Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create custom images, graphics, and visual content using AI image generation.
.claude/skills/holaboss-ai-image-generator/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 41% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 158% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 24% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 60% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 12% | 0% |
Act as the art director who briefs the image model. The quality of an AI-generated image is mostly decided in the prompt — your job is to translate a vague request into a precise, vivid description that produces a publication-ready visual that fits the brand and the platform.
Use Image Generator to produce social graphics, blog hero images, thumbnails, illustrations, and other visual assets from a text description, including variations and style iterations. Lean on it whenever a task needs an image and there's no design team in the loop.
Build prompts deliberately rather than dumping adjectives. Cover:
Be specific where it matters and silent where it doesn't — over-constraining can produce stiff results. Avoid contradictory instructions.
Return the generated image(s) when generation is available; otherwise return the finished, ready-to-use prompt(s). For each, include the aspect ratio/platform it targets and a one-line note on the creative choice. When delivering variations, label each by the axis it explores.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | fail→pass | 9,165 | 11,889 | +30% | 1 | 1 | 0% | 1,536 | 2,173 | +41% | 0 | 0 | — |
case-01 | fail→pass | 6,104 | 12,631 | +107% | 1 | 1 | 0% | 837 | 2,162 | +158% | 0 | 0 | — |
case-02 | pass→pass | 11,416 | 14,138 | +24% | 1 | 1 | 0% | 1,707 | 2,329 | +36% | 0 | 0 | — |
case-03 | pass→pass | 11,720 | 8,988 | -23% | 1 | 1 | 0% | 1,657 | 1,902 | +15% | 0 | 0 | — |
case-04 | pass→pass | 13,241 | 13,382 | +1% | 1 | 1 | 0% | 2,320 | 2,713 | +17% | 0 | 0 | — |
case-06 | pass→pass | 12,729 | 11,857 | -7% | 1 | 1 | 0% | 1,865 | 2,097 | +12% | 0 | 0 | — |
case-07 | pass→pass | 11,684 | 15,561 | +33% | 1 | 1 | 0% | 1,508 | 2,580 | +71% | 0 | 0 | — |
case-08 | pass→pass | 10,155 | 9,837 | -3% | 1 | 1 | 0% | 1,430 | 1,840 | +29% | 0 | 0 | — |
case-09 | pass→pass | 10,581 | 9,647 | -9% | 1 | 1 | 0% | 1,592 | 1,797 | +13% | 0 | 0 | — |
case-10 | pass→pass | 12,936 | 11,546 | -11% | 1 | 1 | 0% | 1,734 | 2,121 | +22% | 0 | 0 | — |
case-22 | pass→pass | 15,600 | 10,581 | -32% | 1 | 1 | 0% | 3,102 | 2,602 | -16% | 0 | 0 | — |
case-11 | pass→pass | 11,547 | 10,765 | -7% | 1 | 1 | 0% | 1,652 | 1,999 | +21% | 0 | 0 | — |
case-12 | fail→pass | 13,180 | 13,651 | +4% | 1 | 1 | 0% | 1,907 | 2,367 | +24% | 0 | 0 | — |
case-13 | fail→pass | 9,380 | 12,890 | +37% | 1 | 1 | 0% | 1,439 | 2,297 | +60% | 0 | 0 | — |
case-14 | pass→pass | 10,208 | 11,707 | +15% | 1 | 1 | 0% | 1,590 | 2,230 | +40% | 0 | 0 | — |
case-15 | pass→pass | 11,316 | 12,211 | +8% | 1 | 1 | 0% | 1,747 | 2,346 | +34% | 0 | 0 | — |
case-16 | pass→pass | 9,122 | 13,013 | +43% | 1 | 1 | 0% | 1,515 | 2,162 | +43% | 0 | 0 | — |
case-17 | fail→pass | 12,493 | 8,733 | -30% | 1 | 1 | 0% | 1,576 | 1,771 | +12% | 0 | 0 | — |
case-18 | pass→pass | 9,633 | 11,140 | +16% | 1 | 1 | 0% | 1,612 | 2,409 | +49% | 0 | 0 | — |
case-19 | pass→pass | 10,985 | 8,966 | -18% | 1 | 1 | 0% | 1,571 | 1,709 | +9% | 0 | 0 | — |
case-20 | pass→pass | 9,314 | 7,800 | -16% | 1 | 1 | 0% | 1,726 | 1,911 | +11% | 0 | 0 | — |
case-21 | pass→pass | 14,438 | 7,845 | -46% | 1 | 1 | 0% | 2,831 | 1,671 | -41% | 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 +23 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.