Youssef
78167667a2
REMOVED (per user direction — imagery is on hold for now): - site/img/generated/ (18 watercolor PNGs + contact sheet + URL cache) - scripts/generate-marketing.mjs - skill/_assets/scripts/generate.mjs (user-recipe drop-in) - skill/references/generated-imagery.md - All imagery markup in site/index.html (transitions, foundation panel plates, harness plates, footer mark) - The .wc / .transition / .slot--hero::before / #study::before / .found-panel__plate / .harness__plate / .foot-mark CSS block from sections.css - Tier C "peer of Tier B" rewrite reverted — back to "last resort, only when characters demand it" - Anti-pattern "Raw generated stills" entry reverted - assets.md Tier C catalog reverted to Nanobanana 2 / Gemini 2.5 Flash baseline - README v0.9.0 release note rewritten without imagery mentions KEPT: - Editorial theme (#23) — open-design-inspired premium · warm cream + coral + Inter Tight × Playfair italic - Open-design steal utilities — Roman-numeral section heads (.num--roman), corner-bracket image annotation (.fig-corners), dark-slab section reversal (.section--invert), pulse animation (.pulse), .tnum tabular-numerals utility, smart-hide banner stub (.banner--smart) - 22 → 23 theme catalog refresh across SKILL.md, custom-theme.md, structure.md, README, site - Version 0.8.0 → 0.9.0 NEW: Spacing pass — normalized internal rhythm across all 23 themes - --section-gap normalized: was 4.5–12rem (huge variance), now 5.5–6.5rem (uniform feel) · Atelier 12 → 6.5 (biggest cut) · Quiet 10 → 6.5 · Linen 10 → 6.5 · Aurora 10 → 6.5 · Garden 9 → 6.5 · Salon 9 → 6.5 · Bloom 9 → 6.5 · Specimen 8 → 6.5 · Almanac 4.5 → 5.5 (slight bump) - --section-head-gap normalized: was 1–3.5rem, now 1.25–1.75rem (smaller spread) - .section { gap: var(--space-2xl) } → var(--space-xl) — content sits closer to its head; less air inside each section - Editorial theme block gains explicit --section-gap + --section-head-gap (was inheriting from :root) Visual: every theme now has the same vertical rhythm (~6rem between sections, ~1.5rem under heads). The page reads as one document instead of 23 different ones.
1.5 KiB
audit verb · what fired and why
Input
A 148-line index.html with inline CSS — a generic AI-shaped landing page for a fictional product called "Nexus." Hits at least 14 anti-patterns from references/anti-patterns.md.
What audit does (and what it doesn't)
- Reads the file. No edits. No new files. The verb is read-only.
- Scores against
anti-patterns.md+ the slop test gates inSKILL.md. - Returns one Markdown report with: tell name · file path + line range · severity · one-line fix.
- Groups by severity (critical / major / minor) and ends with a count.
What loaded
references/anti-patterns.md(the full named-tell library)references/structure.md(for the structural-fingerprint check, gate 9)references/microinteractions.md(for the timing/easing canon, gates 11–18)SKILL.md§ The slop test (gates 1–38)
Why this matters as a verb test
audit is the verb that doesn't change anything. It's the safety verb — the one a user runs before they ask for refine or redesign. The output gives the user enough information to decide which way to take the page.
In this case the audit returns "do not run refine — the structural fingerprint is the problem; run redesign instead." That is the ideal output: a verb that helps the user pick another verb.
Durable artifact
audit-report.md — the kind of output Hallmark produces in a real session. It can be saved to a PR, pasted into a Linear ticket, or used as the brief for the follow-up redesign run.