Install any skill in seconds. Free to start, no credit card required.
Get Started Free →A brand-repositioning strategy for a heritage coffee chain — why, how, what — the governing idea and the moves to prove it. Built as a decision-grade consulting deck for client CMO, board.
.claude/skills/nexu-io-huashu-golden-circle/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 88% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 16% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 158% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 66% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 501% | 0% |
Produce a single-file, one-diagram concept deck. You are visualizing a methodology the way Simon Sinek drew the Golden Circle on a TED flip chart: one geometric master figure carries the entire argument, and every page is that figure seen from a different angle. The visual system, canvas contract, and navigation runtime are locked by example.html. Start from example.html, replace content only — do not rewrite the design or the script. Do not introduce any color or font outside this spec.
Generated from the 中性派「单图母图概念图解 / Diagrammatic Minimalism」spec in references/design-styles.md of huashu-design by 花叔 (alchaincyf), MIT licensed. Lineage: Simon Sinek's Golden Circle TED talk, Bauhaus geometric abstraction, "one diagram rules the room" information architecture.
<div id="stage"> fixed at 1920 × 1080 px, centered withposition: fixed; top: 50%; left: 50%; a fit() function applies translate(-50%, -50%) scale(min(innerWidth/1920, innerHeight/1080)) on load and resize. All inner layout in px — the scaler owns responsiveness.
<section class="slide"> inside #stage with adata-screen-label="01 封面"-style label; exactly one slide carries .active; the dark closing page adds .dark.
←/↑/PageUp previous,→/↓/Space/PageDown next, Home/End first/last; #/N hash routing (1-indexed) read on load + hashchange, written via history.replaceState; click left third = back, rest = forward; fixed counter pill bottom-right, key-hint bottom-left. No external JS, no build step — the file must open inside a sandboxed iframe via file://.
:root — keep the names, re-theme values only)| Token | Value | Role | |---|---|---| | --paper | #FAFAF7 | warm-white page background | | --ink | #141414 | primary text | | --night | #161512 | dark closing-page background | | --night-ink | #F5F2EA | primary text on dark | | --accent | #E8860C | the only accent hue (gold) | | --tint-2 | #F2BA6B | middle-ring tint of the accent | | --tint-3 | #FAE5C3 | outer-ring tint of the accent | | --neutral-2 | #E8E4D9 | de-emphasized ring (warm grey) | | --neutral-3 | #F2EFE7 | de-emphasized outer ring | | --line | #E5E1D6 | hairline rules | | --muted | #8A867C | secondary labels | | --muted-2 | #56524A | body copy | | --faint | #B5B0A4 | tertiary captions | | --night-line | #3A372F | hairlines on the dark page | | --display | 'Jost', 'Manrope', 'PingFang SC', … | uppercase tracked labels | | --sans | 'Manrope', 'PingFang SC', … | headlines + body |
@import(Jost 400–700 + Manrope 400–800); Chinese falls back to PingFang SC / Microsoft YaHei. No other @import, no CDN scripts, no icon fonts.
--tint-2/--tint-3 are tints of the accent, not second colors. A secondaccent hue is the cardinal sin of this style.
The .gc component — nested concentric circles, pure border-radius: 50% divs — is the only diagram in the deck. Keep its geometry verbatim:
.gc { --gc: <size>px } sets the diameter; rings at inset: 0 (WHAT,--tint-3), inset: 16% (HOW, --tint-2), inset: 33% (WHY, solid --accent, flex-centered core label).
.lbl, Jost 600, letter-spacing: .35em) are embeddedinside the bands: WHAT centered at 8% height, HOW at 24.5%, WHY in the core at calc(var(--gc) * 0.085) in --paper white.
.mode-how / .mode-what: target band --accent (its label --paper),other bands --neutral-3/--paper;
.mode-grey: all-neutral counter-example (core --muted);.mode-outline: transparent rings with 2px --accent strokes — darkclosing page only.
.why-disc, the WHY page'szoomed core), faint watermark rings (1px accent border, opacity ≤ 0.3, may bleed off-canvas), mini three-circle inline-SVG glyphs (36px, one band accented to mark a section), and inside-out arrows (inline SVG: dot at the center, 2–2.5px line, solid polygon head). Nothing else — no photos, no charts, no emoji, no decorative illustration.
core/loop/surface…) but there are always exactly three bands, and the innermost band is always the thesis.
.head (top) and .foot (bottom) — 84px strips atleft/right: 80px with a 1px --line border (--night-line on dark), Jost 13px/600/3px-tracking uppercase; left side carries the deck logotype in --ink <b> + a gold / separator; right side of .foot carries NN / 章节名. Content lives in .body (top/bottom: 84px, left/right: 80px).
the WHAT evidence cards at 18px radius); everything else is a circle or a straight hairline.
| Layout | Role | |---|---| | cover | left headline (gold full stop 。) + hairline meta row; right master figure | | contents | hairline rows: gold number + mini glyph + title + uppercase EN label | | master | the figure at ≈660px + inside-out arrow piercing the rings + per-ring notes | | why | solid accent disc (zoomed core) + one belief statement, ≤ 2 sentences | | how | .mode-how figure + three hairline-separated pillar columns | | what | .mode-what figure + 2×2 evidence cards with glyph markers | | big-number | 300px Manrope-800 numeral (accent unit glyph) + stat rail + watermark rings | | comparison | grey circle (inward arrow, outside-in) vs gold circle (outward arrow, inside-out) | | quote | 66px/800 quote, key word in accent, offset watermark rings bleeding right | | closing | dark page, .mode-outline figure + CTA in a 2px accent outlined capsule |
full stop 。 is the accent. Body (.lede) 20px/1.85 --muted-2 with <strong> ink emphasis — statements, not bullet walls; any list ≤ 3 short lines.
.kicker, .caption, ring labels): Jost 600, 12–15px, 3–5pxtracking, uppercase — the connective tissue of the style.
to single glyphs and key words, never whole headings.
Accent budget: outside the figure itself, ≤ 3 gold touches per page.
to on the circle, it does not belong in this deck.
three-band framework (what is the WHY / HOW / WHAT of their argument), plus 1–2 protagonist numbers for the evidence page. The master figure is the deck; name its bands before writing any page.
example.html, retitle, rename the ring labels if needed, thenreplace each section's content following the layout enumeration. Keep the .gc component, tokens, hairline chrome, and the script intact. Re-theme by changing token values only — and keep --tint-2/--tint-3 as tints of whatever single accent you choose.
page, confirm the figure's band names read correctly, then batch the rest.
#/N hash stay in sync; no overflow beyond 1920×1080; every page shows the figure or one of its locked derivatives; ring labels sit inside their bands at every --gc size used; exactly one accent hue in the whole file; no leftover demo ("同心"/"TONGXIN") text; grep for TODO.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | fail→pass | 14,687 | 23,106 | +57% | 1 | 1 | 0% | 2,623 | 4,937 | +88% | 0 | 0 | — |
case-01 | fail→fail | 54,529 | 65,805 | +21% | 1 | 1 | 0% | 8,293 | 10,823 | +31% | 0 | 0 | — |
case-02 | fail→fail | 86,368 | 43,680 | -49% | 1 | 1 | 0% | 8,284 | 10,814 | +31% | 0 | 0 | — |
case-03 | fail→pass | 21,344 | 10,758 | -50% | 1 | 1 | 0% | 3,478 | 4,041 | +16% | 0 | 0 | — |
case-05 | fail→pass | 13,665 | 22,703 | +66% | 1 | 1 | 0% | 2,316 | 5,968 | +158% | 0 | 0 | — |
case-06 | pass→pass | 12,769 | 15,316 | +20% | 1 | 1 | 0% | 2,526 | 4,902 | +94% | 0 | 0 | — |
case-07 | pass→pass | 21,233 | 7,579 | -64% | 1 | 1 | 0% | 3,003 | 3,638 | +21% | 0 | 0 | — |
case-08 | fail→pass | 15,839 | 16,952 | +7% | 1 | 1 | 0% | 2,741 | 4,557 | +66% | 0 | 0 | — |
case-09 | fail→pass | 9,356 | 49,526 | +429% | 1 | 1 | 0% | 1,788 | 10,754 | +501% | 0 | 0 | — |
case-19 | fail→pass | 17,387 | 7,768 | -55% | 1 | 1 | 0% | 2,455 | 3,869 | +58% | 0 | 0 | — |
case-10 | fail→fail | 17,932 | 38,484 | +115% | 1 | 1 | 0% | 2,434 | 10,748 | +342% | 0 | 0 | — |
case-11 | fail→pass | 8,489 | 36,692 | +332% | 1 | 1 | 0% | 1,336 | 10,751 | +705% | 0 | 0 | — |
case-12 | pass→pass | 11,208 | 4,562 | -59% | 1 | 1 | 0% | 1,806 | 3,031 | +68% | 0 | 0 | — |
case-13 | fail→fail | 12,100 | 39,221 | +224% | 1 | 1 | 0% | 1,862 | 10,740 | +477% | 0 | 0 | — |
case-14 | fail→pass | 20,405 | 3,580 | -82% | 1 | 1 | 0% | 1,857 | 3,054 | +64% | 0 | 0 | — |
case-15 | fail→pass | 11,745 | 5,866 | -50% | 1 | 1 | 0% | 1,963 | 3,332 | +70% | 0 | 0 | — |
case-16 | fail→pass | 12,596 | 36,725 | +192% | 1 | 1 | 0% | 1,656 | 10,757 | +550% | 0 | 0 | — |
case-17 | fail→pass | 18,323 | 12,446 | -32% | 1 | 1 | 0% | 2,546 | 4,385 | +72% | 0 | 0 | — |
case-18 | fail→pass | 15,300 | 29,472 | +93% | 1 | 1 | 0% | 2,402 | 7,012 | +192% | 0 | 0 | — |
case-20 | fail→pass | 13,614 | 10,401 | -24% | 1 | 1 | 0% | 1,911 | 4,065 | +113% | 0 | 0 | — |
case-21 | fail→pass | 9,165 | 7,922 | -14% | 1 | 1 | 0% | 1,388 | 3,776 | +172% | 0 | 0 | — |
case-22 | pass→fail | 25,861 | 50,918 | +97% | 1 | 1 | 0% | 3,940 | 10,761 | +173% | 0 | 0 | — |
case-23 | pass→fail | 34,122 | 39,117 | +15% | 1 | 1 | 0% | 5,477 | 10,748 | +96% | 0 | 0 | — |
case-24 | pass→fail | 23,094 | 40,905 | +77% | 1 | 1 | 0% | 5,262 | 10,751 | +104% | 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. 24 cases were attempted. The headline lift of +46 percentage points is the difference between those two pass rates over the 24 comparable cases. 3 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.