PABCD Initiative Documentation Hub pabcd_initiative codexclaw cli-jaw

Design Grammar Port: Hero Composition, Serif Renaissance, KR Serif, OpenAI Grammar

Source: devlog/260709_design_grammar_port.md

Design Grammar Port: Hero Composition, Serif Renaissance, KR Serif, OpenAI Grammar

Date: 2026-07-09

Kind: SoT reinforcement + downstream port (dev-frontend), 4 PABCD work-phases

Research: codexclaw devlog/_plan/260709_design_grammar_research/ (000 + 001),

5 GPT-5.5 explorers, Tier-2 opened sources + live computed-style measurements.

What changed (identical blocks in SoT, codexclaw, cli-jaw-skills)

WPFileAddition
1layout-discipline.md## Hero Composition Grammar (verified 2026-07-09) — split hero (left bold + right boxed mockup) demoted to paid-conversion-LP-only; replacement menu: centered-stacked-over-media, product-as-stage, editorial opener, evolved split w/ canvas, full-bleed consumer hero. Product visual = stage, never a right-column card.
1anti-slop.mdSplit-hero template tell (Stripe->Linear exhausted lineage).
2aesthetics.md### Serif Discipline (verified 2026-07-09) rewrite — domain-gated three-role system (display serif LIGHT 330-400 + sans UI + mono accent); AI/editorial/trust lane legitimate; dashboard ban + Fraunces/Instrument ban kept; counter-cases (Runway/Attio/Mistral) recorded.
2anti-slop.md"tasteslop" serif-shortcut tell (WIRED 2026-06-05).
3korea-2026.md### Korean Serif / Myeongjo Display (verified 2026-07-09) — MaruBuri/Noto Serif KR/Nanum/Chosun ranking, myeongjo display + Pretendard UI pairing, explicit no-KR-AI-wave disclaimer.
4aesthetics.md### Expressive vs Functional Layers + ### AI-Brand Grammar Vocabulary (OpenAI in-house Feb-2025 + ABC Dinamo + Studio Dumbar; Anthropic by Geist; DeepMind by MultiAdaptor/Colophon).
4motion.md## Soft-Focus Organic Background + Capsule Label (Level 5+) + ## Product-Led Hero Motion (Level 6+).
4asset-requirements.md## Mockup Production Pipeline (Rotato-class tools; AI scene + real screenshot compositing; real-UI-pixels rule).

jawcode: N/A (no dev-frontend skill family).

Addendum (same day): SKILL.md layer + dev-uiux-design reflection

WPFileAddition
5dev-uiux-design/references/design-isms.mdTwo new isms: AI Serif Editorial (three-role serif system, tasteslop gate) + Organic Capsule (OpenAI announcement grammar, layer split). Numbering adapted: codexclaw 1.14/1.15, SoT/cli-jaw 2.12/2.13.
5dev-uiux-design/references/product-personalities.mdOpenAI (2026 warm-sans organic) + Anthropic / Claude (serif bookish) DNA profiles.
6dev-uiux-design/SKILL.mdIsm/profile row counts updated; UX-TYPE-01 typography stance extended with the three-role serif system (added as a new stance bullet in SoT/cli-jaw, which lacked the line); MaruBuri Korean-serif-display bullet in Font Selection.
6dev-frontend/SKILL.mdHero-composition bullet in §5 Anti-Slop; 5 Modular References rows extended (layout-discipline, aesthetics, korea-2026, motion, asset-requirements).

Noted gap (out of scope, not fixed here): SoT/cli-jaw dev-uiux-design lag

codexclaw — no Liquid Glass ism (1.12), no Liquid Editorial kit (1.13), no

Aside profile, no UX-DEFAULT-ISM-01 section; their design-isms use a 2.x

numbering scheme. A separate backfill port would be needed to converge them.

Convergence addendum (same day): gap CLOSED

The backfill port ran as a 2-work-phase PABCD loop:

typography-line-breaks.md, design-read-example.md (newly created in cli-jaw),

ux-states.md byte-synced codexclaw -> SoT + cli-jaw (all six now

byte-identical x3; source files carry zero tool-specific mentions).

Pre-overwrite reverse-drift checks found only numbering/header variants

(2.x->1.x isms, 7-12->1-6 ux-states, "The Problem" rename) — no substantive

downstream content lost.

  • SKILL.md (section-additive, wholesale copy avoided): SoT gained

UX-DEFAULT-ISM-01 (No-Brief Default Direction) + the step-3 expressive-surface

pointer; cli-jaw gained Lazy-User Gate (UX-LAZY-01), UX State Contract

(UX-STATE-01), IA Chooser (UX-IA-01), UX-DEFAULT-ISM-01, and the UX-INTENT-01

compact ambiguity flow (its old 6-question Socratic intro was replaced; the

old Rules bullets survive inside "Optional deepening" except the "max 6

questions" cap, superseded by the ONE-fork rule). Ism/profile rows updated to

true counts (15 / 10). codexclaw's Native-tool-support paragraph and

frontmatter were deliberately NOT ported (agent-neutral).

Remaining known drift: minor wording differences in shared SKILL.md prose

(SoT keeps its richer Lazy-User phrasing; codexclaw keeps tool-specific

paragraphs). Capability level is now equal across the three repos.

dev-frontend convergence addendum (same day)

3-work-phase PABCD loop, criterion = capability parity (not byte parity where

structure deliberately differs):

  • WP1 backfill/syncs: color-system.md + liquid-glass.md added to SoT+cli-jaw;

preflight-full.md added to cli-jaw; theme-switching, examples/README,

layout-discipline, mobile-native synced to the newest variant (reviewer caught

two same-line-count SoT drifts: stale RN 0.84 content, old Bento phrasing);

scripts/README direction REVERSED after diff proof — SoT's §12-aligned version

was newest and now lives in all three.

  • WP2 forward ports: typography-wrapping + anti-slop byte-synced x3 (SoT's 16

unique anti-slop lines verified as phrasing variants whose substance exists in

the codexclaw version); motion.md full-synced (892L) with downstream-only

neutralizations — generic image-gen/image-to-video wording replaces the ima2

example, and codexclaw-local devlog paths became descriptive attributions.

  • WP3 reverse ports + SKILL.md parity: SoT's korea-2026 "Verified 2026

Additions" block ported to codexclaw+cli-jaw (korea-2026 now byte-identical

x3); react/perf moved-content ports to cli-jaw judged N/A (equivalent inline

sections exist); discoverability rows for color-system/liquid-glass added to

SoT+cli-jaw SKILL.md and preflight-full row to cli-jaw.

Accepted structural residuals: SoT SKILL.md keeps its 2026-07-02 refactor

(§8-10 consolidated into "React Behavior Rules", §6 connection budgets moved to

performance-budget.md) while codexclaw/cli-jaw keep the inline layout; motion.md

downstream copies differ from codexclaw only by the two neutralized spots.

Review trail

  • WP1: reviewer Curie (gpt-5.4) FAIL->fix->PASS (contradiction between paid-LP

exception and absolute rule — rule rescoped "Outside the paid-LP exception

above"; scope-creep suspicion rebutted with git evidence: Hangul bullet was a

pre-existing uncommitted 260708 port line).

  • WP2, WP3: Curie PASS, no blockers.
  • WP4 final gate: FRESH reviewer Hypatia (gpt-5.4) FAIL->fix->PASS (cross-ref

renamed to the exact heading Frame Sequence Format Guide); brand

attributions fact-checked against opened sources.

  • All WPs: blocks byte-identical across the 3 repos, insertions-only diffs

(rewrites accounted), code-fence parity.

Key sources

Wallpaper/Fast Company/Creative Review (OpenAI Feb-2025 rebrand), studiodumbar.com,

geist.co/work/anthropic, multiadaptor.com + colophon-foundry.org (DeepMind),

WIRED "AI Has Come for Serif Fonts" (2026-06-05), live computed styles claude.ai /

medium.com / manus.im / runwayml.com / attio.com / mistral.ai (2026-07-09),

LogRocket "Linear Design" (2026-02-03), Nordcraft, Rectangle "Linear effect",

live hero inspections of 11 sites (2026-07-09), hangeul.naver.com MaruBuri,

notofonts/noto-cjk, Apple Newsroom Liquid Glass (2025-06-09), WWDC25 session 356.