arkaos 4.27.0 → 4.29.0

package/config/evals/marketing.yaml

@@ -1731,6 +1731,80 @@
   tags:
   - human-writing
   - imported
+- id: marketing-human-writing-07
+  department: marketing
+  prompt: 'Rewrite this launch-post paragraph: ''Here''s the thing: shipping fast isn''t
+    the problem. Shipping the wrong thing is. What if I told you the answer isn''t more
+    process? When the feedback loop tightens, the product improves. The implications are
+    significant. Not a framework. Not a methodology. A habit.'''
+  expected_properties:
+  - Identifies throat-clearing opener ("Here's the thing:")
+  - Identifies binary contrasts ("isn't the problem... is", "the answer isn't")
+  - Identifies rhetorical setup ("What if I told you")
+  - Identifies false agency ("the feedback loop tightens, the product improves")
+  - Identifies vague declarative ("The implications are significant")
+  - Identifies negative listing ("Not a framework. Not a methodology. A habit.")
+  - Rewrite states points directly and names the human actor
+  rubric: 'Should apply Rule 8 (No Formulaic Structures) and the structural-patterns
+    catalogue. Should flag every structural pattern in the paragraph: throat-clearing
+    opener, two binary contrasts, rhetorical setup, false agency (nobody tightens the
+    loop or improves the product in the sentence — the rewrite must name who does),
+    vague declarative, and negative listing. The rewrite should state each point as a
+    direct claim with a human subject, without introducing new AI patterns (no em
+    dashes, no fresh contrasts). Should not merely reword phrases while keeping the
+    structures.'
+  tags:
+  - human-writing
+- id: marketing-human-writing-08
+  department: marketing
+  prompt: 'Limpa este parágrafo de um email de lançamento: ''Vamos mergulhar no que
+    torna este produto verdadeiramente único. Não se trata de mais uma ferramenta,
+    trata-se de uma nova forma de trabalhar. Basicamente, a plataforma ganha vida
+    com os teus dados — e os resultados falam por si. As implicações são profundas.'''
+  expected_properties:
+  - Responds in European Portuguese (pt-PT)
+  - Identifies "Vamos mergulhar" as throat-clearing
+  - Identifies "Não se trata de X, trata-se de Y" as binary contrast
+  - Identifies stacked -mente adverbs ("verdadeiramente", "basicamente")
+  - Identifies false agency ("a plataforma ganha vida", "os resultados falam")
+  - Identifies em dash used as connector
+  - Identifies vague declarative ("As implicações são profundas")
+  - Rewrite is idiomatic pt-PT with named actors and direct claims
+  rubric: 'Should apply references/pt-pt-anti-slop.md and respond in pt-PT (never
+    pt-BR). Should flag: "Vamos mergulhar" (abertura de limpar a garganta),
+    "Não se trata de... trata-se de" (contraste binário), "verdadeiramente" and
+    "basicamente" (advérbios a cortar por defeito), "a plataforma ganha vida" and
+    "os resultados falam por si" (falsa agência — nomear quem faz), the em dash
+    connector (Rule 1), and "As implicações são profundas" (declarativa vaga).
+    The rewrite must be natural European Portuguese with direct statements and
+    human subjects, and must not translate English constructs literally.'
+  tags:
+  - human-writing
+- id: marketing-human-writing-09
+  department: marketing
+  prompt: 'Score this draft with the Slop Score rubric and tell me if it ships:
+    ''In today''s fast-paced landscape, teams need to navigate complexity. Here''s
+    why that matters: productivity isn''t about tools. It''s about focus. Really,
+    the data tells us everything. Make no mistake, this changes everything. The
+    stakes are high.'''
+  expected_properties:
+  - Applies the five Slop Score dimensions (Directness, Rhythm, Trust, Authenticity, Density)
+  - Scores each dimension 1-10 with a one-line justification
+  - Total is clearly below the 35/50 threshold
+  - Verdict is revise, not ship
+  - Names the specific offending patterns per dimension
+  - Offers a revised version or the concrete revision steps
+  rubric: 'Should apply the Slop Score rubric from the human-writing skill: rate
+    Directness, Rhythm, Trust, Authenticity, and Density 1-10 each with brief
+    justification tied to specific text (cliches "in today''s fast-paced" and
+    "navigate", emphasis crutches "Here''s why that matters" and "Make no
+    mistake", binary contrast "isn''t about tools. It''s about focus", false
+    agency "the data tells us", vague declarative "The stakes are high",
+    metronomic short sentences). The total must land clearly below 35/50 and the
+    verdict must be revise-before-shipping, with a rewrite or concrete steps.
+    Should not score generously to avoid the revision.'
+  tags:
+  - human-writing
 - id: marketing-customer-research-01
   department: marketing
   prompt: I have 20 customer interview transcripts. Help me analyze them.

package/config/skills-provenance.yaml

@@ -60,6 +60,20 @@
 baseline_date: 2026-07-14
 
 derived:
+  # Design absorption campaign W1 — slop gates, anti-patterns, critique
+  # protocol and registers derived from nutlope/hallmark (MIT, primary
+  # source in frontmatter) and pbakaus/impeccable (Apache-2.0, second
+  # source recorded in docs/THIRD-PARTY-NOTICES.md). Both skills were
+  # reclassified OUT of first_party deliberately: their bodies now carry
+  # substantial ported content.
+  departments/brand/skills/design-review:
+    origin: community
+    source: https://github.com/nutlope/hallmark
+    license: MIT
+  departments/brand/skills/ux-audit:
+    origin: community
+    source: https://github.com/nutlope/hallmark
+    license: MIT
   # Marketing skills integration campaign — derived from
   # coreyhaines31/marketingskills (MIT). Frontmatter metadata must match
   # each entry field-for-field (test_registry_matches_frontmatter_field_for_field).
@@ -272,7 +286,6 @@ first_party:
   - departments/brand
   - departments/brand/skills/archetype-finder
   - departments/brand/skills/colors
-  - departments/brand/skills/design-review
   - departments/brand/skills/design-system
   - departments/brand/skills/identity-system
   - departments/brand/skills/logo-brief
@@ -280,7 +293,6 @@ first_party:
   - departments/brand/skills/naming-evaluate
   - departments/brand/skills/positioning-statement
   - departments/brand/skills/primal-audit
-  - departments/brand/skills/ux-audit
   - departments/brand/skills/voice-guide
   - departments/brand/skills/wireframe
   - departments/community

package/departments/brand/agents/brand-director.yaml

@@ -66,6 +66,8 @@ expertise:
     - Nielsen Heuristics
     - Dieter Rams 10 Principles
     - Double Diamond
+    - Hallmark slop gates (58) + pre-emit critique (P/H/E/S/R/V)
+    - Impeccable design laws + registers (brand vs product)
   depth: master
   years_equivalent: 14
 

package/departments/brand/agents/design-ops/design-ops-lead.yaml

@@ -64,6 +64,8 @@ expertise:
     - Style Dictionary
     - shadcn/ui conventions
     - Tailwind CSS configuration
+    - Hallmark slop gates (58) + pre-emit critique (P/H/E/S/R/V)
+    - Impeccable design laws + registers (brand vs product)
   depth: master
   years_equivalent: 12
 

package/departments/brand/agents/design-ops/wcag-auditor.yaml

@@ -62,6 +62,8 @@ expertise:
     - EAA / EN 301 549
     - axe-core ruleset
     - Section 508
+    - Hallmark slop gates (58) + pre-emit critique (P/H/E/S/R/V)
+    - Impeccable design laws + registers (brand vs product)
   depth: master
   years_equivalent: 11
 

package/departments/brand/agents/ux-designer.yaml

@@ -70,6 +70,8 @@ expertise:
     - WCAG 2.1 AA
     - Two-Part Conversion Formula
     - Microinteractions (trigger-rules-feedback-loops)
+    - Hallmark slop gates (58) + pre-emit critique (P/H/E/S/R/V)
+    - Impeccable design laws + registers (brand vs product)
     - runtime:frontend-design (Anthropic skill)
   knowledge_sources:
     - "[[Area 02 - Design]]"

package/departments/brand/agents/ux-strategist.yaml

@@ -54,6 +54,8 @@ expertise:
     - Blue Ocean / Value Innovation
     - Provisional Personas
     - Lean UX
+    - Hallmark slop gates (58) + pre-emit critique (P/H/E/S/R/V)
+    - Impeccable design laws + registers (brand vs product)
   knowledge_sources:
     - "[[Personas/Jaime Levy]]"
     - "[[4 Elementos Centrais da UX Strategy]]"

package/departments/brand/agents/visual-designer.yaml

@@ -57,6 +57,8 @@ expertise:
     - Brand Identity Process (Wheeler)
     - ArkaOS Design Tokens
     - WCAG AA contrast
+    - Hallmark slop gates (58) + pre-emit critique (P/H/E/S/R/V)
+    - Impeccable design laws + registers (brand vs product)
     - runtime:frontend-design (Anthropic skill)
     - runtime:ui-ux-pro-max (community skill)
   depth: expert

package/departments/brand/references/uiux-knowledge-and-tools.md

@@ -6,7 +6,8 @@
 >
 > Read this BEFORE any UI/UX work. It defines the non-negotiable KB-first
 > rule, the canonical knowledge sources, the concrete design tokens, the
-> motion system, the new tooling, and the squad validation order.
+> motion system, the new tooling, the squad validation order, the canonical
+> reflex-reject lists (§11) and the hard design laws (§12).
 
 ---
 
@@ -132,7 +133,10 @@ write material findings back to `[[Area 02 - Design]]`:
 - Dieter Rams' 10 principles
 - Atomic Design (full atoms→pages)
 - Microinteractions (trigger→rules→feedback→loops/modes)
-- Color theory & typography science (OKLCH, harmony, modular scale)
+- Color theory & typography science (OKLCH, harmony, modular scale) —
+  **partially filled in-repo**: hard floors/ceilings now live in §12 and
+  the reflex-reject discipline in §11; deeper craft references live under
+  `departments/brand/skills/design-review/references/`.
 - Gestalt & visual hierarchy (F/Z scanning, proximity, contrast)
 
 ---
@@ -168,6 +172,23 @@ for any other brief?"* If yes, revise and state what changed and why.
 Quality floor, never announced: responsive to mobile, visible keyboard
 focus, `prefers-reduced-motion` respected, WCAG AA contrast.
 
+**Pre-emit critique (scored — hallmark protocol).** Before handing back
+any design artifact, score it 1–5 on six axes:
+
+| Axis | Question |
+|---|---|
+| **P** Philosophy | Does the design have a stated POV the brief earns? |
+| **H** Hierarchy | Does the eye land where the content says it should? |
+| **E** Execution | Are spacing, contrast, states, and breakpoints finished? |
+| **S** Specificity | Does this look like *this brief* — or a generic page? |
+| **R** Restraint | Is every bold move deliberate; is everything else quiet? |
+| **V** Variety | Does it differ structurally from the previous output? |
+
+Any axis **< 3 triggers a revision pass** — never ship and explain.
+Stamp the six scores in the artifact via the `[arka:design-dna]` companion
+stamp (§9). The 58 pass/fail slop gates that operationalize this rubric
+live in `departments/brand/skills/design-review/references/slop-test.md`.
+
 ## 9. Design Marker Contract (`[arka:design]`)
 
 Every UI/design-producing turn MUST emit the structured marker on its own
@@ -192,6 +213,21 @@ A bare `[arka:design] <token>` is a LEGACY marker: tolerated in WARN mode,
 counted separately in telemetry, and treated as missing once the frontend
 gate goes hard.
 
+**Companion stamp `[arka:design-dna]` (informational).** Design-producing
+work additionally stamps the first non-empty line of produced CSS with:
+
+```
+/* [arka:design-dna] macrostructure=<name> genre=<genre> anchor=<oklch|hex> display=<font> body=<font> critique=P#H#E#S#R#V# */
+```
+
+Key=value, single line, greppable. It records the structural choice, the
+palette anchor, the type pairing, and the six pre-emit critique scores
+(§8), and feeds the structural-diversification rule (consecutive outputs
+must differ on at least one axis; project memory at
+`.arka/design/log.json`). This stamp is **NOT parsed by the frontend gate**
+(`core/workflow/frontend_gate.py`) and carries no gate semantics — the
+structured `[arka:design]` marker above remains the only enforced line.
+
 ## 10. Aesthetic Directions Catalog (optional, curated)
 
 When a project has NO design system yet and the brief leaves the visual
@@ -207,3 +243,150 @@ job: load `gsap-core` + `gsap-timeline` for any animation work,
 `gsap-scrolltrigger`/`gsap-plugins` as the work demands, `gsap-react` on
 React, and treat `gsap-performance` as the review bar
 (`npx skills add https://github.com/greensock/gsap-skills` when absent).
+
+## 11. Canonical Reflex-Reject (fonts + aesthetic lanes)
+
+> Derived from [impeccable](https://github.com/pbakaus/impeccable)
+> (Apache-2.0) and [hallmark](https://github.com/nutlope/hallmark) (MIT) —
+> license texts at `departments/brand/skills/design-review/references/`.
+> This section is the SINGLE canonical home of the list. Skills and gates
+> point here; they never restate it (drift risk).
+
+These are **training-data defaults**: every LLM reaches for them by
+reflex, which is exactly why using them by reflex produces monoculture.
+The semantics are **never-by-reflex**, not never:
+
+1. **Identity-preservation carve-out** — when the existing brand has
+   already committed to a font or lane, identity wins. The list governs
+   greenfield decisions only.
+2. **Deliberate-pick carve-out** — a listed font may be used when a named
+   theme or brief *deliberately* selects it and the choice is stated with
+   its reason (e.g. Playfair Display as display-only in an editorial
+   theme; Space Grotesk in a brutalist direction). Reflex is the crime,
+   not the font.
+3. **Product-register carve-out** — product UI (design SERVES the task)
+   has explicit permission for system fonts and familiar sans defaults
+   (Inter, SF Pro, system-ui). The ban list bites on BRAND surfaces
+   (design IS the product). See
+   `departments/brand/skills/design-review/references/design-registers.md`.
+
+### Reflex-reject fonts
+
+**Sans:** Inter · DM Sans · Space Grotesk · Outfit · Plus Jakarta Sans ·
+Instrument Sans · Syne · Roboto · Open Sans · Lato · Poppins · Source Sans ·
+Nunito · Montserrat · Raleway · Work Sans · system-ui-as-brand · Arial ·
+Helvetica-as-default
+**Serif:** Fraunces · Newsreader · Lora · Crimson / Crimson Pro / Crimson
+Text · Playfair Display · Cormorant / Cormorant Garamond · DM Serif Display ·
+DM Serif Text · Instrument Serif · Merriweather · Source Serif ·
+Georgia-as-default
+**Mono:** IBM Plex Mono · Space Mono · Courier New · Consolas-as-default ·
+system-mono-as-brand
+**Family-wide:** IBM Plex Sans · IBM Plex Serif
+
+Selection discipline when rejecting a reflex pick: run the 4-step
+physical-object font procedure in `design-registers.md` (three concrete
+brand-voice words → reject reflex picks → browse a real catalog for the
+brand as a physical object → cross-check).
+
+### Reflex-reject aesthetic lanes
+
+- **Editorial-typographic** — display serif (often italic) + small mono
+  labels + ruled separators + monochromatic restraint. The 2026 fingerprint:
+  three rule-separated columns, an italic serif headline, lowercase
+  track-spaced metadata, no imagery. Every Stripe-adjacent and
+  Notion-adjacent brand has landed here. Requires a register reason that
+  *demands* it (a literal magazine), not a mood.
+
+Lanes saturate and desaturate over time — entries move in and out on
+evidence, mirroring upstream. The **second-order slop check** applies: if
+the aesthetic family is guessable from category-plus-anti-references
+("fintech that's not navy-and-gold → terminal-native dark mode"), it is
+the trap one tier deeper. Rework until neither the first-order nor the
+second-order answer is obvious.
+
+## 12. Design Laws (hard floors and ceilings)
+
+> Derived from [impeccable](https://github.com/pbakaus/impeccable)
+> (Apache-2.0) and [hallmark](https://github.com/nutlope/hallmark) (MIT).
+> Compact canonical table — one law per line. The enforcement surfaces
+> (slop gates, design-slop detector, QG visual review) cite these; deeper
+> rationale lives in `design-review/references/`.
+
+**Typography**
+- Body contrast ≥ 4.5:1; large text (≥18px, or bold ≥14px) ≥ 3:1;
+  placeholders need 4.5:1 too — muted-gray-on-tinted-near-white is the
+  single biggest readability failure.
+- Prose line length 65–75ch (dense data UI may run denser).
+- Hero/display ceiling: `clamp()` max ≤ 6rem. Above that the page shouts.
+- Display letter-spacing floor ≥ −0.04em — tighter and letters touch.
+- `text-wrap: balance` on h1–h3; `text-wrap: pretty` on long prose.
+- Max 3 font families per page (display + body + one outlier). Four = slop.
+- Pair on a contrast axis (serif+sans, geometric+humanist) or use one
+  family with committed weight contrast — never two similar sans.
+- Light-on-dark compensation runs on three axes: line-height +0.05–0.1,
+  letter-spacing +0.01–0.02em, optionally body weight one notch up.
+- Commit to extremes: weight 200 next to 800 reads as intentional;
+  400 next to 600 reads as a default setting.
+
+**Color**
+- OKLCH for new palettes. Pure gray is dead: tint neutrals 0.005–0.015
+  chroma toward the brand hue (never default-tint warm/cool).
+- The warm-neutral band (L 0.84–0.97, C < 0.06, hue 40–100) is the
+  saturated AI default — cream/sand/paper by any token name
+  (`--paper`, `--cream`, `--sand`, `--linen`, `--parchment`…) is a tell.
+  Warmth is carried by accent + typography + imagery, not body bg.
+- Pick a color strategy BEFORE colors, on the commitment axis:
+  **Restrained** (neutrals + one accent ≤10%) · **Committed** (one
+  saturated color, 30–60% of surface) · **Full palette** (3–4 named
+  roles) · **Drenched** (the surface IS the color).
+- Gray text on colored backgrounds washes out: use a darker shade of the
+  background's own hue or a transparency of the text color.
+- Button text vs fill must differ by > 5% lightness or > 0.05 chroma in
+  OKLCH (catches the black-on-black bug).
+- Heavy alpha/transparency is a design smell — it usually means an
+  incomplete palette.
+
+**Layout**
+- Nested cards are always wrong; cards only when truly the best affordance.
+- Flexbox for 1D, Grid for 2D; breakpoint-free grids via
+  `repeat(auto-fit, minmax(280px, 1fr))`.
+- Semantic z-index scale (dropdown → sticky → modal-backdrop → modal →
+  toast → tooltip); never 999/9999.
+- Vary spacing for rhythm: generous separations, tight groupings —
+  monotonous spacing is a tell.
+- Touch targets ≥ 44×44 (extend hit area via `::before { inset: -10px }`
+  when the visual must stay smaller).
+
+**Motion**
+- Ease out with exponential curves (ease-out-quart/quint/expo); no bounce,
+  no elastic, no browser-default `ease`.
+- Exits run ~75% of the enter duration.
+- `prefers-reduced-motion` alternative is mandatory for every animation.
+- Never animate CSS layout properties without real need; motion conveys
+  state on product surfaces, never decoration.
+- The uniform fade-and-rise reveal on every scrolled section is a tell —
+  each reveal fits what it reveals; zero motion is not the fix either.
+- Reveal animations enhance an already-visible default — never gate
+  content visibility on a class-triggered transition (ships blank in
+  hidden tabs and headless renderers).
+
+**Interaction**
+- Every interactive element ships all 8 states: default, hover,
+  focus-visible, active, disabled, loading, error, success. Missing any =
+  unfinished.
+- Skeletons for loading, not spinners in content; empty states teach the
+  interface.
+- Modal as first thought is laziness — exhaust inline/progressive
+  alternatives first.
+
+**Absolute bans (match-and-refuse — rewrite the element, never ship):**
+side-stripe borders (`border-left/right` > 1px as accent) · gradient text
+(`background-clip: text`) · glassmorphism-as-default · the hero-metric
+template (big number + small label + gradient accent) · identical
+icon+heading+text card grids · tiny uppercase tracked eyebrow above every
+section · numbered section markers (01/02/03) as default scaffolding ·
+text overflowing its container at any breakpoint.
+
+These laws are floors, not aspirations: §8's three default looks plus any
+absolute-ban hit are constitution violations under `excellence-mandate`.

package/departments/brand/skills/design-review/SKILL.md

@@ -13,6 +13,10 @@ description: >
   dev/clean-code-review wins; trying to break flows or find abuse vectors ->
   dev/adversarial-review wins.
 allowed-tools: [Read, Write, Edit, Bash, Grep, Glob, Agent]
+metadata:
+  origin: community
+  source: https://github.com/nutlope/hallmark
+  license: MIT
 ---
 
 # Design Review — `/brand design-review`
@@ -41,6 +45,12 @@ Do this BEFORE reviewing, in this order, and record what actually loaded:
 3. **Read the brand guidelines / design system** — the review compares
    against the project's OWN tokens (`design-system.yaml`, brandbook,
    or §3 of the squad reference as fallback).
+4. **Load the review references in this directory** —
+   `references/design-registers.md` (pick the register FIRST: brand vs
+   product changes what "good" means), `references/critique-protocol.md`
+   (two-isolated-subagent orchestration for deep reviews),
+   `references/slop-test.md` + `references/anti-patterns.md` (the
+   mechanical gates and the named-tell dictionary used below).
 
 ### Graceful degradation (honest, never silent)
 
@@ -102,11 +112,26 @@ For every captured surface, annotate deviations with exact values
 - **Benchmark contrast** — for each major finding, one line on what the
   named benchmark does instead.
 
+## Slop gates (pre-verdict — MANDATORY)
+
+Before issuing any verdict, run the reviewed surfaces through
+`references/slop-test.md`: 58 pass/fail gates (every answer must be
+**no**) plus the six-axis pre-emit rubric (P/H/E/S/R/V, squad reference
+§8). Anti-pattern findings are named from
+`references/anti-patterns.md` with their Critical/Major/Minor severity —
+never described vaguely. When produced CSS carries the
+`[arka:design-dna]` stamp, verify the recorded critique scores against
+what the screenshots actually show; a stamp that flatters the artifact
+is itself a finding.
+
 ## Verdict
 
 `ON-BRAND` / `DEVIATIONS (n)` / `OFF-BRAND` — with the annotated
 screenshot list, exact-value deviation table, and fix list ranked by
 visual impact. A review with zero captured screenshots is INVALID.
+`OFF-BRAND` and every rejected surface MUST cite the failed slop-gate
+numbers and named anti-patterns (e.g. "gates 12, 41; purple-gradient
+hero [Critical]") — verdicts on evidence, never on vibes.
 
 ## Output