Skip to content

Anti-AI-slop design skill for Claude Code, Cursor, and Codex.

CSS

CI runs refcheck on every pull request; package.json and the old verb notes catch up

Youssefa8242c2695bb
diff --git a/.github/workflows/verify.yml b/.github/workflows/verify.yml
new file mode 100644
index 0000000..01de1fb
--- /dev/null
+++ b/.github/workflows/verify.yml
@@ -0,0 +1,21 @@
+name: verify
+on:
+ push:
+ branches: [main]
+ pull_request:
+jobs:
+ verify:
+ runs-on: ubuntu-latest
+ steps:
+ - uses: actions/checkout@v4
+ - uses: actions/setup-node@v4
+ with:
+ node-version: 20
+ - name: References, gates, fixtures, versions and links agree
+ run: node skills/hallmark/scripts/refcheck.mjs
+ - name: Every theme validates
+ run: node skills/hallmark/scripts/theme.mjs validate skills/hallmark/references/themes/*.md
+ - name: The theme index is current
+ run: |
+ node skills/hallmark/scripts/theme.mjs index
+ git diff --exit-code -- skills/hallmark/references/themes/index.json
diff --git a/package.json b/package.json
index 642c7a6..73b3dfb 100644
--- a/package.json
+++ b/package.json
@@ -1,23 +1,21 @@
{
"name": "hallmark",
- "version": "1.2.0",
+ "version": "2.4.0",
"description": "A design skill for AI coding assistants. Makes the UIs they generate look made, not generated. Powered by Together AI.",
"keywords": [
"claude",
"claude-code",
- "cursor",
"codex",
+ "opencode",
+ "agent-skills",
"skill",
"design",
+ "redesign",
"ui",
"anti-ai-slop",
"typography",
"oklch",
- "together-ai",
- "variants",
- "open-models",
- "glm",
- "kimi"
+ "together-ai"
],
"license": "MIT",
"type": "module",
@@ -29,8 +27,8 @@
"references": "skills/hallmark/references",
"harnesses": [
"claude-code",
- "cursor",
- "codex"
+ "codex",
+ "opencode"
]
},
"scripts": {
diff --git a/site/_tests/verbs/audit/audit-report.md b/site/_tests/verbs/audit/audit-report.md
index b021dce..520f326 100644
--- a/site/_tests/verbs/audit/audit-report.md
+++ b/site/_tests/verbs/audit/audit-report.md
@@ -4,9 +4,9 @@
## Verbs loaded
-- [`anti-patterns.md`](../../../skills/hallmark/references/anti-patterns.md) — the named-tells library to score against
-- [`structure.md`](../../../skills/hallmark/references/structure.md) — for the structural-fingerprint check (gate 9)
-- [`microinteractions.md`](../../../skills/hallmark/references/microinteractions.md) — the timing/easing canon
+- `anti-patterns.md` — the named-tells library to score against
+- `structure.md` — for the structural-fingerprint check (gate 9)
+- [`microinteractions.md`](../../../../skills/hallmark/references/microinteractions.md) — the timing/easing canon
(No write paths. `audit` does not edit.)
diff --git a/site/_tests/verbs/redesign/notes.md b/site/_tests/verbs/redesign/notes.md
index 5abf161..867ff83 100644
--- a/site/_tests/verbs/redesign/notes.md
+++ b/site/_tests/verbs/redesign/notes.md
@@ -69,4 +69,4 @@ The `studied: no · context: redesign` flag tells the next Hallmark run that the
The audit recommended `redesign` over `refine` because the page's structural fingerprint *was* the slop. `refine` would have left the centered-hero + 3-card + gradient-CTA shape intact — fixing the colour tokens and easings, but the page would still look AI-generated. `redesign` was the right call.
-If the user had said "I love the shape, just fix the colours" → `refine` (see [`../refine/`](../refine/)). If they say "throw it out, keep my words" → `redesign` (this page). The two verbs have surgical-vs-restructural roles, and the audit decides which.
+If the user had said "I love the shape, just fix the colours" → `refine` (see `../refine/`). If they say "throw it out, keep my words" → `redesign` (this page). The two verbs have surgical-vs-restructural roles, and the audit decides which.
diff --git a/site/_tests/verbs/study/diagnosis.md b/site/_tests/verbs/study/diagnosis.md
index 5768736..fc25558 100644
--- a/site/_tests/verbs/study/diagnosis.md
+++ b/site/_tests/verbs/study/diagnosis.md
@@ -4,10 +4,10 @@
## References loaded
-- [`study.md`](../../../skills/hallmark/references/study.md) — the vision-extraction protocol + structured-fields schema + refusal heuristics
-- [`study-examples.md`](../../../skills/hallmark/references/study-examples.md) — three worked examples (Pentagram-style portfolio, Klim-specimen, Rauno-personal)
-- [`macrostructures.md`](../../../skills/hallmark/references/macrostructures.md) — to name the page-shape
-- [`component-cookbook.md`](../../../skills/hallmark/references/component-cookbook.md) — to name the hero / pitch / footer archetypes
+- `study.md` — the vision-extraction protocol + structured-fields schema + refusal heuristics
+- `study-examples.md` — three worked examples (Pentagram-style portfolio, Klim-specimen, Rauno-personal)
+- [`macrostructures.md`](../../../../skills/hallmark/references/macrostructures.md) — to name the page-shape
+- [`component-cookbook.md`](../../../../skills/hallmark/references/component-cookbook.md) — to name the hero / pitch / footer archetypes
## Diagnosis (10 structured fields)
@@ -33,7 +33,7 @@
## Anti-patterns the screenshot has — do NOT carry over
-> 1. The hover-state on the number-label uses a slow ease-in-out 800 ms colour fade. Per [`microinteractions.md`](../../../skills/hallmark/references/microinteractions.md) § The timing canon, hover state should be 150–200 ms. **Carry the *idea* (colour-shift on hover); shorten the *duration*.**
+> 1. The hover-state on the number-label uses a slow ease-in-out 800 ms colour fade. Per [`microinteractions.md`](../../../../skills/hallmark/references/microinteractions.md) § The timing canon, hover state should be 150–200 ms. **Carry the *idea* (colour-shift on hover); shorten the *duration*.**
>
> 2. The footer text is set at 11 px with 0.06 em tracking — under the 14 px floor for body copy. **Bump to 12–13 px on rebuild, keep the tracking.**
diff --git a/site/_tests/verbs/study/notes.md b/site/_tests/verbs/study/notes.md
index cc95350..4c01650 100644
--- a/site/_tests/verbs/study/notes.md
+++ b/site/_tests/verbs/study/notes.md
@@ -8,23 +8,23 @@ A described screenshot (`input-description.md`) of a Pentagram-style design stud
> *"Extract the DNA — macrostructure, archetypes, type-pairing role, colour anchor, rhythm — and produce a diagnosis report, then optionally rebuild the user's content using the extracted DNA. Never copies pixels. Never claims to identify exact fonts. Refuses obvious template-marketplace screenshots."*
-Five-step pipeline (per [`SKILL.md`](../../../skills/hallmark/SKILL.md) § `hallmark study`):
+Five-step pipeline (per [`SKILL.md`](../../../../skills/hallmark/SKILL.md) § `hallmark study`):
1. **Refuse-or-proceed check** — public reference, not a paid template → proceed.
-2. **Vision pass** — read the image into the structured-fields schema in [`study.md`](../../../skills/hallmark/references/study.md). Output ten fields.
+2. **Vision pass** — read the image into the structured-fields schema in `study.md`. Output ten fields.
3. **Diagnosis report** — return the named macrostructure + archetypes + anti-patterns to NOT carry over. See [`diagnosis.md`](diagnosis.md).
4. **Confirmation question** — ask before building.
5. **Build** — produce `output.html` + `output.css` with the user's actual six projects, stamped `studied: yes`.
## What loaded
-- [`references/study.md`](../../../skills/hallmark/references/study.md) — the protocol, schema, and refusal heuristics.
-- [`references/study-examples.md`](../../../skills/hallmark/references/study-examples.md) — the three worked examples (Pentagram-style portfolio, Klim-specimen, Rauno-personal). The Pentagram example matched our screenshot most closely; the skill imitated the diagnosis structure.
-- [`references/macrostructures.md`](../../../skills/hallmark/references/macrostructures.md) — to name the page-shape (Specimen).
-- [`references/component-cookbook.md`](../../../skills/hallmark/references/component-cookbook.md) — for archetype names (H4 Margin-Number Display, Ft1 Single-Line Colophon).
-- [`references/typography.md`](../../../skills/hallmark/references/typography.md) — for the type-role vocabulary (italic-display + sans-body) and free / paid candidate pairings.
-- [`references/color.md`](../../../skills/hallmark/references/color.md) — for the warm-orange-at-low-chroma colour-anchor.
-- [`references/microinteractions.md`](../../../skills/hallmark/references/microinteractions.md) — for the hover-duration anti-pattern callout (the screenshot's 800 ms is too slow; should be 150–200 ms).
+- `references/study.md` — the protocol, schema, and refusal heuristics.
+- `references/study-examples.md` — the three worked examples (Pentagram-style portfolio, Klim-specimen, Rauno-personal). The Pentagram example matched our screenshot most closely; the skill imitated the diagnosis structure.
+- [`references/macrostructures.md`](../../../../skills/hallmark/references/macrostructures.md) — to name the page-shape (Specimen).
+- [`references/component-cookbook.md`](../../../../skills/hallmark/references/component-cookbook.md) — for archetype names (H4 Margin-Number Display, Ft1 Single-Line Colophon).
+- `references/typography.md` — for the type-role vocabulary (italic-display + sans-body) and free / paid candidate pairings.
+- `references/color.md` — for the warm-orange-at-low-chroma colour-anchor.
+- [`references/microinteractions.md`](../../../../skills/hallmark/references/microinteractions.md) — for the hover-duration anti-pattern callout (the screenshot's 800 ms is too slow; should be 150–200 ms).
## What `study` did *not* do
@@ -58,4 +58,4 @@ The screenshot maps neatly to study-examples.md Example A (Pentagram-style edito
- The confirmation question.
- The `studied: yes` stamp.
-If the input had been a paid Squarespace-template marketplace screenshot, the verb would have refused. If the input had been ambiguous (an unknown agency page or a personal portfolio with no clear ownership), the verb would have asked before extracting. Both branches are documented in [`study.md`](../../../skills/hallmark/references/study.md); this test exercises the proceed-with-public-reference branch.
+If the input had been a paid Squarespace-template marketplace screenshot, the verb would have refused. If the input had been ambiguous (an unknown agency page or a personal portfolio with no clear ownership), the verb would have asked before extracting. Both branches are documented in `study.md`; this test exercises the proceed-with-public-reference branch.
diff --git a/skills/hallmark/scripts/refcheck.mjs b/skills/hallmark/scripts/refcheck.mjs
index e10b362..6b41dd2 100644
--- a/skills/hallmark/scripts/refcheck.mjs
+++ b/skills/hallmark/scripts/refcheck.mjs
@@ -317,6 +317,25 @@ for (const [name, ok, msg] of existsSync(FIX) ? expectations : []) {
else if (!ok(j)) add('error', target, 1, 'fixture', msg);
}
+const REPO = resolve(ROOT, '..', '..');
+const pkgFile = join(REPO, 'package.json');
+if (existsSync(pkgFile)) {
+ let pkg = null;
+ try { pkg = JSON.parse(read(pkgFile)); } catch {}
+ if (pkg && pkg.name === 'hallmark') {
+ if (pkg.version !== VERSION) add('error', pkgFile, 1, 'version', `package.json says ${pkg.version} where SKILL.md says ${VERSION}`);
+ const docs = ['README.md', 'ROADMAP.md', ...(existsSync(join(REPO, 'docs')) ? readdirSync(join(REPO, 'docs')).filter((n) => n.endsWith('.md')).map((n) => join('docs', n)) : [])].map((n) => join(REPO, n)).filter(existsSync);
+ for (const f of docs) {
+ const src = read(f);
+ for (const m of src.matchAll(/\]\(([^)\s]+)\)/g)) {
+ const t = m[1].split('#')[0].split('?')[0];
+ if (!t || /^(https?:|mailto:|data:)/.test(m[1])) continue;
+ if (!existsSync(resolve(dirname(f), t))) add('error', f, lineAt(src, m.index), 'link', `${m[1]} does not resolve`);
+ }
+ }
+ }
+}
+
const shown = ONLY_ERRORS ? findings.filter((x) => x.level === 'error') : findings;
const errors = findings.filter((x) => x.level === 'error').length;
const warns = findings.length - errors;