case-13 | fail→pass | 9,617 | 2,136 | -78% | 1 | 1 | 0% | 1,522 | 463 | -70% | 0 | 0 | — |
case-14 | pass→pass | 6,609 | 4,064 | -39% | 1 | 1 | 0% | 1,029 | 841 | -18% | 0 | 0 | — |
case-15 | pass→pass | 4,333 | 2,095 | -52% | 1 | 1 | 0% | 623 | 393 | -37% | 0 | 0 | — |
case-16 | fail→pass | 5,954 | 1,833 | -69% | 1 | 1 | 0% | 758 | 457 | -40% | 0 | 0 | — |
case-17 | pass→pass | 7,661 | 1,659 | -78% | 1 | 1 | 0% | 1,103 | 392 | -64% | 0 | 0 | — |
case-18 | fail→pass | 10,074 | 5,638 | -44% | 1 | 1 | 0% | 1,498 | 992 | -34% | 0 | 0 | — |
case-19 | fail→pass | 7,724 | 2,929 | -62% | 1 | 1 | 0% | 1,346 | 660 | -51% | 0 | 0 | — |
case-20 | fail→pass | 13,915 | 3,000 | -78% | 1 | 1 | 0% | 798 | 576 | -28% | 0 | 0 | — |
case-21 | pass→pass | 13,693 | 8,050 | -41% | 1 | 1 | 0% | 2,224 | 1,518 | -32% | 0 | 0 | — |
case-22 | pass→pass | 17,647 | 6,335 | -64% | 1 | 1 | 0% | 2,947 | 1,186 | -60% | 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, and 21 counted toward the lift figure. The other 1 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 +50 percentage points is the difference between those two pass rates over the 21 comparable cases. 2 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.