Release v0.5.7: Visual Craft and artifact checks

.github/scripts/validate_repository.py

@@ -16,6 +16,7 @@
     "screen-spec",
     "acceptance-verifier",
     "design-direction",
+    "visual-craft",
     "verified-delivery",
     "release-readiness",
     "handdrawn-diagram",

.github/workflows/release.yml

@@ -26,6 +26,11 @@ jobs:
         with:
           fetch-depth: 0
 
+      - name: Set up Node for rendered-detector fixtures
+        uses: actions/setup-node@v4
+        with:
+          node-version: "22"
+
       - name: Set up Python
         uses: actions/setup-python@v5
         with:

.github/workflows/validate.yml

@@ -21,6 +21,11 @@ jobs:
       - name: Check out repository
         uses: actions/checkout@v4
 
+      - name: Set up Node for rendered-detector fixtures
+        uses: actions/setup-node@v4
+        with:
+          node-version: "22"
+
       - name: Set up Python
         uses: actions/setup-python@v5
         with:

README.md

@@ -4,8 +4,8 @@
 
 **PRD·저장 가능한 작업 계획·요구사항 검증·Git 작업을 위한 선택형 스킬 모음.**
 
-![Version](https://img.shields.io/badge/version-0.5.6-6C5CE7?style=for-the-badge)
-![Skills](https://img.shields.io/badge/core_skills-9-00B894?style=for-the-badge)
+![Version](https://img.shields.io/badge/version-0.5.7-6C5CE7?style=for-the-badge)
+![Skills](https://img.shields.io/badge/core_skills-10-00B894?style=for-the-badge)
 ![Platform](https://img.shields.io/badge/platform-Codex-111827?style=for-the-badge)
 ![License](https://img.shields.io/badge/license-Apache--2.0-0984E3?style=for-the-badge)
 
@@ -58,6 +58,8 @@ Codex 앱에서는 Plugins를 열어 WIGTN 마켓플레이스를 선택한 뒤 
 ```text
 "OAuth 로그인 기능 PRD 뽑아줘"
 "이 PRD로 화면정의서 만들어줘"
+"프론트 UI 만들어줘"
+"이번 실험을 테크리포트 문서로 작성해줘"
 "요구사항이 실제 구현에 반영됐는지 검증해줘"
 "변경사항 검증하고 커밋해줘"
 ```
@@ -96,18 +98,21 @@ acceptance-verifier ── 요구사항 ↔ 코드·테스트 증거 매핑
 release-readiness ──── 안전한 커밋·푸시·PR
 ```
 
+UI·문서 제작에는 `visual-craft`가 구성과 문구를 보조합니다. 파일 생성·렌더링은 해당 형식의 워크플로를 따릅니다. 오타 수정이나 백엔드 작업에는 적용하지 않습니다.
+
 문서나 발표가 필요할 때는 `handdrawn-diagram`, `wigtn-presentation`을 독립적으로 사용할 수 있습니다.
 
 ---
 
-## 포함된 9개 Core 스킬
+## 포함된 10개 Core 스킬
 
 | 스킬 | 하는 일 | 자연어 호출 예시 | 호출 정책 |
 |---|---|---|---|
 | `product-spec` | 구현 가능한 PRD 생성, 리뷰, 딥다이브 | “PRD 뽑아줘” | 자동 |
 | `screen-spec` | 요청한 IA, User Flow, 화면 명세, self-contained lo-fi HTML 와이어프레임 또는 개발 핸드오프 | “화면정의서 만들어줘” | 자동 |
 | `work-planner` | 요구사항을 저장·재개 가능한 dependency-aware WorkGraph로 변환 | “이 계획을 WorkGraph로 저장해줘” | 자동·저장형 전용 |
 | `acceptance-verifier` | PRD 요구사항과 코드·실행 테스트 증거를 연결해 충족 여부 판정 | “요구사항 반영됐는지 검증해줘” | 자동 |
+| `visual-craft` | UI·문서 제작 시 정보 구조, 자연스러운 문구, 줄바꿈과 과도한 장식 검수 | “프론트 UI 만들어줘”, “문서 만들어줘” | 자동·제작 및 실질적 수정 |
 | `design-direction` | 기존 디자인 시스템을 먼저 읽고 신규 UI 방향 제안 | “이 UI 디자인 방향 잡아줘” | 자동·제한적 |
 | `verified-delivery` | 구현에 비례적인 requirement→code→executed-check 근거 추가 | `$verified-delivery로 구현해줘` | **명시 호출 전용** |
 | `release-readiness` | 변경 범위를 보존하며 검증, 커밋, 푸시, PR 수행 | “커밋해줘”, “PR 올려줘” | 자동·제한적 |
@@ -120,6 +125,20 @@ release-readiness ──── 안전한 커밋·푸시·PR
 
 ---
 
+## v0.5.7: Visual Craft와 자동 검사
+
+`visual-craft`를 추가해 Core 스킬이 9개에서 10개로 늘었습니다. UI·문서 제작과 실질적인 수정에 적용하며, 기존 9개 스킬의 호출 정책은 유지합니다.
+
+- 프론트 UI와 문서·보고서·발표자료 제작 요청에서 자동 선택을 허용합니다. `$wigtn-plugins-with-codex:visual-craft`로 명시할 수도 있습니다.
+- 공통 기준만 먼저 읽고, UI 배치와 문서 구성에 필요한 참조를 선택해 읽습니다. 기존 디자인 시스템·브랜드·출처·요구사항을 보존합니다.
+- 실제 산출물의 줄바꿈·과밀 구성·장식·가독성을 검수합니다. 모든 작업에 별도 디자인 계획이나 도구 설치를 요구하지 않습니다.
+- HTML·Markdown 소스 검사와 브라우저의 읽기 전용 배치 검사를 제공합니다. 확실한 오류와 문맥 검토 항목을 나누며, 검사 통과를 디자인 품질 점수로 취급하지 않습니다.
+- 자동 선택 허용과 정적 검증은 실제 모델의 호출 성공률이나 디자인 품질 개선을 보장하지 않습니다.
+
+[변경 내용과 검증 범위](docs/RELEASE-v0.5.7-KO.md) · [English release notes](docs/RELEASE-v0.5.7.md) · [설계 근거](docs/VISUAL-CRAFT.md)
+
+---
+
 ## v0.5.6: 요구사항 원문 보존과 단일 문서 범위 수정
 
 - 요구사항 importer가 `unit_price`, `ROUND_HALF_UP`, 곱셈 기호와 본문 내부 공백을 보존합니다. 서로 다른 본문을 같게 정규화해 이전 검증을 재사용하던 문제를 수정했습니다.

docs/RELEASE-v0.5.7-KO.md

@@ -0,0 +1,42 @@
+# v0.5.7: Visual Craft와 읽기 전용 자동 검사
+
+UI·문서·보고서·발표자료를 제작하거나 실질적으로 다듬을 때 구성, 정보 밀도,
+문구, 줄바꿈과 과도한 장식을 검토하는 `visual-craft`를 추가했습니다.
+Core 스킬은 9개에서 10개로 늘며, 기존 스킬과 호출 정책은 유지합니다.
+
+## 적용 방식
+
+공통 기준을 읽고 요청한 산출물에 맞는 UI 또는 문서 참조를 선택합니다.
+기존 브랜드·인용문·수치·요구사항을 보존하며, 파일 생성과 렌더링은 해당
+형식의 전문 워크플로를 따릅니다. 일반 대화·백엔드·오타 수정에는 개입하지 않습니다.
+`$wigtn-plugins-with-codex:visual-craft`로 명시하거나 자연어 제작 요청에서
+Codex가 선택하도록 사용할 수 있습니다.
+
+## 자동 검사
+
+- HTML·Markdown 검사: 중복 ID, 문서 언어, 내부 링크 대상, 입력 라벨,
+  제목 단계, 강제 줄바꿈과 반복 기호를 검사합니다. Python 표준 라이브러리만
+  사용하며 파일 수정·스크립트 실행·외부 요청을 하지 않습니다.
+- 브라우저 검사: 허용된 브라우저의 현재 화면에서 페이지 넘침, 잘린 글자,
+  작은 글자와 작은 클릭 영역을 측정합니다. 별도 브라우저 실행기나 훅은 추가하지 않습니다.
+- 확실한 소스 오류와 문맥 검토 항목을 구분합니다. 의도적인 말줄임,
+  동적 라우트, 인용문과 브랜드를 경고 제거 목적으로 바꾸지 않습니다.
+
+## 검증 범위
+
+릴리스 검증은 `bash scripts/validate.sh`로 실행합니다. 자동 검사 회귀 테스트
+17개와 그중 Node 기반 테스트의 합성 DOM 측정 조건 10개를 포함합니다.
+CI에 Node를 명시해 측정 조건 테스트를 실행하도록 했습니다.
+
+로컬 HTML 시안 3개는 소스 검사상 검출 항목이 없었습니다. 실제 브라우저
+통합 검증과 디자인 품질 비교는 아직 완료하지 않았습니다. 자동 선택 설정과
+키워드 예시는 실제 호출 성공률을 입증하지 않으며, 모델 성능·속도·토큰 개선을
+주장하지 않습니다. 자동 검사는 시각 검수나 전체 접근성 검사를 대체하지 않습니다.
+
+과거 full8/full9 실험의 스킬 구성은 유지합니다. Knowledge Wiki는 공동 버전
+정책에 따라 manifest 버전만 변경합니다.
+
+임시 개인용 `visual-craft`를 설치했다면 배포본과 비교해 로컬 수정 여부를
+확인한 뒤 백업하고 중복 등록을 정리하세요. 업데이트 후 새 Codex 작업을 시작하세요.
+
+[설계 근거와 출처](VISUAL-CRAFT.md)

docs/RELEASE-v0.5.7.md

@@ -0,0 +1,47 @@
+# v0.5.7: Visual Craft and read-only artifact checks
+
+UI and document creation can now use `visual-craft` to review hierarchy, density,
+natural copy, wrapping and decorative clutter. The Core catalog grows from nine
+to ten skills. Existing skills and their invocation policies remain available.
+
+## Artifact-specific guidance
+
+The skill supports frontend interfaces and authored documents, reports and slides.
+It loads UI or document references according to the requested artifact, respects
+existing brands and format-specific workflows, and preserves source quotations,
+data and requirements. It excludes ordinary chat, backend work and typo-only edits.
+Invoke it explicitly with `$wigtn-plugins-with-codex:visual-craft`, or let Codex
+select it for a matching creation or substantial refinement request.
+
+## Automatic checks
+
+- The Python standard-library checker reads explicit HTML/Markdown files and
+  reports duplicate IDs, language declarations, fragment targets, source-level
+  labels, heading structure and contextual copy patterns. It does not execute
+  embedded scripts, fetch links or edit files.
+- A read-only browser expression measures viewport overflow, clipped text, small
+  text and small targets within the current viewport. It uses an already-authorized
+  browser; no browser launcher, hook or third-party package is installed.
+- Confirmed source defects and contextual review findings are separate. Intentional
+  truncation, dynamic hash routes, quotations and brand marks must not be rewritten
+  simply to clear a warning. Findings are not aesthetic scores.
+
+## Validation and limitations
+
+- The release runs `bash scripts/validate.sh`, including 17 detector regression
+  tests and ten synthetic DOM geometry assertions inside the Node fixture test.
+  CI explicitly provisions Node so the geometry fixture can run.
+- Source checks of three local HTML samples found no findings. Real-browser
+  integration and visual quality comparison remain unverified. Geometry fixtures
+  do not replace browser testing, keyboard checks or a full accessibility audit.
+- Automatic selection metadata and lexical fixtures do not establish model routing
+  accuracy. No GPT-6 quality, speed or token-saving improvement is claimed.
+- Historical full8/full9 ablation catalogs retain their original skill membership;
+  they do not represent the new ten-skill package.
+- Knowledge Wiki receives only the matching version under the shared-version policy.
+
+If you installed the temporary personal `visual-craft` copy, compare it with the
+packaged skill and archive it after the plugin update, preserving any local edits.
+Start a new Codex task to pick up the updated plugin.
+
+[Design rationale and research sources](VISUAL-CRAFT.md)

docs/VISUAL-CRAFT.md

@@ -0,0 +1,101 @@
+# Visual Craft: scope and validation
+
+Release: v0.5.7. No model-quality or token-saving claim.
+
+## Design choice
+
+One implicitly discoverable skill supports frontend UI and authored documents.
+The entrypoint carries common decisions; UI and document guidance load by artifact.
+No third-party runtime dependency, hook, renderer, style menu or mandatory planning file is added.
+Read-only checks use Python and the already-authorized browser workflow.
+Existing format and product skills retain responsibility for their own outputs.
+The new skill does not force WIGTN branding on unrelated documents.
+
+The default excludes typo-only edits, ordinary chat and backend work. Direction-only
+requests still belong to design-direction; PRD and screen content contracts belong
+to their existing skills. A relevant craft pass can accompany those artifacts without
+starting a second workflow. Skill selection remains a model decision, not a keyword hook.
+
+## Research provenance
+
+Reviewed upstream main on 2026-09-22. These are independently written instructions
+informed by the following design approaches, not vendored upstream skill files.
+Upstream links may change after this review.
+
+- [Anthropic frontend-design](https://github.com/anthropics/skills/blob/main/skills/frontend-design/SKILL.md):
+  brief-specific design, content-led hierarchy, meaningful structural markers and plain copy.
+- [Taste-skill](https://github.com/Leonxlnx/taste-skill/blob/main/skills/taste-skill/SKILL.md):
+  concrete examples of decorative dots, broken headings and detached explanatory text.
+  Do not import its global font bans, motion defaults, punctuation replacement in quotes,
+  or forced asymmetric layouts.
+- [Impeccable](https://github.com/pbakaus/impeccable):
+  focused simplification, typography, layout and copy review with bounded visual checks.
+  Do not import its binary launcher, installer, hooks or whole command suite.
+
+Korean writing, source fidelity, publication formats and cooperation with existing
+WIGTN contracts are local additions. An em dash in a quotation, a status dot and the
+WIGTN logo period are intentionally different from decorative new copy.
+
+## Validation and practical limits
+
+Run `bash scripts/validate.sh` for packaging, resource links, explicit/implicit policy
+and repository regressions. The trigger fixtures are lexical examples only: changing
+their mapping does not establish actual Codex routing. No automatic visual-quality
+score is introduced. Historical full8/full9 experiment catalogs retain their old
+skill membership; neither represents the ten-skill package introduced in v0.5.7.
+
+Before claiming a quality improvement, compare generated artifacts with and without
+the skill using the same brief, assets, model and settings. Inspect rendered outputs,
+preserved content and task completion; include latency and input usage rather than
+rewarding deletion alone. Useful behavioral cases include:
+
+| Request | Observable outcome to inspect |
+|---|---|
+| Korean product landing page | Natural heading wraps at narrow/wide widths; functional CTA; no invented proof |
+| Dense operations dashboard | Required columns and actions remain usable; no marketing hero added |
+| Technical report from supplied measurements | Exact data, sources and limitations survive; no fabricated improvement |
+| Word report with a long table | Valid requested file; readable repeated headers and page breaks |
+| Generic slides | Requested deck format, no accidental WIGTN brand |
+| WIGTN deck | Approved brand and logo dot preserved; readable slides |
+| Existing legal text and quotations | Source text unchanged despite punctuation preferences |
+| Small CSS or typo fix | Only requested change, no redesign or new documents |
+
+These cases define a follow-up behavioral evaluation, not completed model runs.
+
+## Checks executed on 2026-09-22
+
+- Full `bash scripts/validate.sh`: PASS, including ten core skill manifests,
+  all resource links, 65 lexical routing examples and existing repository contracts.
+- `git diff --check`: PASS.
+- Personal installation of this skill only: validated with `quick_validate.py`.
+- Real Codex app-server `skills/list` with forceReload: one enabled user-scope
+  `visual-craft` entry, no related discovery errors. No model generation was run.
+- The entrypoint is 55 lines; the UI and document references are selected by task.
+  File length is an implementation fact, not a measured token saving.
+
+Before v0.5.7, local testing used a personal copy with the unqualified
+`$visual-craft` default prompt. After installing v0.5.7, archive that copy only after
+checking for local edits and confirming the packaged skill is enabled, so duplicate
+discovery does not remain. The plugin uses `$wigtn-plugins-with-codex:visual-craft`.
+
+## Automatic detectors added on 2026-09-23
+
+The skill now routes artifact completion through `references/automatic-checks.md`.
+`scripts/audit-source.py` checks explicit HTML/Markdown files and emits bounded JSON
+findings with rule, severity and source line. `scripts/audit-rendered.js` is a
+read-only expression for an existing authorized browser evaluation API. It reports
+viewport overflow and contextual clipping, small text and small target findings.
+Neither script installs a hook, fetches resources or modifies the artifact.
+
+Source checks deliberately do not claim browser visibility, full Markdown parsing,
+JSX support or document-format validation. Rendered checks cover only the current
+viewport, excluding iframe contents and shadow roots. Contrast, keyboard behavior,
+copy correctness and aesthetic quality are not certified. Browser permission blocks
+remain blocks; the detector is not an alternate browser launcher.
+
+Validation: 17 regression tests passed, including a Node VM fixture with ten DOM
+geometry assertions. Tests cover positive findings, protected quotations/code,
+native labels, inert templates, decoded fragment targets, bounded reporting,
+unsupported inputs and read-only operation. Geometry fixtures do not constitute a
+real-browser integration test. The three existing smoke-test HTML files returned
+no source findings; this is not a rendered or interaction pass.

plugins/wigtn-knowledge-wiki/.codex-plugin/plugin.json

@@ -1,6 +1,6 @@
 {
   "name": "wigtn-knowledge-wiki",
-  "version": "0.5.6",
+  "version": "0.5.7",
   "description": "Opt-in, gated Codex session knowledge capture for an explicitly authorized personal wiki namespace.",
   "author": {
     "name": "WIGTN"

plugins/wigtn-plugins-with-codex/.codex-plugin/plugin.json

@@ -1,15 +1,15 @@
 {
   "name": "wigtn-plugins-with-codex",
-  "version": "0.5.6",
-  "description": "Selective product contracts for Codex: compact PRDs, saved WorkGraphs, scoped screen artifacts, acceptance evidence, explicit verified delivery, and safe release actions.",
+  "version": "0.5.7",
+  "description": "Selective product contracts and visual craft for Codex: PRDs, saved WorkGraphs, screen artifacts, UI and document composition, acceptance evidence, and safe release actions.",
   "author": {
     "name": "WIGTN"
   },
   "skills": "./skills/",
   "interface": {
     "displayName": "WIGTN Plugin for Codex",
     "shortDescription": "Product contracts, evidence, and safe delivery.",
-    "longDescription": "Adds compact product contracts, saved WorkGraph planning, scoped screen artifacts, evidence integrity, and release authority boundaries while leaving ordinary Codex coding unconstrained. Verified delivery remains explicit-only.",
+    "longDescription": "Adds compact product contracts, saved WorkGraph planning, scoped screen artifacts, UI and document craft, evidence integrity, and release authority boundaries. Visual craft applies to artifact creation and substantial refinement; ordinary fixes stay lightweight. Verified delivery remains explicit-only.",
     "developerName": "WIGTN",
     "category": "Productivity",
     "capabilities": [],