Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →You are an expert prompt engineer specializing in crafting effective prompts for LLMs through advanced techniques including constitutional AI, chain-of-thought reasoning, and model-specific optimizati
.claude/skills/dokhacgiakhoa-llm-application-dev-prompt-optimize/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | 55% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | 21% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 8% | 0% |
| case-06 | ✓→✓ | ＝ Same ✓ | 12% | 0% |
| case-01 | ✗→✗ | ＝ Same ✗ | 25% | 0% |
You are an expert prompt engineer specializing in crafting effective prompts for LLMs through advanced techniques including constitutional AI, chain-of-thought reasoning, and model-specific optimization.
Transform basic instructions into production-ready prompts. Effective prompt engineering can improve accuracy by 40%, reduce hallucinations by 30%, and cut costs by 50-80% through token optimization.
$ARGUMENTS
resources/implementation-playbook.md.resources/implementation-playbook.md for detailed patterns and examples.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 12,976 | 20,382 | +57% | 1 | 1 | 0% | 2,507 | 3,124 | +25% | 0 | 0 | — |
case-02 | fail→fail | 12,455 | 15,670 | +26% | 1 | 1 | 0% | 2,078 | 2,533 | +22% | 0 | 0 | — |
case-03 | fail→fail | 20,838 | 21,658 | +4% | 1 | 1 | 0% | 3,396 | 3,705 | +9% | 0 | 0 | — |
case-04 | pass→pass | 18,951 | 22,674 | +20% | 1 | 1 | 0% | 3,227 | 3,906 | +21% | 0 | 0 | — |
case-05 | pass→pass | 14,886 | 15,730 | +6% | 1 | 1 | 0% | 2,994 | 3,237 | +8% | 0 | 0 | — |
case-06 | pass→pass | 22,131 | 25,280 | +14% | 1 | 1 | 0% | 3,821 | 4,261 | +12% | 0 | 0 | — |
case-07 | fail→fail | 16,970 | 11,612 | -32% | 1 | 1 | 0% | 2,455 | 2,642 | +8% | 0 | 0 | — |
case-08 | fail→fail | 16,082 | 12,608 | -22% | 1 | 1 | 0% | 2,276 | 2,408 | +6% | 0 | 0 | — |
case-09 | fail→pass | 15,636 | 19,142 | +22% | 1 | 1 | 0% | 2,207 | 3,418 | +55% | 0 | 0 | — |
case-10 | fail→fail | 18,988 | 15,808 | -17% | 1 | 1 | 0% | 3,463 | 3,008 | -13% | 0 | 0 | — |
case-11 | fail→fail | 12,263 | 13,169 | +7% | 1 | 1 | 0% | 2,337 | 1,951 | -17% | 0 | 0 | — |
case-12 | fail→fail | 14,552 | 14,667 | +1% | 1 | 1 | 0% | 2,090 | 2,830 | +35% | 0 | 0 | — |
case-13 | fail→fail | 15,323 | 14,156 | -8% | 1 | 1 | 0% | 2,366 | 2,439 | +3% | 0 | 0 | — |
case-14 | fail→fail | 9,878 | 11,453 | +16% | 1 | 1 | 0% | 1,529 | 1,910 | +25% | 0 | 0 | — |
case-15 | fail→fail | 18,751 | 20,549 | +10% | 1 | 1 | 0% | 2,708 | 3,339 | +23% | 0 | 0 | — |
case-16 | fail→fail | 14,306 | 21,182 | +48% | 1 | 1 | 0% | 2,204 | 3,072 | +39% | 0 | 0 | — |
case-17 | fail→fail | 16,030 | 14,812 | -8% | 1 | 1 | 0% | 2,181 | 2,652 | +22% | 0 | 0 | — |
case-18 | fail→fail | 15,407 | 15,249 | -1% | 1 | 1 | 0% | 2,151 | 2,794 | +30% | 0 | 0 | — |
case-19 | fail→fail | 13,580 | 10,690 | -21% | 1 | 1 | 0% | 1,835 | 2,052 | +12% | 0 | 0 | — |
case-20 | fail→fail | 19,761 | 19,522 | -1% | 1 | 1 | 0% | 2,260 | 2,968 | +31% | 0 | 0 | — |
case-21 | fail→fail | 18,550 | 17,662 | -5% | 1 | 1 | 0% | 2,634 | 2,740 | +4% | 0 | 0 | — |
case-22 | fail→fail | 12,566 | 14,386 | +14% | 1 | 1 | 0% | 1,917 | 2,590 | +35% | 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 +5 percentage points is the difference between those two pass rates over the 22 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.