Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Write and confirm a short implementation plan before starting any large or risky change. Use this for tasks that touch three or more files, alter public APIs, database schemas, auth, payments, build systems, or CI, for migrations and multi-step refactors, and whenever the honest estimate is "this will take many steps". Also trigger it when the user asks "how would you approach this", says "plan", or when the request is ambiguous enough that two reasonable engineers would build different things.
.claude/skills/adityaarakeri-plan-first/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 5% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 29% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -20% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 5% | 0% |
Convert uncertainty into explicit decisions before implementation.
Keep the plan concise and include:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→fail | 16,412 | 15,134 | -8% | 1 | 1 | 0% | 3,161 | 2,656 | -16% | 0 | 0 | — |
case-08 | fail→fail | 9,027 | 14,271 | +58% | 1 | 1 | 0% | 1,746 | 2,946 | +69% | 0 | 0 | — |
case-01 | fail→pass | 20,895 | 13,296 | -36% | 1 | 1 | 0% | 2,468 | 2,597 | +5% | 0 | 0 | — |
case-03 | fail→fail | 22,513 | 13,498 | -40% | 1 | 1 | 0% | 3,342 | 573 | -83% | 0 | 0 | — |
case-04 | fail→fail | 13,838 | 15,363 | +11% | 1 | 1 | 0% | 2,612 | 3,066 | +17% | 0 | 0 | — |
case-05 | fail→fail | 11,723 | 13,439 | +15% | 1 | 1 | 0% | 2,142 | 2,659 | +24% | 0 | 0 | — |
case-06 | fail→fail | 15,906 | 14,252 | -10% | 1 | 1 | 0% | 2,808 | 2,707 | -4% | 0 | 0 | — |
case-07 | fail→fail | 9,888 | 5,351 | -46% | 1 | 1 | 0% | 2,043 | 615 | -70% | 0 | 0 | — |
case-09 | fail→fail | 14,342 | 10,812 | -25% | 1 | 1 | 0% | 2,833 | 2,345 | -17% | 0 | 0 | — |
case-10 | fail→pass | 14,995 | 17,334 | +16% | 1 | 1 | 0% | 2,650 | 3,420 | +29% | 0 | 0 | — |
case-11 | fail→pass | 14,392 | 9,256 | -36% | 1 | 1 | 0% | 2,606 | 2,025 | -22% | 0 | 0 | — |
case-12 | fail→fail | 14,662 | 9,828 | -33% | 1 | 1 | 0% | 2,433 | 1,956 | -20% | 0 | 0 | — |
case-13 | pass→pass | 12,833 | 12,172 | -5% | 1 | 1 | 0% | 2,140 | 2,210 | +3% | 0 | 0 | — |
case-14 | fail→pass | 17,830 | 14,380 | -19% | 1 | 1 | 0% | 3,157 | 2,541 | -20% | 0 | 0 | — |
case-15 | fail→pass | 14,366 | 14,033 | -2% | 1 | 1 | 0% | 2,616 | 2,755 | +5% | 0 | 0 | — |
case-16 | pass→fail | 10,055 | 5,968 | -41% | 1 | 1 | 0% | 2,180 | 530 | -76% | 0 | 0 | — |
case-17 | pass→pass | 10,968 | 11,662 | +6% | 1 | 1 | 0% | 2,225 | 2,246 | +1% | 0 | 0 | — |
case-18 | fail→pass | 9,697 | 11,480 | +18% | 1 | 1 | 0% | 2,267 | 2,661 | +17% | 0 | 0 | — |
case-19 | fail→pass | 13,068 | 9,952 | -24% | 1 | 1 | 0% | 3,199 | 2,364 | -26% | 0 | 0 | — |
case-20 | pass→pass | 3,529 | 3,298 | -7% | 1 | 1 | 0% | 988 | 1,212 | +23% | 0 | 0 | — |
case-21 | pass→pass | 2,365 | 3,684 | +56% | 1 | 1 | 0% | 527 | 1,189 | +126% | 0 | 0 | — |
case-22 | pass→pass | 11,478 | 12,824 | +12% | 1 | 1 | 0% | 2,445 | 2,995 | +22% | 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 19 counted toward the lift figure. The other 3 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 +27 percentage points is the difference between those two pass rates over the 19 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/3/2026 | +23% |
| gemini-3.6-flash | verified | 8/3/2026 | +45% |
Other measured skills in the registry, with their headline benchmark lift.