Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Open Design's operating review: growth, burn, and the concrete path to sustainability without losing the open ethos. Built as a decision-grade corporate strategy deck for leadership team.
.claude/skills/nexu-io-simple-deck/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -3% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 22% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 6% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 16% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 11% | 0% |
Produce a single-file horizontal-swipe HTML deck using the seed and layout library.
simple-deck/
├── SKILL.md ← you're reading this
├── assets/
│ └── template.html ← seed: tokens + slide primitives + proven nav script (READ FIRST)
└── references/
├── layouts.md ← 8 paste-ready slide layouts + theme-rhythm rules
└── checklist.md ← P0/P1/P2 self-review (rhythm spot-check at bottom)assets/template.html end-to-end through the <style> block AND the <script> block. The script solves five iframe-specific bugs (real scroller detection, dual capture-phase listeners, auto-focus, no scrollIntoView, position persistence) — do not rewrite it.references/layouts.md so you know the 8 layouts. Pay special attention to the "Theme rhythm" section — it's the rule that prevents the deck from feeling sleepy.:root variables in the seed.Copy assets/template.html to the project root as index.html. Replace the six :root variables with the active design system's tokens. Replace the page <title>.
Default: 6 slides unless the brief says otherwise.
| Audience / format | Slides | |---|---| | Product overview / lightning talk (5–10 min) | 6 | | Pitch deck (15 min) | 8–10 | | Investor update / longer talk (20–30 min) | 12–18 |
Then write out the rhythm before any HTML — for example, 8 slides:
01 hero light center Cover
02 light Problem
03 hero dark center Big stat
04 light Three points
05 dark Pipeline
06 hero light center Quote
07 light Before / after
08 hero dark center AskA healthy sequence has:
hero dark AND ≥ 1 hero light (for 8+ slides)Show this rhythm sketch to the user before writing slide HTML — they can redirect cheaply.
For each planned slide, copy the matching <section> from layouts.md into the body. Replace bracketed text with real, specific copy. No filler / no lorem. If a slide feels empty, the layout is wrong — pick a different one.
Tag each slide with data-screen-label="01 Cover", "02 Problem", etc., in the order you wrote them. (The seed's first three slides already do this — extend the pattern.)
Run through references/checklist.md. The "Theme rhythm spot-check" at the end is non-negotiable:
bashgrep 'class="slide' index.html
Read the resulting class list. If you see light × 4 in a row, swap one to dark. If no hero dark exists in an 8+ slide deck, promote one big-stat or closing slide.
Write the completed deck HTML to index.html.
Then send one short ordinary assistant summary naming index.html and describing the deck. Do not output the full HTML source in chat and do not emit a source-code <artifact> block.
light | dark | hero light | hero dark). Bare class="slide" = regression.var(--font-display). .h-hero / .h-xl / .h-md already enforce.scrollIntoView(). Breaks iframe.data-screen-label on every slide.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 40,603 | 14,114 | -65% | 1 | 1 | 0% | 8,267 | 1,527 | -82% | 0 | 0 | — |
case-02 | fail→fail | 49,754 | 52,763 | +6% | 1 | 1 | 0% | 8,073 | 9,284 | +15% | 0 | 0 | — |
case-03 | fail→fail | 36,338 | 36,711 | +1% | 1 | 1 | 0% | 8,254 | 9,268 | +12% | 0 | 0 | — |
case-04 | pass→pass | 7,191 | 11,637 | +62% | 1 | 1 | 0% | 1,296 | 2,612 | +102% | 0 | 0 | — |
case-05 | pass→pass | 10,289 | 29,955 | +191% | 1 | 1 | 0% | 1,470 | 6,367 | +333% | 0 | 0 | — |
case-06 | pass→fail | 13,498 | 36,046 | +167% | 1 | 1 | 0% | 2,318 | 9,235 | +298% | 0 | 0 | — |
case-07 | fail→pass | 18,419 | 8,477 | -54% | 1 | 1 | 0% | 2,583 | 2,507 | -3% | 0 | 0 | — |
case-08 | fail→pass | 15,087 | 12,816 | -15% | 1 | 1 | 0% | 2,317 | 2,820 | +22% | 0 | 0 | — |
case-09 | fail→pass | 13,031 | 7,256 | -44% | 1 | 1 | 0% | 1,990 | 2,100 | +6% | 0 | 0 | — |
case-10 | fail→pass | 9,039 | 5,647 | -38% | 1 | 1 | 0% | 1,441 | 1,677 | +16% | 0 | 0 | — |
case-11 | fail→fail | 15,828 | 10,205 | -36% | 1 | 1 | 0% | 2,251 | 2,459 | +9% | 0 | 0 | — |
case-12 | fail→pass | 12,183 | 6,445 | -47% | 1 | 1 | 0% | 1,673 | 1,857 | +11% | 0 | 0 | — |
case-13 | fail→pass | 16,145 | 9,297 | -42% | 1 | 1 | 0% | 2,187 | 2,034 | -7% | 0 | 0 | — |
case-14 | pass→pass | 15,665 | 44,586 | +185% | 1 | 1 | 0% | 2,124 | 7,901 | +272% | 0 | 0 | — |
case-15 | fail→pass | 13,197 | 4,185 | -68% | 1 | 1 | 0% | 1,877 | 1,694 | -10% | 0 | 0 | — |
case-16 | fail→fail | 35,022 | 33,514 | -4% | 1 | 1 | 0% | 8,223 | 9,237 | +12% | 0 | 0 | — |
case-17 | pass→pass | 9,390 | 2,814 | -70% | 1 | 1 | 0% | 1,586 | 1,342 | -15% | 0 | 0 | — |
case-18 | fail→pass | 7,296 | 1,653 | -77% | 1 | 1 | 0% | 1,193 | 1,215 | +2% | 0 | 0 | — |
case-19 | fail→pass | 15,870 | 5,410 | -66% | 1 | 1 | 0% | 2,074 | 1,927 | -7% | 0 | 0 | — |
case-20 | fail→fail | 25,257 | 16,938 | -33% | 1 | 1 | 0% | 5,472 | 4,168 | -24% | 0 | 0 | — |
case-21 | fail→pass | 10,667 | 9,078 | -15% | 1 | 1 | 0% | 1,428 | 1,872 | +31% | 0 | 0 | — |
case-22 | fail→fail | 15,709 | 11,155 | -29% | 1 | 1 | 0% | 2,108 | 2,473 | +17% | 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 +41 percentage points is the difference between those two pass rates over the 21 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.