refine oh-my-frontend orchestration

.claude-plugin/marketplace.json

@@ -1,7 +1,7 @@
 {
   "$schema": "https://json.schemastore.org/claude-code-marketplace.json",
   "name": "fe-dudu",
-  "version": "0.2.1",
+  "version": "0.2.2",
   "description": "Frontend engineering and workflow orchestration Agent Skills for TypeScript and React",
   "owner": {
     "name": "fe-dudu"
@@ -32,8 +32,8 @@
     {
       "name": "oh-my-frontend",
       "source": "./skills/oh-my-frontend",
-      "description": "Optional frontend workflow and orchestration for project memory, approval, parallel work, specialist review, and risk-based verification.",
-      "version": "0.1.1",
+      "description": "Lightweight frontend triage and orchestration for risk classification, project memory, consequential decisions, safe parallel work, and runtime evidence.",
+      "version": "0.2.0",
       "author": {
         "name": "fe-dudu"
       },

README-ko.md

@@ -3,14 +3,14 @@
 [![skills.sh](https://skills.sh/b/fe-dudu/skills)](https://skills.sh/fe-dudu/skills)
 [![Claude Code Marketplace](https://img.shields.io/badge/Claude%20Code-Marketplace-8A63D2)](./.claude-plugin/marketplace.json)
 
-TypeScript, React, 협업형 프론트엔드 작업을 위한 두 계층 Agent Skills입니다. 코드 원칙만 필요하면 `frontend-engineering`을 단독으로 사용하고, 작업 오케스트레이션이 필요하면 `oh-my-frontend`를 추가합니다. 각 스킬은 짧은 `SKILL.md`를 제공하며, 상세 지침은 필요할 때만 reference에서 읽습니다.
+TypeScript, React, 협업형 프론트엔드 작업을 위한 두 계층 Agent Skills입니다. `oh-my-frontend`는 프론트엔드 task의 가벼운 triage 진입 계층이며, 명확한 기계적 변경은 Level 0으로 끝내고 위험이 있을 때만 memory, 승인, Worker, specialist evidence를 추가합니다. 구현 규칙은 `frontend-engineering`에 둡니다. 각 스킬은 짧은 `SKILL.md`를 제공하며, 상세 지침은 필요할 때만 reference에서 읽습니다.
 
 ## 스킬
 
 | 스킬 | 용도 |
 | --- | --- |
 | [frontend-engineering](./skills/frontend-engineering/) | 읽기 쉽고 명확하며 예측 가능한 TypeScript·React 구현 원칙입니다. |
-| [oh-my-frontend](./skills/oh-my-frontend/) | 선택형 프론트엔드 workflow·orchestration 계층입니다. |
+| [oh-my-frontend](./skills/oh-my-frontend/) | 프론트엔드 triage·위험 기반 orchestration 계층입니다. |
 
 <details>
 <summary><strong>frontend-engineering</strong> — 코드 원칙과 구현 품질</summary>
@@ -26,15 +26,15 @@ TypeScript, React, 협업형 프론트엔드 작업을 위한 두 계층 Agent S
 
 <details>
 <summary><strong>oh-my-frontend</strong> — 문서·승인·에이전트 오케스트레이션</summary>
-프로젝트 메모리, 승인, 병렬 작업, specialist review, 위험 기반 검증을 위한 선택형 프론트엔드 workflow·orchestration 계층입니다. 사람과 에이전트의 작업을 조율하며, `frontend-engineering`의 구현 원칙을 대체하지 않습니다.
+프론트엔드 task의 triage와 프로젝트 메모리, 승인, 병렬 작업, specialist review, 위험 기반 검증을 담당합니다. 모든 프론트엔드 task가 Level 0 분류를 위해 들어올 수 있지만, 변경 위험이 있을 때만 무거운 workflow를 추가합니다. `frontend-engineering`의 구현 원칙을 대체하지 않습니다.
 
-- Domain language, business rule, feature, decision, architecture를 `/docs`에 장기 메모리로 관리합니다.
-- Planner 대화, 필요한 승인, 범위가 제한된 Worker brief, 안전한 병렬 wave를 관리합니다.
+- 저장소의 canonical 구조에서 domain language, business rule, feature, decision, architecture를 관리하며, durable 문서 관례가 없을 때만 최소 구조를 추천합니다.
+- 관련 blocking question을 묶고, 필요한 승인, 범위가 제한된 Worker brief, 안전한 병렬 lane을 관리합니다.
 - Component architecture, UI state, browser·visual verification, accessibility, debugging을 다룹니다.
 - Data fetching, routing, performance, security, responsive UI, form, i18n, compatibility, observability를 라우팅합니다.
 - TDD나 과도한 테스트 코드를 강제하지 않고 위험 기반으로 테스트합니다.
 
-`docs/*ko/`의 한국어 mirror는 로컬 열람용이며 공개하거나 설치 패키지에 포함하지 않습니다.
+`docs/*.ko/`의 한국어 mirror는 로컬 열람용이며 공개하거나 설치 패키지에 포함하지 않습니다.
 
 [SKILL.md](./skills/oh-my-frontend/SKILL.md)부터 읽습니다.
 </details>

README.md

@@ -4,16 +4,18 @@
 [![Claude Code Marketplace](https://img.shields.io/badge/Claude%20Code-Marketplace-8A63D2)](./.claude-plugin/marketplace.json)
 
 Two-layer Agent Skills for TypeScript, React, and coordinated frontend work.
-Use `frontend-engineering` alone for code principles, or add `oh-my-frontend` as
-an orchestration layer. Each skill has a short `SKILL.md`; detailed guidance is
+Use `oh-my-frontend` as the lightweight triage entry for frontend tasks. It may
+classify mechanical work as Level 0 and adds memory, approval, workers, or
+specialist evidence only when risk requires them. Use `frontend-engineering` for
+implementation rules. Each skill has a short `SKILL.md`; detailed guidance is
 loaded from references only when needed.
 
 ## Skills
 
 | Skill | Use it for |
 | --- | --- |
 | [frontend-engineering](./skills/frontend-engineering/) | Standalone readable, explicit TypeScript and React implementation rules |
-| [oh-my-frontend](./skills/oh-my-frontend/) | Optional frontend workflow and orchestration layer |
+| [oh-my-frontend](./skills/oh-my-frontend/) | Frontend triage and risk-based orchestration layer |
 
 <details>
 <summary><strong>frontend-engineering</strong> — Code principles and implementation quality</summary>
@@ -31,18 +33,19 @@ Start with [SKILL.md](./skills/frontend-engineering/SKILL.md).
 
 <details>
 <summary><strong>oh-my-frontend</strong> — Documentation, approval, and agent orchestration</summary>
-Optional frontend workflow and orchestration for project memory, approval,
-parallel work, specialist review, and risk-based verification. It coordinates
-people and agents; it does not replace `frontend-engineering` implementation
-rules.
-
-- Durable `/docs` memory: domain language, business rules, features, decisions, and architecture
-- Planner conversation, focused approval, bounded Worker briefs, and safe parallel waves
+Frontend task triage and risk-based orchestration for project memory, approval,
+parallel work, specialist review, and verification. It may enter any frontend
+task, exits clear mechanical work at Level 0, and adds workflow only when the
+changed risk requires it. It does not replace `frontend-engineering`
+implementation rules.
+
+- Repository-canonical project memory: domain language, business rules, features, decisions, and architecture; recommend a minimal structure only when no durable convention exists
+- Batched blocking questions, focused approval, bounded Worker briefs, and safe parallel lanes
 - Component architecture, UI state, browser/visual verification, accessibility, and debugging
 - Data fetching, routing, performance, security, responsive UI, forms, i18n, compatibility, and observability routing
 - Risk-based testing without mandatory TDD or excessive test code
 
-Korean mirrors under `docs/*ko/` are local reading notes and are not published or included in installed packages.
+Korean mirrors under `docs/*.ko/` are local reading notes and are not published or included in installed packages.
 
 Start with [SKILL.md](./skills/oh-my-frontend/SKILL.md).
 </details>

Taskfile.yml

@@ -1,7 +1,6 @@
 version: '3'
 
 vars:
-  SKILLS_DIR: ./skills
   LINT_DIR: ./lint
   EVALS_DIR: ./evals
 
@@ -24,6 +23,8 @@ tasks:
       - ./skill-lint ../skills
     sources:
       - '*.go'
+      - 'go.mod'
+      - 'go.sum'
       - '../skills/**/*.md'
     generates:
       - skill-lint
@@ -35,6 +36,8 @@ tasks:
       - go build -o skill-lint .
     sources:
       - '*.go'
+      - 'go.mod'
+      - 'go.sum'
     generates:
       - skill-lint
 
@@ -51,16 +54,23 @@ tasks:
       - go build -o skill-eval .
     sources:
       - '*.go'
+      - 'go.mod'
       - 'evals.json'
     generates:
       - skill-eval
 
+  eval:test:
+    desc: Run static eval grader unit tests
+    dir: '{{.EVALS_DIR}}'
+    cmds:
+      - go test ./...
+
   eval:grade:
     desc: Grade a skill-creator workspace with static assertions
     deps:
       - eval:build
     cmds:
-      - '{{.EVALS_DIR}}/skill-eval --workspace {{.CLI_ARGS}}'
+      - '{{.EVALS_DIR}}/skill-eval {{.CLI_ARGS}}'
 
   check:
     desc: Run discovery, linter, and linter tests

evals/README.md

@@ -7,20 +7,22 @@ Static assertion grader for skill-creator eval workspaces. It applies the assert
 Prerequisites: [Task](https://taskfile.dev) and Go 1.25+.
 
 ```bash
-task eval:grade -- /path/to/workspace
-task eval:grade -- /path/to/workspace/iteration-1
+task eval:grade -- --skill oh-my-frontend --workspace /path/to/workspace
+task eval:grade -- --skill frontend-engineering --workspace /path/to/workspace/iteration-1
+task eval:test
 ```
 
 ## Adding cases
 
-Add cases under the `frontend-engineering` entry in `evals.json`.
+Add cases under the corresponding skill entry in `evals.json`. Eval IDs are
+scoped to a skill, so `--skill` is required when grading. The grader rejects
+workspace `eval-*` directories that are not defined for the selected skill.
 
 | Field | Required | Meaning |
 | --- | --- | --- |
 | `id` | yes | Numeric ID matching `eval-N` in the evaluation workspace |
 | `prompt` | yes | User-like task prompt |
 | `should_trigger` | no | Whether the skill should activate; defaults to `true` |
-| `expected_output` | no | Human-readable expected behavior |
 | `assertions` | no | Static checks against generated output |
 
 Supported assertions: