The skills CLI copies everything under the directory containing SKILL.md.
With SKILL.md at the repo root, installs were pulling in docs/, site/,
package.json, README, ROADMAP, vercel.json, LICENSE. Moving the skill
into skills/hallmark/ scopes the install to just SKILL.md + references/.
The CLI's priority search hits skills/ first, so `npx skills add
nutlope/hallmark` is unchanged. Updated README install paths,
package.json (entry/references/files), and the internal links that
crossed the new boundary.
Remove the "refuses competitor pages" policy from the study verb.
Touches the verb summary in README and SKILL.md, both refusal tables
in study.md, the URL refuse list, the schema's refusal enum, the
emission-attestation parenthetical, and the worked examples plus the
test fixtures.
Paid-template refusal and the signature-work soft-refusal stay —
those carry real copyright weight.
Add a brand-first flow as the lead Next item — from a short product
description, Hallmark generates a complete brand (palette, type, voice,
Nanobanana imagery) into a design.md, then builds the whole site
against it.
Rewrite the data-viz item as "Charts reference for analytics pages" —
names the actual AI tells (rainbow palettes, 3D donuts, dual-axis
spaghetti) and what the new reference file would do.
Fold the proposed hallmark extract verb into study as a third input
mode (codebase path, alongside screenshot and URL). Same verb, same
design.md output, no new surface area.
Drop the Won't-do section.
README: 147 → 81 lines. Drop the comparison table, the long
"What's distinct" bullet list, the references/ inventory, the
v1.0.0 release-note blockquote, and the duplicate Update section.
Keep the verb table, the example grid, and install paths.
ROADMAP: 201 → 50 lines. Fold the Status/Build/Why-it-matters
tri-headers into one or two sentences per item. Drop the stale
"What's already shipped" list, the "Measurement" section, and
the duplicate Deferred entry. Keep the Won't-do list — that's
where the discipline lives.
The description was an unquoted YAML plain scalar containing multiple
colon-space sequences ("Four verbs: ", "Fires when ANY: ", "Does NOT
fire on: ", "Decisive signal: "). YAML reads ': ' inside a plain
scalar as a nested mapping, so strict parsers abort with 'mapping
values are not allowed here'. The skills CLI parses this frontmatter
during 'npx skills add', so install failed.
Wrapped the value in double quotes (no internal " or backslash, so
this is lossless and stays one line). Frontmatter now parses; name,
description, and version keys intact.
Co-Authored-By: WOZCODE <contact@withwoz.com>
Codex enforces a 1024-char cap on the frontmatter description; ours was 1813,
causing "invalid description: exceeds maximum length of 1024 characters" on
load. Rewrote to 927 chars, lossless on routing signals — all five fire
triggers, all three don't-fire conditions, and the decisive heuristic are
preserved. Stripped duplicate verb listings, parenthetical examples on each
trigger, and padding prose.
README + site install panes now document `~/.codex/skills/hallmark/` (personal)
and `.codex/skills/hallmark/` (project-scoped) for Codex, replacing the prior
"manual / git clone" placeholder.
A subagent generating a Hallmark page with explicit "preserve structural parity with this reference build" instructions inherited the banned tag-left / header-right pattern (`.workbench__head` with `grid-template-columns: minmax(0, 0.4fr) minmax(0, 1fr)` and three `01 · The tour`-style eyebrows) from a launch-day reference that pre-dated gates 66/67/68/69. Gate 66 didn't catch it because the wrapper used `.workbench__head` rather than the canonical `.section__head` named in the original wording.
Two fixes:
1. **gate 66 (slop-test.md) — bind on content shape, not class.** The rule now reads "any <header>, <div>, or <section> wrapper — regardless of class name (.section__head, .workbench__head, .pricing__head, .catches__head, .foo__intro, anything macrostructure- or section-specific) — that contains both an eyebrow/label/number/kicker element AND a heading element MUST resolve to single-column layout." Class-name allowlists are out; the rule binds wherever the *shape* appears.
2. **Both rules (gate 66 + anti-patterns.md "Eyebrow on every section") — explicitly non-bypassable by parity instructions.** New clause in both: instructions to "preserve structural parity", "mirror this reference build", or "match the prior build" do NOT exempt the new build from the rule. If the reference ships the banned pattern, silently flatten it in the new build — the rules win over parity. Reference builds often pre-date these gates; the gates are authoritative.
No content/scope change beyond hardening the wording. Local Hallmark mirror updated.
The skills CLI's frontmatter parser rejected SKILL.md with "Nested mappings are not allowed in compact mappings" because the description contained " ANY of: (1) " — the colon-space sequence inside an unquoted YAML scalar made the parser think it had hit a nested key/value pair. End result: 'npx skills add nutlope/hallmark' cloned successfully but then bailed with "No skills found. Skills require a SKILL.md with name and description."
Fix: rephrase "Fires when ANY of:" → "Fires when ANY of these match —". Single character semantic — colon→em-dash. Frontmatter now parses cleanly under js-yaml. No content semantics change.
Also tightened the repo to what actually helps the skill or a future user — per the new instruction. Added to .gitignore + git rm --cached:
- .hallmark/ — runtime state from my dogfooding runs, has no business being committed
- site/_launch-comparison/ — one-off launch-day comparison artifact (Hallmark / Impeccable / no-skills); not linked from the marketing site, README, or anywhere else
Working-directory copies preserved (rm --cached, not rm). docs/talk-slides.md left in place — it's user-authored content, not mine to remove.
The published install command in the README — npx skills add nutlope/hallmark — was failing for users because the skills CLI looks for SKILL.md at the repo root, not at skill/SKILL.md where Hallmark kept it. The clone succeeded but the CLI reported "No skills found." Manual cp was the only working install path.
Flattened the layout:
- skill/SKILL.md → SKILL.md
- skill/references/ → references/
- (skill/ directory removed; was empty after moves)
git mv preserved rename history for all 76+ files (SKILL.md + 28 reference files + 21 macrostructures + 46 components + 7 verb/genre subfolders).
Updated every cross-reference (36 occurrences across 9 files):
- README.md — 4 links (SKILL.md + references/ + the install snippet still reads the same)
- ROADMAP.md — 4 links to reference files
- package.json — "files" field + "skill" entry/references paths
- site/js/main.js — 3 source comments
- references/design-md.md — one broken link (was pointing skill/references/export-formats.md from inside references/ — now just export-formats.md as a sibling)
- site/_tests/verbs/study/notes.md, diagnosis.md and verbs/audit/audit-report.md, custom/README.md — test-artifact relative paths shortened by one segment (e.g. ../../../skill/SKILL.md → ../../../SKILL.md)
No skill content changes. After this lands, 'npx skills add nutlope/hallmark' should clone + install correctly without any manual cp fallback.
- docs/screenshots/hero-*.jpg: 8 fresh hero-only captures of each worked example (1440×900 via Chrome headless on the local dev server). Replaces the mid-video stills in _tests/_thumbs/*-still.jpg which had video-player overlay and cropped headlines. Each is unique — no duplicates. README "Different briefs, different shapes" section now uses these and each thumbnail is a clickable link to the live deployed example.
- site/index.html: the Study verb's brief-pill swapped from "/hallmark study this landing page" to "/hallmark study together.ai · or paste a screenshot" so URL mode is visually concrete next to the lede that already names both modes. One-line change.
- ROADMAP.md: new "Now — actively working on" section at the top. N.1 covers tightening the existing 22 themes + adding 4–6 new distinct ones + lowering the bar to route to custom. N.2 covers first-class Nanobanana image generation + a new image-led theme (working title: Plate) + token discipline for generated stills.
The frontmatter description opened with "Use this skill when ANY of these fire" — agent-facing trigger logic. Users typing /hall in the slash picker saw the truncated preview as "Use this skill when ANY of these fire — (1) the user names one of Hallmark's explicit verbs..." which says nothing about what Hallmark actually does.
Prepended a one-line user-facing summary so the picker preview reads as: "Anti-AI-slop design skill — generate distinctive landing pages, audit existing UIs, redesign templated pages, or study an admired URL / screenshot to extract its DNA." plus a usage hint with the four verbs. The full 5-condition trigger logic stays underneath unchanged so Claude's auto-fire behaviour is preserved.
Each .ex-card div is now an <a target="_blank" rel="noopener"> wrapping the same content. The hover-play video, prompt line, and .ex-card styles all work unchanged (the class already had color: inherit + focus-visible). Added one CSS line — text-decoration: none on .ex-card — so the <a> doesn't underline the card.
Linked destinations:
- examples/tally/ — SaaS, modern-minimal (from /Users/youssef/test-hallmark-9)
- examples/wayfare/ — travel booking, atmospheric (from test-hallmark-12)
- _tests/09-slow-pour/ — small-batch coffee subscription (already in site)
- examples/bananastudio/ — creative studio with playful motion (from test-hallmark-6)
- _tests/06-anya-portfolio/ — software architect personal site (already in site)
- examples/najm/ — Moroccan fashion (from test-hallmark-7)
- _tests/11-soroe-ceramics/ — small ceramics studio (already in site)
- examples/hyperlane/ — developer infrastructure (from test-hallmark-15)
site/examples/ adds ~308 KB total across the 5 copied projects. The other 3 already live under site/_tests/ so no copying needed for them.
When the user ran `study <URL>` and then said "build it", Hallmark fell back to a catalog theme (the "closest cousin" hint in the diagnosis) instead of using the studied DNA — paper/accent/fonts/macrostructure from the source were discarded.
- study.md: rewrote both diagnosis templates and the worked example. The "closest catalog theme" line is now framed as an explicit pivot option ("just say 'use X instead'"), not a recommendation. The "build it" handoff explicitly uses the studied DNA as the system.
- SKILL.md § 2.6: new Condition 0 (highest priority) — when a study diagnosis was emitted earlier in the conversation and the user asks to build from it ("build it" / "make it" / "use this DNA"), skip catalog/custom dispatch entirely. The DNA's tokens, macrostructure, and archetypes become the locked system. Diversification suspended for the build. Pivot phrases ("use Linen instead" / "ignore the DNA") route back to normal dispatch.
- SKILL.md Step 6: stamp's `theme:` field is `studied-DNA (source: <URL or image>)` with inline OKLCH/font values for studied builds.
- slop-test.md: new gate 69 auto-fails any build that names a catalog theme in its stamp when a recent study exists and the user didn't explicitly pivot.
- Gate count bumped 68 → 69 in SKILL.md + preview-examples.md.
- Bloom: .section__head was text-align: center with .section-label justify-content: center and .section__title margin-inline: auto — every section read as centered-everything (the AI fingerprint). Flipped to text-align: left, justify-content: flex-start, margin-inline: 0. Hero (.hero--marquee) stays centered via its own rules; the atmospheric glow does the centering work without needing every section head to repeat it. Tag stays above the title in the same column (still passes gate 66).
- Newsprint: removed border-top: 3px double var(--color-ink) from .section__head — the double rule above each section tag was reading as decorative noise on a layout that already has a strong 1 px bottom border. Single bottom border kept.
Two sticky-at-top:0 elements in the same stacking context (.banner and Almanac's .section__head) overlapped during scroll because they shared --z-sticky (200) and the section head paints last in DOM order. Visible on the marketing site as the section header drawing on top of the nav as you scroll.
site/ — actual fix for the live page:
- tokens.css: split --z-sticky into --z-sticky (200, in-page) + --z-sticky-nav (300, top nav); add --banner-height: 44px so secondary stickies can offset by a single token
- components.css: .banner now uses --z-sticky-nav and reads --banner-height for its height (kept at 44 px)
- sections.css: Almanac .section__head docks at top: var(--banner-height) instead of top: 0; z-index stays at --z-sticky so the banner always out-paints when their boxes momentarily overlap
skill/ — prevent the pattern in future generated pages (without forcing sticky on every output):
- slop-test.md: new gate 68 auto-fails any second sticky-at-top:0 element when a sticky page-level nav exists; fires only when the page actually has sticky elements (S3, F2, sticky TOCs), trivial pass otherwise
- components/s3-sticky-pinned.md: recipe rewritten to use top: var(--banner-height) + --z-sticky; pairing rule explained
- components/f2-sticky-scroll-stack.md: top now calc(var(--banner-height) + var(--space-xl)) so the sticky pane docks below the nav with breathing room
- SKILL.md + preview-examples.md: gate count bumped 67 → 68
Only Almanac uses sticky section heads in the live site — no other themes affected.
Manifesto and Sport themes set --lh-tight: 0.94 paired with text-transform: uppercase and condensed display faces (Anton / Inter Tight 900). On a two-line .section__title the cap-tops of line N+1 collide with the baseline of line N — most visible on the marketing site's "SAME PROMPT, TWO / DIFFERENT OUTPUTS." headline where the comma and the D fuse into one glyph.
- tokens.css: bump --lh-tight to 1.02 for both themes (safe floor for all-caps condensed display)
- slop-test.md: new gate 67 auto-fails any display-size element that pairs text-transform: uppercase with line-height < 1.0
- typography.md: explicit "Floor for all-caps display heads is 1.0 — recommended 1.02–1.08" line added to body-text rules
- SKILL.md + preview-examples.md: gate count bumped 66 → 67
- macrostructures.md (37 KB monolith) split into 21 per-macro files in references/macrostructures/; slim index lists names + summaries; SKILL.md Step 2 instructs to pick from index, then load ONLY the matching per-item file
- component-cookbook.md (55 KB monolith) split into 46 per-archetype files in references/components/; slim index lists names + summaries; SKILL.md Step 2 instructs to pick archetype codes, then load ONLY the matching files (typical build: 5–7 archetype files vs the whole cookbook)
- SKILL.md § 3 load-list rewritten: macros + components moved from eager-load to index-then-pick; slop-test.md gated to Step 7 only with explicit "do not pre-load" language; hero-enrichment.md no longer auto-loaded at Step 4
- 4 worked Step 5 preview blocks extracted from SKILL.md to references/preview-examples.md (lazy-loaded)
- gate count bumped 65 → 66 across SKILL.md (gate 66 was added in prior commit)
Expected token reduction per build: ~30–40 K (~30–40 % of the previous ~100 K-per-build cost), driven primarily by no longer loading 21 macros or 46 archetypes when only one of each is picked.
After a study diagnosis, the user can now say "lock the DNA" to write
a portable design.md from the extracted DNA — same format as the
default verb's lock-the-system flow, seeded from the studied source.
URL mode populates exact tokens + fonts; image mode populates
estimated bands + candidate fonts. URL-mode emission asks the user
to attest the source is theirs or a public reference for their own
brand — third-party URLs are refused at the emission step even when
the diagnosis itself was allowed. Image mode emits without asking.
Updates skill protocol, design-md trigger list, public verb table,
README comparison, and the site's Study row.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
URL mode reads HTML/CSS via WebFetch — names exact fonts and exact
colour values; trades the rhythm pass for everything else getting
more accurate. Auto-refuses template-marketplace URLs; falls back
to asking for a screenshot if the URL is auth-walled or an SPA shell.
Also drops a stale external reference from ROADMAP.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Banner Install hover-stick:
- Smooth-scroll on a sticky banner can leave the cursor over the
pill after click; Safari/Chrome occasionally keep :hover stuck
even after the pointer moves away.
- Scope the hover rule to (hover: hover) and (pointer: fine) so a
touch tap never sticks the inverted state.
- Add :focus:not(:focus-visible) reset so synthesised mouse-focus
doesn't keep the inverted look either.
- Add a JS click handler that blur()s the pill and toggles
pointer-events for one frame, forcing the browser to release any
sticky :hover/:focus state.
Atelier hero header:
- Quote copy: "Restraint, repeated, becomes a signature." → "A
workshop, not a template." Shorter, declarative, fits Atelier's
workshop voice.
- Override the hero quote to use the upright display face
(var(--font-display), normal style, --display-weight 900),
matching the "Atelier" wordmark in the banner and every section
title. Italic var(--font-serif) was making the header read as a
pull-quote rather than the page's primary statement; Atelier's
voice is declarative, not quoted.
With / Without rename:
- Section 06 label "Before / After" → "With / Without". Both panels
exist now (Sonnet 4.6 alone vs Sonnet 4.6 + Hallmark) — there is
no temporal before/after, only a presence/absence comparison.
- HTML comment updated to match.
Diptych screenshot centring:
- .ba__frame img object-position: top left → center. The full hero
composition reads inside the 16:10 frame instead of clipping to
the top-left corner. Other diptychs (.diptych__half, .brief-output)
intentionally stay top-aligned — they're page screenshots where
the masthead matters; the .ba__frame is hero-only.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The previous mobile fix (display: block + width: 100% on the button)
overflowed the frame because .code has white-space: nowrap and the
button rendered inline-flex on the same nowrap line as the command,
spilling past the right edge.
Cleaner pattern, matching what gh repo clone / npmjs / Vercel
install panes do on touch devices:
- Hide .code__copy entirely below 45rem.
- Bind a click handler on the whole [data-copy-source] pre so the
command area itself is tap-to-copy. cursor: pointer signals it.
- Add a transient "Copied ✓" badge inside the pre on success
(anchored top/right/centred-vertically) — replaces the button's
label-swap on mobile. Animates in and clears after 2.2 s.
- Restore the original .code flex + align-items: center so the
command sits vertically centred in the frame regardless of height.
JS:
- Refactor attachCopyButtons to share a copyFromSource helper.
- Source binding uses data-copy-bound; button binding uses
data-copy-btn-bound (separate flags so re-attached cloned heroes
don't double-bind).
- Button click stops propagation so a single tap on the visible
button (desktop) doesn't fire both the button and source handlers.
- Both surfaces flash data-state="copied" together so CSS feedback
works on whichever surface is visible at the active breakpoint.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The shared prompt in the Before/After Hallmark gallery used the old
Quiet Hour natural-wine brief — stale now that the before/after
thumbs were swapped to the DevForge dev-conference landing pages.
Updated the prompt to match the new content.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The before/after demo in the comparison gallery (Section 06) was
broken because the on-disk files had drifted from the HTML refs:
- HTML: <img src="_tests/_thumbs/before-quiet-hour.png" />
- Disk: before-quite-hours.jpeg (JPEG, typo "quite", plural "hours")
- HTML: <img src="_tests/_thumbs/after-quiet-hour.png" />
- Disk: after-quite-hour.png (typo "quite")
Re-encoded the before file from JPEG to PNG via sips and renamed
both to the correct singular "quiet-hour" spelling that the HTML
references. The Quiet Hour DevForge before/after now renders.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Hero install copy button:
- On coarse-pointer (touch) devices the copy button sits visible at
all times. With its absolute right-aligned position it overlapped
the npx command text on narrow viewports — most visible on
Specimen but applied to every theme using .code.
- Below 45rem, drop the absolute positioning and let the button flow
as a full-width row beneath the command (GitHub / Vercel / npmjs
pattern). The command keeps its horizontal scroll for long URLs.
Banner GitHub star spacing on mobile:
- The star uses a negative inline-start margin (-0.75rem) tuned for
the install/star pair on desktop. On mobile the install link is
hidden, so that same margin pulls the star too close to the
wordmark.
- Below 50rem, loosen the start margin to -0.25rem (+8 px breathing
room from the wordmark, matching the requested gap).
Specimen eyebrow:
- Drop " · Powered by Together AI" from the Specimen hero eyebrow
copy. The Together AI credit lives in the footer; repeating it in
the hero made the lede read like a sponsorship banner.
Footer bottom spacing:
- .page added padding-block-end: var(--space-4xl) (10 rem ≈ 160 px),
and .foot--colophon added another --space-2xl (4 rem ≈ 64 px) of
internal bottom padding. Combined ≈ 224 px of empty space below
"Made by Together AI" on every theme.
- Drop the page-wrapper bottom padding to 0 — the colophon owns its
own end-spacing.
- Cap the colophon's own bottom padding at --space-md (16 px) on
desktop and mobile. The credit line now hugs ≤ 20 px from the
viewport bottom.
Stale stats in footer (caught while in there):
- 23 themes → 22 (Pastel removed earlier).
- 60 slop-test gates → 65 (caught up to slop-test.md's actual count).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The "The page is the demo" interlude rendered with white-on-light text
on Coral, Violet, Plume, Editorial — and dark-on-dark on Aurora, Halo
— because .interlude__phrase em was reaching for --color-accent-ink.
That token is "text laid over the accent fill" (paper-coloured on
most themes), so it disappears on the paper-2 band.
Use --color-accent directly. The accent itself is mid-L on every
theme (40-78% L), so it has the right contrast against either light
or dark paper.
Same root cause caught in two more places:
- .diptych__half[data-state="after"] .diptych__cap (the small label
under the "after" diptych panel — same paper background)
- [data-theme="midnight"] .opening__title em (works visually on
Midnight because its accent and accent-ink are equal, but the token
usage was semantically wrong; aligning for consistency)
Six themes specifically affected by the visible bug: Coral, Violet,
Plume, Editorial (light, near-white em on light paper) and Aurora,
Halo (dark, near-black em on dark paper). All six now read clean.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Repo handover prep. Replaces:
- github.com/Luffixos/hallmark → github.com/nutlope/hallmark
(banner star link, footer link, JS REPO constant for star count)
- npx skills add hallmark → npx skills add nutlope/hallmark
(README install section, four install panes in site/index.html)
No skill internals touched — SKILL.md and references/ unchanged.
The local installed skill at ~/.claude/skills/hallmark/ continues
to work; the URL change is content-only.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Phrase-triggered design-system extraction. The user iterates freely on
the first builds; when the system is settled, they say "lock the system"
(or "give me a design.md", "make this portable") and Hallmark writes a
portable ~45-line design.md at the project root.
Why opt-in:
- Zero churn — no design.md flickering as the user swaps themes between
iterations.
- The locked system reflects a settled choice, not build #1's hot take.
- Mirrors how design teams actually work: formalise after patterns hold.
Mechanics:
- skill/references/design-md.md (new) — trigger phrases, scope rules,
the tight format template, no-overwrite policy.
- SKILL.md Step 3 — design-md.md added to load-conditionally.
- SKILL.md Step 5 — preview block carries one quiet CTA line. Skipped
on component-scope or when design.md already exists.
- SKILL.md Step 6 — opt-in design.md bullet alongside tokens.css and
multi-format-export rules. Default verb does NOT auto-emit. Multi-
page redesign keeps producing the heavyweight design.md (unchanged).
- README — one bullet under What's distinct.
Page-scope only; component-scope skips. Once design.md exists every
Hallmark run defers to it (existing pre-flight rule, unchanged) and
the diversification rule inverts to consistency.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Both favicon SVGs (dark + light) updated with hand-tuned path
coordinates for the Hn wordmark glyphs. No structural change to
the SVG; just nudged stroke positions.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The README and SKILL.md had drifted out of sync with the current
slop-test.md (now 65 gates) and the verb state (refine was removed;
audit · redesign · study + default = four verbs total).
- README: revert lingering "refine" references; verb count 5 → 4
- README: gate count 55/60 → 65 across body + side notes
- README: section title "Eight pages, eight different shapes" →
"Different briefs, different shapes" — the old phrasing implied
the skill could only produce these eight specific sites
- README: replace eight <video src=...mp4> tags with <img> stills
(first-frame extracts via ffmpeg from the existing .mp4 files;
no autoplay overhead, faster page render on the readme view)
- SKILL.md: drop "refine" from the frontmatter trigger description;
sync "55-gate" → "65-gate" across routing file + stamp examples
No version bump — pure documentation drift catching up to the
current rule-set state.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Description was firing on broad catch-alls — "design a", "build a
landing page", "make a dashboard", and the universal trap "any
request that will produce HTML / CSS / JSX / Tailwind output". Net
effect: Hallmark hijacked generic UI prompts that should have gone
to frontend-design or whatever else is loaded.
Tightened to four conditions:
1. User explicitly names a verb — audit · refine · redesign · study
2. User says a UI "feels AI-generated" / "looks templated" / "looks
like every other AI page", or asks to "make it less AI-generated"
3. User attaches a screenshot of a design they admire (study verb)
4. User invokes by name — "use Hallmark" or /hallmark
Added an explicit DO-NOT list — "design a button", "build me a form",
"add a navbar" — pointing those at frontend-design or other UI skills.
Hallmark stays proactive (matches the peer-skill convention) but the
phrase list is on-purpose, not catch-all. Skill discoverability
preserved; default-builder behaviour removed.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The previous wording said "almost always asks" with two carve-outs,
and the "user already named all three" exception let the model skip
the gate on five-word briefs by wishfully reading "podcast site" as
audience+use+tone. Removed that exception. The only silent verbs
are now audit / study / redesign --mood. Updated the prompt format
to match the canonical 1-2-3 layout (Audience · Use case · Tone)
with bolded labels. Added an explicit no-length-threshold clause:
short briefs get the same gate, the user waves through with "go ahead".
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Default to no section eyebrow. Use one only when the section is
genuinely ordinal (numbered manifesto, multi-step process, long
document chapters) or when the section name is too short to read
as a heading on its own. At most 1–2 per page.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Youssef