tobyilee

Use this skill whenever the user wants to back up or restore their global Claude Code configuration under ~/.claude — global CLAUDE.md, settings.json, user-level commands/skills/subagents, installed plugin marketplaces, and install manifests. Trigger on "claude backup", "backup claude", "backup my claude config", "snapshot claude", "restore claude", "claude 백업", "클로드 백업", "claude 복구", "claude 복원", "설정 백업", "글로벌 설정 백업", "내 claude 설정 저장", "claude 설정 복원", "plugin 백업", "설치된 플러그인 백업", "백업으로 되돌려", "이전 백업으로 복원", "원래 상태로 되돌려", "backup ~/.claude", "restore my plugins", "migrate claude to another machine". Also trigger when the user is about to do something risky to their global Claude config (big settings edit, plugin overhaul, reinstalling Claude Code) and mentions wanting a snapshot/checkpoint first, or when they mention losing plugins and wanting the previous state back. Do NOT trigger for project-level backups, git stashes, or generic "backup my repo" — this skill is specifically for the ~/.claude user directory.

This skill should be used when the user asks to "create a prd", "write prd for", "plan this feature", "write requirements", "spec out a feature", "product requirements document", "feature spec", "define user stories", "PRD 작성", "PRD 만들어", "PRD 써줘", "요구사항 문서", "기능 명세", "기능 정의서", "프로덕트 요구사항", "스펙 문서 만들어", "유저 스토리 작성", or needs guidance on structured feature planning and requirements documentation. Do NOT trigger for ad-hoc TODO lists, design docs (use a separate design-doc workflow), or sprint-planning tickets — this skill produces a single PRD per feature.

Use this skill to delegate tasks to OpenAI Codex CLI running in the background. Trigger on "delegate to codex", "send to codex", "ask codex to", "run with codex", "have codex do", "codex에게", "codex로", "코덱스에게", "코덱스로", "use codex for", "let codex handle", "codex한테 시켜", "codex로 돌려봐". Also trigger when the user wants a second opinion from another AI specifically mentioning Codex or OpenAI, wants to compare Claude's approach with Codex, or asks to run a task with an external OpenAI agent in parallel. Do NOT trigger for general "delegate" or "외부 AI" requests without mentioning Codex/OpenAI — those may be better handled by gemini-delegate or other tools.

Use this skill to delegate tasks to Google Gemini CLI running in the background. Trigger on "delegate to gemini", "send to gemini", "ask gemini to", "run with gemini", "have gemini do", "gemini에게", "gemini로", "제미나이에게", "제미나이로", "use gemini for", "let gemini handle", "gemini한테 시켜", "gemini로 돌려봐". Also trigger when the user wants a second opinion from another AI specifically mentioning Gemini or Google, wants to compare Claude's approach with Gemini, or asks to run a task with an external Google agent in parallel. Do NOT trigger for general "delegate" or "외부 AI" requests without mentioning Gemini/Google — those may be better handled by codex-delegate or other tools.

Use this skill when the user wants to develop features using Test-Driven Development with an agentic Red-Green-Refactor cycle. Trigger on "start TDD", "do TDD", "TDD로 개발해줘", "TDD 시작", "TDD 팀 만들어", "테스트 주도 개발", "red green refactor", "test-driven development", "테스트 먼저 작성하고 싶어", "write tests first then implement", "테스트부터 짜줘", "TDD 방식으로 구현해줘". Also trigger when a user describes a feature and says they want it built incrementally with tests, e.g. "이 기능 테스트 먼저 만들고 하나씩 구현하자", "build this with failing tests first", or "한 단계씩 테스트 작성하면서 개발하고 싶어". Do NOT trigger for simply writing unit tests after implementation, running existing tests, or debugging test failures — those are not TDD workflows.

Use this skill to send a task to Codex running in a cmux pane and get the result back as a file. Trigger on "toby codex", "toby codex에게", "toby codex한테", "toby codex로", "toby codex에게 해줘", "toby codex한테 시켜", "toby codex로 보내", "toby codex에게 물어봐". Only trigger when the user explicitly says "toby codex" — the "toby" prefix is required to distinguish from codex-delegate. Do NOT trigger on bare "codex" mentions without the "toby" prefix. This skill communicates with a persistent Codex pane in cmux (started by toby-team-starter) and collects the result via a file. Do NOT trigger for codex-delegate (which launches a new codex process in the background).

Use this skill to send a task to Gemini running in a cmux pane and get the result back as a file. Trigger on "toby gemini", "toby gemini에게", "toby gemini한테", "toby gemini로", "toby gemini에게 해줘", "toby gemini한테 시켜", "toby gemini로 보내", "toby gemini에게 물어봐". Only trigger when the user explicitly says "toby gemini" — the "toby" prefix is required to distinguish from gemini-delegate. Do NOT trigger on bare "gemini" mentions without the "toby" prefix. This skill communicates with a persistent Gemini pane in cmux (started by toby-team-starter) and collects the result via a file. Do NOT trigger for gemini-delegate (which launches a new gemini process in the background).

Use this skill to spawn a multi-agent workspace with Codex and Gemini running alongside Claude in cmux. Trigger on "toby team", "toby team 시작", "toby team 준비", "toby team 시작해", "toby team 준비해", "toby 팀 시작", "toby 팀 준비", "agent team", "에이전트 팀", "codex랑 gemini 같이", "start agent team", "팀 시작", "멀티 에이전트 시작", "codex gemini 같이 띄워", "에이전트 팀 만들어", "spawn agents", "launch agents", "에이전트 pane 열어", "agent pane", "start multi-agent", "codex gemini pane". Also trigger when the user wants Codex and Gemini running side-by-side in terminal panes for parallel work, or asks to set up a multi-agent development environment. Do NOT trigger when the user just wants to delegate a single task to codex or gemini — use codex-delegate or gemini-delegate for that. This skill is specifically for spawning persistent agent panes in cmux.

Integrate all finished chapters into a single book-ready manuscript — polish chapter transitions, unify terminology, insert cross-chapter callbacks, and write front/back matter (preface, epilogue, references). Produce a manuscript + book_manifest.json for EPUB build. Use when chapters are individually complete and need to become a coherent book.

Orchestrate a full book-writing workflow from topic to finished EPUB with cover image. Use when the user asks to "write a book", "draft a book", "author a book", "책 쓰기", "책 저술해줘", "책 만들어줘", "전자책 만들어줘", "EPUB 생성", provides a topic/audience/outline and asks to turn it into a book, or says "~에 대한 책을 써줘". Also triggers on follow-ups like "다시 저술", "계획 수정", "챕터 다시 써줘", "표지 바꿔", "책 업데이트", "특정 챕터 보완", "이전 책 개선", "리서치만 다시". Coordinates research, planning, review, chapter writing in the active genre voice (auto-detected; defaults to tech-book = Toby's style), style enforcement, editing, cover design, and EPUB assembly. Author defaults to Toby-AI (overridable via user prompt — include "저자: {이름}").

Write a single book chapter in Toby's distinctive Korean writing style (평어체 기반, 청유형 적극 활용, 수사적 질문, 상황 가정, 감정적 공감 표현). Use when drafting or revising a book chapter, rewriting prose in Toby's voice, translating an outline into full chapter text, or polishing chapter drafts to match 토비 문체. Triggers on "챕터 써줘", "이 섹션 저술", "토비 스타일로 써줘", "write a chapter in Toby's voice", "rewrite in 평어체", or when a chapter outline needs to become full prose.

Build a standards-compliant EPUB 3 file from a manuscript markdown, cover image, and book_manifest.json using the bundled pandoc-based script. Output is 책-제목-v{version}.epub with the manifest's author (defaults to Toby-AI), accompanied by a paired 책 소개 markdown ({책-제목}-v{version}.md) in the same folder. Use when converting the final book manuscript to EPUB, rebuilding after edits, or testing EPUB output.

Verify concrete factual claims in a tech-book chapter draft — numbers, statistics, benchmarks, quotes, version numbers, release years, API signatures, superlatives — against the reference document, and resolve "(사실 확인 필요)" markers. Use when checking a chapter for factual accuracy, validating version/API claims in fast-moving tech content, or resolving fact-check annotations. Triggers on "팩트체크", "사실 확인", "fact check", "출처 검증", "버전 맞는지 확인".

Whole-book acceptance gate run by a fresh-context reviewer before EPUB build (Phase 4.5). Cross-reads the integrated manuscript against the plan and escalation logs, scans for forbidden leftover markers, checks terminology/cross-reference/promise integrity, and emits {slug}/05_acceptance.md with per-criterion PASS/BLOCK + routing. Genre-agnostic; complements (does not replace) per-chapter fact/continuity loops. Triggers on "수락 검수", "통권 검수", "manuscript acceptance", "최종 통과 게이트".

Coordinate parallel research agents (web, paper, community) and synthesize findings into a unified reference document. Use when collecting background material for a book, whitepaper, or deep-dive article — triggers on "리서치 조율", "여러 소스에서 자료 모아줘", "reference 문서 만들어줘" style requests. De-duplicates, groups by theme, flags conflicts, preserves citations.

Review chapter drafts against the active genre profile's style checklist (defaults to tech-book = Toby's voice) — detects voice deviations and provides concrete before/after rewrite suggestions. Use when reviewing a chapter draft for style consistency, enforcing the genre voice, or polishing prose.

Final build step — render all Marp slides, validate SSML, synthesize the course manifest, and package bundle.zip. Use when coherence review has passed and the course is ready for deployment. Triggered by the asset-builder agent.

Orchestrate end-to-end online course generation from a topic. Invoke when user asks to "create a course", "design a curriculum", "build online lectures", "강의 만들어", "코스 설계", "커리큘럼 짜", "섹션별 퀴즈 포함 강의", "TTS 스크립트 강의", or any request to produce structured learning material with slides + notes + scripts + quizzes. Also invoke for follow-ups — "재실행", "섹션만 다시", "업데이트", "이전 결과 개선", "코스 수정". Do NOT invoke for single-artifact requests (just a slide, just a quiz) — those should call the specific authoring skill directly.

Synthesize TTS audio from course transcripts with pluggable engines (OpenAI gpt-4o-mini-tts default, edge-tts fallback). Use when turning a class transcript.txt into per-slide MP3s plus a concatenated full.mp3, handling [slide N] / [pause:ms] markers, retries, and per-slide speaker_affect overlays. Triggered after coherence-review passes and before/within asset-build, or as a standalone step. Also invoke for "오디오 생성", "TTS 합성", "강의 음성 만들어", "voice 변경", "speed 조정".

Write TTS-ready transcripts with [slide N] cues, [pause:ms] markers, and optional SSML. Convert raw code/URLs/long numbers to speakable forms. Use when turning slides+beats into a narrator script. Triggered by the script-writer agent.

Author slide decks in Marp Markdown and render to HTML. Use when turning class beats into 4-7 slides per class, applying Marp frontmatter/layout rules, preventing overflow, or rendering via the marp CLI. Triggered by the slide-author agent.

Cross-artifact QA for the course pipeline. Verify LO coverage, Bloom balance, slide↔script cue alignment, note↔slide consistency, tone, quiz factual validity, and transcript speakability. Use when reviewing generated course assets before build. Triggered by the coherence-reviewer agent.

Plan a single class as a beat sheet (hook → teach → example → practice → recap) used jointly by slide, note, and script authors. Use when turning a class spec into beats, allocating duration across beats, or assigning visual/speaker cues. Triggered by the class-planner agent.

Design course-level and section-level curriculum scaffolding. Use when writing Learning Objectives, choosing Bloom's Taxonomy levels, splitting a topic into sections, or breaking sections into classes. Triggered by curriculum-architect and section-designer agents.

에이전트 팀의 Closed-Loop Communication(폐쇄 루프 커뮤니케이션)을 강제하는 스킬. SendMessage로 작업 핵심 정보를 주고받을 때 발신→수신확인(ACK)→발신검증의 3단 루프를 보장한다. 팀원 간 핸드오프, 작업 지시, 인터페이스 질의, 의존 결과 요청 시 반드시 사용. '핸드오프', 'ACK', '메시지 확인', '전달 보장', '오해 방지' 관련 작업에 트리거. Team Big Five 조율 메커니즘 2.

에이전트 팀의 Mutual Performance Monitoring(상호 성과 모니터링)과 Backup Behavior(백업 행동)를 수행하는 스킬. 팀원 산출물을 Shared Mental Model 기준으로 교차 검증하고, 경계면 불일치·목표 이탈을 감지하며, 정체·과부하·실패한 팀원을 백업한다. 'QA', '교차 검증', '경계면 점검', '정합성 확인', '팀원 백업', '재할당', '드리프트 감지' 작업에 반드시 트리거. Team Big Five 구성요소 2·3.

에이전트 팀의 Shared Mental Model(공유 정신 모델)을 _workspace/SMM.md 파일로 구축·유지하는 스킬. 팀이 시작할 때, 팀원 간 산출물이 어긋날 때, 용어/인터페이스/목표를 합의해야 할 때 반드시 사용. 'SMM', '공유 정신 모델', '단일 진실 원천', '팀원 간 정합성', '인터페이스 합의' 요청 시 트리거. Team Big Five 조율 메커니즘 1.

Team Science의 Team Big Five 이론(팀 리더십·상호 성과 모니터링·백업 행동·적응성·팀 지향성 + 공유 정신 모델·폐쇄 루프 커뮤니케이션·상호 신뢰)을 에이전트 팀에 적용해 고성과로 과제를 수행하는 오케스트레이터. 여러 에이전트의 협업이 필요한 모든 과제 — 조사·분석·설계·구현·검증·작성 등 — 에 사용. '팀으로', '에이전트 팀', '팀 빅파이브', 'team bigfive', '고성과 팀', '협업으로 처리', '여러 에이전트로' 요청 시 반드시 트리거. 후속 작업: 재실행·다시 실행·업데이트·수정·보완·부분 재실행·이전 결과 개선 요청 시에도 반드시 이 스킬을 사용.