| name | yoda |
| description | Learning-science-based code review and educational content generation. /yoda review for Before/After/Why reviews, /yoda share to spread to a team in 4 formats (md/web/wiki/slides). 코드 리뷰, 코드 분석, 코드 설명, 팀 공유, 온보딩, 기술 세미나, 지식 전파 요청 시 활성화 |
| argument-hint | [review <target> | share <input> --format md|web|wiki|slides --audience junior|mid|senior] |
| model | opus |
user: "이 ViewModel 리뷰해줘"
assistant: "/yoda review 모드로 6단계 파이프라인을 실행하여 Before/After/Why 기반 3계층 리뷰를 수행하겠습니다."
user: "이 코드를 팀에 공유할 마크다운 문서로 만들어줘"
assistant: "/yoda share --format md 모드로 Worked Examples와 이중 코딩 원칙을 적용한 교육 마크다운 문서를 생성하겠습니다."
user: "방금 리뷰한 내용으로 세미나 슬라이드 만들어줘"
assistant: "/yoda share --from-review --format slides 모드로 인지적 도제 모델링을 적용한 15장 Marp 프레젠테이션을 생성하겠습니다."
user: "Swift Concurrency 패턴을 주니어한테 설명하는 위키 글 써줘"
assistant: "/yoda share --format wiki --audience junior 모드로 4막 내러티브 구조와 높은 비계 수준의 위키 페이지를 생성하겠습니다."
Yoda
Unified skill for learning-science-based code review and educational content generation.
Where conventional code review focuses on "What" (what is wrong), yoda also addresses "Why" (why it is wrong, why it should be fixed this way). It turns code review into a learning experience through Before/After/Why triples, 3-layer progressive disclosure, curiosity triggers, ZPD calibration, and per-format learning-science principles.
Respond to the user in Korean. All output is written in Korean.
Reference documents
Read the references below before running, so the rules they define are applied consistently.
review-reference.md — Review mode 6-phase pipeline, 3-layer output structure, severity classification detailsshare-reference.md — Share mode per-format generation strategy, --from-review pipeline, ZPD calibration detailslearning-science.md — 4 core principles, per-format additional principles, curiosity trigger rules, ZPD calibration matrix
Mode selection
Analyze keywords in the user input to auto-select the mode.
| Mode | Keywords | Description |
|---|
| review | 리뷰, 분석, 검토, 코드 봐줘, 코드 확인, 점검 | Code analysis + 3-layer learning-science-based report generation |
| share | 공유, 전파, 세미나, 슬라이드, 위키, 문서, 온보딩, 마크다운 | Educational content generation (supports 4 formats) |
4 core principles
Apply to all output. See learning-science.md for detailed rules.
| # | Principle | Application |
|---|
| 1 | Before/After/Why triple | For every finding, place the problem code (Before) + fixed code (After) + reason (Why) adjacently. Start Why with a curiosity trigger |
| 2 | Signaling/labeling | Attach a severity label (🔴🟡🔵🟢💡) + cognitive error type ([Slip]/[Rule]/[Knowledge]/[Lapse]) to every finding |
| 3 | Coherence -- remove the unnecessary | Remove what the reader already knows, information unrelated to the finding, decorative phrasing, and disclaimers |
| 4 | Chunking | One section/slide/card = one concept. Keep total findings to 7 or fewer |
| 5 | Specification grounding | Ground findings in concrete tests/contracts/specifications. Label speculation-based findings as "확인 필요" |
| 6 | Growth framing | Frame as a learning opportunity rather than a defect. Use normalizing phrasing + growth language |
Review mode
Entry point
/yoda review <파일|디렉토리>
- Targets a single source file or a directory path.
- For a directory, recursively traverse the source files underneath and review them all.
Internal pipeline (6 phases)
| Phase | Name | Summary |
|---|
| 1 | Context collection | Read the target code + git log/blame + explore tests/dependencies + infer author intent (run implicitly) |
| 2 | Multi-perspective analysis | Independent analysis through 5 lenses: structure/clarity/safety/performance/tests. Collect findings exhaustively first (no cap during analysis); apply the top-N selection (up to 3 per lens, 7 or fewer total) only at the output stage |
| 3 | Finding structuring | Structure each finding as a Before/After/Why triple (curiosity trigger → principle link → real impact) |
| 4 | Severity + error type classification | 🔴🟡🔵🟢💡 severity + [Slip]/[Rule]/[Knowledge]/[Lapse] cognitive error type tags |
| 5 | Mental model visualization | Generate one Mermaid diagram only when an architecture change is needed (conditional) |
| 6 | Metacognition prompt | Stimulate thinking with 2 questions on scalability/change-readiness + transfer/application |
Output: 3-layer progressive disclosure
| Layer | Target reader | Time required |
|---|
| Layer 1: Key summary | Busy lead, quick check | 30s |
| Layer 2: Detailed analysis | Code author, reviewer | 5-10 min |
| Layer 3: Deep insight | Developer with the will to learn | +5 min |
See review-reference.md and templates/md-template.md for detailed pipeline rules and output examples.
Share mode
Entry point
/yoda share <입력> [--format md|web|wiki|slides] [--audience junior|mid|senior]
Input types
| Input | Detection method | Handling |
|---|
--from-review | Check for flag presence | Extract the previous review result from the conversation and convert it |
| File/directory | Verify path existence with Glob | Run independent analysis, then generate content |
| Free text | None of the two above apply | Explore the codebase with Grep, then generate based on real examples |
Per-format generation & delegation
If --format is not given, present the format choices to the user and ask.
| Format | Generation method | Reinforced principles |
|---|
| md | Generate directly into docs/yoda/YYYY-MM-DD-[slug].md with the Write tool | Worked Examples (Sweller) + dual coding (Mayer) + elaboration questions (Dunlosky) |
| web | Generate interactive HTML directly, or delegate to the frontend-design skill | Progressive disclosure (Nielsen) + retrieval practice (Roediger & Karpicke) + personalization (Mayer) |
| wiki | Generate wiki markdown with the Write tool. Publish directly when Confluence MCP is connected | Storytelling (4-act narrative) + social constructivism (Vygotsky) + elaboration questions |
| slides | Generate 15 Marp markdown slides directly with the Write tool | Signaling reinforcement + coherence maximization + cognitive apprenticeship modeling (Collins et al.) |
확장: pptx 스킬이 설치되어 있으면 --format pptx로 발표자 노트 포함 PowerPoint를 생성할 수 있다. 미설치 시 이 옵션은 표시하지 않는다.
ZPD calibration (--audience)
Following Vygotsky's Zone of Proximal Development (ZPD) theory, calibrate content per audience level. Defaults to mid when unspecified.
| audience | Before/After/Why adjustment | Terminology handling |
|---|
| junior | Explain Why in detail. Place rich comments in the After code | Include term explanations. Spell out full names alongside abbreviations |
| mid | Standard Before/After/Why triple. Name the principle + explain core impact | Use shared team vocabulary |
| senior | Turn Why into a question (prompt thinking without giving the answer) | Free use of abbreviations/jargon |
See share-reference.md and templates/ for detailed per-format generation strategies and the --from-review pipeline.
Usage
/yoda review ChatRoomViewModel.swift
/yoda review Sources/Feature/Chat/
/yoda share --from-review --format md
/yoda share --from-review --format slides
/yoda share --from-review --format slides
/yoda share Sources/Feature/Chat/ --format wiki --audience junior
/yoda share "Swift Concurrency 에러 처리" --format web
/yoda review UserProfileViewModel.swift
/yoda share --from-review --format slides
Limitations
- Static analysis only; code is not executed.
- Severity classification involves qualitative judgment.
- For new files without git history, some steps of context collection (Phase 1) are skipped.
--format pptx is available only when the pptx skill is installed.
