모델 순위: LMArena Leaderboard 기반 (2025년 12월 기준)
Claude Code, ChatGPT GPTs, Gemini Gems를 위한 종합 AI 프롬프트 엔지니어링 스킬 모음입니다.
설정 없이 바로 사용할 수 있는 링크입니다:
| 플랫폼 | 링크 |
|---|---|
| ChatGPT GPTs | 두부경 종합 프롬프트 생성기 |
| Gemini Gems | 프롬프트 생성기 Gem |
이 저장소는 다음 모델에 최적화된 프롬프트 엔지니어링 자료를 제공합니다:
| 모델 | 커버리지 |
|---|---|
| GPT-5.2 / GPT-5.2-Codex | XML 패턴, reasoning_effort, Compaction |
| Claude 4.5 (Opus/Sonnet/Haiku) | 명시적 지시, Extended Thinking |
| Gemini 3 | Constraints First, 멀티모달 컨텍스트 |
| Veo 3.1 | 오디오 포함 동영상 생성 |
| Gemini Image | 이미지 생성 |
출처: LMArena Leaderboard - 사용자 투표 기반 순위
| 목적 | 1순위 | 2순위 | 3순위 |
|---|---|---|---|
| 코딩/개발 | Claude Opus 4.5 | GPT-5.2 | Gemini 3 Pro |
| 수학/논리 | GPT-5.2 | Gemini 3 Flash | Claude Opus 4.5 |
| 글쓰기/창작 | Gemini 3 Pro | Gemini 3 Flash | Claude Opus 4.5 |
| 이미지 생성 | GPT Image 1.5 | Gemini 3 Pro Image | Flux 2 Max |
| 동영상 생성 | Veo 3.1 | Sora 2 Pro | Veo 3 |
| 웹 검색/리서치 | Gemini 3 Pro Grounding | GPT-5.2 Search | GPT-5.1 Search |
| 팩트체크 | GPT-5.2 Thinking | Gemini 3 Pro Grounding | Perplexity Sonar Pro |
prompt-engineering-skills/
├── README.md # 이 파일
├── LICENSE # MIT 라이선스
│
├── skills/ # 핵심 스킬 파일
│ ├── prompt-engineering-guide.md # 모델별 프롬프트 전략
│ ├── image-prompt-guide.md # 이미지 생성 가이드
│ ├── gpt-5.2-prompt-enhancement.md # GPT-5.2 전용 패턴
│ ├── claude-4.5-prompt-strategies.md # Claude 4.5 전용 전략
│ ├── gemini-prompt-strategies.md # Gemini/Veo/Nano Banana 전략
│ └── context-engineering-collection.md # CE 원칙
│
├── commands/ # Claude Code 커맨드
│ ├── prompt.md # /prompt 커맨드
│ └── prompt-sync.md # /prompt-sync 동기화 커맨드
│
├── instructions/ # GPTs/Gems 시스템 프롬프트
│ ├── GPTs-Prompt-Generator.md # ChatGPT GPTs용
│ └── Gems-Prompt-Generator.md # Gemini Gems용
│
└── examples/ # 사용 예시
├── gpt-5.2-examples.md
├── claude-4.5-examples.md
└── image-generation-examples.md
macOS / Linux:
git clone https://github.com/tofukyung/prompt-engineering-skills.git /tmp/pes && \
mkdir -p ~/.claude/skills ~/.claude/commands && \
cp /tmp/pes/skills/*.md ~/.claude/skills/ && \
cp /tmp/pes/commands/*.md ~/.claude/commands/ && \
rm -rf /tmp/pes && \
echo "✅ 설치 완료! 이제 모든 프로젝트에서 /prompt 사용 가능"Windows PowerShell:
git clone https://github.com/tofukyung/prompt-engineering-skills.git $env:TEMP\pes; `
New-Item -ItemType Directory -Force -Path "$env:USERPROFILE\.claude\skills", "$env:USERPROFILE\.claude\commands" | Out-Null; `
Copy-Item "$env:TEMP\pes\skills\*.md" "$env:USERPROFILE\.claude\skills\"; `
Copy-Item "$env:TEMP\pes\commands\*.md" "$env:USERPROFILE\.claude\commands\"; `
Remove-Item -Recurse -Force "$env:TEMP\pes"; `
Write-Host "✅ 설치 완료! 이제 모든 프로젝트에서 /prompt 사용 가능" -ForegroundColor Green| 설치 위치 | 적용 범위 | 사용 시점 |
|---|---|---|
~/.claude/ (글로벌) |
모든 프로젝트에서 사용 가능 | 개인용 도구로 항상 사용 |
.claude/ (프로젝트 폴더) |
해당 프로젝트에서만 사용 | 팀 공유, 프로젝트 특화 설정 |
주의: repo를 포크/클론하면 해당 프로젝트에만 설치됩니다. 모든 프로젝트에서 사용하려면 위의 글로벌 설치 명령어를 실행하세요.
# 글로벌 설치 (모든 프로젝트에서 사용)
cp skills/*.md ~/.claude/skills/
cp commands/*.md ~/.claude/commands/그 후 Claude Code에서 /prompt 커맨드 사용.
설치 후 최신 버전으로 업데이트하려면:
/prompt-sync
또는 원클릭 동기화:
macOS / Linux:
git clone --depth 1 https://github.com/tofukyung/prompt-engineering-skills.git /tmp/pes-sync && \
cp /tmp/pes-sync/skills/*.md ~/.claude/skills/ && \
cp /tmp/pes-sync/commands/*.md ~/.claude/commands/ && \
rm -rf /tmp/pes-sync && echo "✅ 동기화 완료!"Windows PowerShell:
git clone --depth 1 https://github.com/tofukyung/prompt-engineering-skills.git $env:TEMP\pes-sync; `
Copy-Item "$env:TEMP\pes-sync\skills\*.md" "$env:USERPROFILE\.claude\skills\"; `
Copy-Item "$env:TEMP\pes-sync\commands\*.md" "$env:USERPROFILE\.claude\commands\"; `
Remove-Item -Recurse -Force "$env:TEMP\pes-sync"; `
Write-Host "✅ 동기화 완료!" -ForegroundColor Green- GPT Editor 접속
skills/폴더의 파일을 Knowledge에 업로드instructions/GPTs-Prompt-Generator.md내용을 Instructions에 붙여넣기
- Gems 설정 접속
skills/폴더의 파일을 지식 파일에 업로드instructions/Gems-Prompt-Generator.md내용을 지침에 붙여넣기
메인 스킬 파일:
- 모델별 프롬프트 전략
- 각 모델의 필수 XML 블록
- Context Engineering 원칙
- 목적별 템플릿
- 품질 체크리스트
공냥이(@specal1849)님의 자료를 기반으로 한 종합 이미지 생성 가이드:
- 시그널(Signal)과 신뢰도(Faithful) 개념
- 프롬프트 형식 비교 (JSON/XML/Markdown/자연어)
- 스타일별 템플릿 (제품/푸드/패션/캐릭터 등)
- 조명 및 카메라 기법
- 인포그래픽 제작 가이드
Agent Skills for Context Engineering을 기반으로 한 Context Engineering 원칙:
- Attention Budget 관리
- 컨텍스트 저하 방지
- 멀티 에이전트 패턴
- 도구 설계 원칙
| 기능 | 패턴 |
|---|---|
| 장황함 제어 | <output_verbosity_spec> |
| 범위 제약 | <design_and_scope_constraints> |
| 불확실성 처리 | <uncertainty_and_ambiguity> |
| 도구 사용 | <tool_usage_rules> |
Codex 핵심 원칙: "Less is More" - 최소한의 프롬프팅이 최적
| 기능 | 패턴 |
|---|---|
| 기본 행동 | <default_to_action> |
| 명시적 지시 | 모든 행동에 필수 |
| Extended Thinking | 복잡한 추론 시 활성화 |
| 병렬 도구 호출 | <use_parallel_tool_calls> |
| 기능 | 패턴 |
|---|---|
| Constraints First | 제약 조건을 최상단에 배치 |
| Temperature | 1.0 권장 |
| 멀티모달 | 이미지/오디오 컨텍스트 네이티브 지원 |
| Veo 오디오 | 대화, 음향효과, 배경음 |
공냥이 (@specal1849)님의 자료 기반:
Muratcan Koylan의 Agent Skills for Context Engineering 기반
OpenAI, Anthropic, Google 공식 문서 참조
MIT License - LICENSE 참조
기여를 환영합니다! PR 제출 전 기여 가이드라인을 확인해 주세요.
- 현재: 1.8.0
- 최종 업데이트: 2026-01-01
- 변경사항 v1.8.0:
- 동영상 프롬프트 JSON 구조 통일: 모든 동영상에 스토리보드 형식(scenes 배열) 적용
- GPTs 기본 모델: Sora 2, Gems 기본 모델: Veo 3.1
- 스토리보드 필수 요소 체크리스트 추가 (sequence, duration, description, camera, audio)
- 변경사항 v1.7.0:
- 동영상 생성 시 스토리보드 단계 필수화 (생략 금지)
- 스토리보드 JSON 구조 상세화 (lighting, character action, camera motion 등)
- 변경사항 v1.6.0:
- 명시적 요소 확장: 사용자 입력에서 암시적 요소를 구체적으로 풀어쓰기
- Agent 모드 프롬프트 지원 추가
- 변경사항 v1.5.0:
- 모든 채널(GPTs, Gems, Skills) 버전 1.5.0으로 통일
- 동영상 프롬프트 JSON+자연어 형식으로 통일 (이미지와 동일)
- gpt-image 모델명 통일 (DALL-E/GPT Image 1.5 → gpt-image)
/prompt-sync명령어 추가: 자동 업데이트 동기화- 원클릭 설치/동기화 스크립트 추가
- 진행 상황 표시 (🔍→🧠→✅) 추가
- $ARGUMENTS 처리 규칙 추가
- 변경사항 v1.4.0:
- README 버전 정보 동기화
- GPTs-Prompt-Generator: v5.2.0 (워크플로우 전면 개편, 모든 작업 금지사항 적용)
- Gems-Prompt-Generator: v4.3.0 (네이티브 이미지 생성 호출 명시)
- commands/prompt.md: v4.1.0 (금지사항 강화, 개선 옵션 UI)
- 변경사항 v1.3.0:
- 검색/리서치 모델 추천 추가 (Search Arena 기준)
- 팩트체크: GPT-5.2 Thinking 고정 추천
- research-prompt-guide.md 스킬 파일 추가
- 변경사항 v1.2.0:
- 빠른 프리셋 제거 → LMArena 기반 목적별 추천으로 대체
- 모든 문서에 LMArena 출처 명시
Model Rankings: Based on LMArena Leaderboard (December 2025)
A comprehensive collection of AI prompt engineering skills for Claude Code, ChatGPT GPTs, and Gemini Gems.
Use these links without any setup:
| Platform | Link |
|---|---|
| ChatGPT GPTs | Tofukyung Comprehensive Prompt Generator |
| Gemini Gems | Prompt Generator Gem |
This repository provides production-ready prompt engineering resources optimized for:
| Model | Coverage |
|---|---|
| GPT-5.2 / GPT-5.2-Codex | XML patterns, reasoning_effort, Compaction |
| Claude 4.5 (Opus/Sonnet/Haiku) | Explicit instructions, Extended Thinking |
| Gemini 3 | Constraints First, multimodal context |
| Veo 3.1 | Video generation with audio |
| Gemini Image | Image generation |
Source: LMArena Leaderboard - User voting based rankings
| Purpose | 1st | 2nd | 3rd |
|---|---|---|---|
| Coding/Dev | Claude Opus 4.5 | GPT-5.2 | Gemini 3 Pro |
| Math/Logic | GPT-5.2 | Gemini 3 Flash | Claude Opus 4.5 |
| Writing/Creative | Gemini 3 Pro | Gemini 3 Flash | Claude Opus 4.5 |
| Image Generation | GPT Image 1.5 | Gemini 3 Pro Image | Flux 2 Max |
| Video Generation | Veo 3.1 | Sora 2 Pro | Veo 3 |
| Web Search/Research | Gemini 3 Pro Grounding | GPT-5.2 Search | GPT-5.1 Search |
| Fact-Check | GPT-5.2 Thinking | Gemini 3 Pro Grounding | Perplexity Sonar Pro |
prompt-engineering-skills/
├── README.md # This file
├── LICENSE # MIT License
│
├── skills/ # Core skill files
│ ├── prompt-engineering-guide.md # Multi-model prompt strategies
│ ├── image-prompt-guide.md # Image generation guide
│ ├── gpt-5.2-prompt-enhancement.md # GPT-5.2 specific patterns
│ ├── claude-4.5-prompt-strategies.md # Claude 4.5 specific strategies
│ ├── gemini-prompt-strategies.md # Gemini/Veo/Nano Banana strategies
│ └── context-engineering-collection.md # CE principles
│
├── commands/ # Claude Code commands
│ ├── prompt.md # /prompt command
│ └── prompt-sync.md # /prompt-sync sync command
│
├── instructions/ # GPTs/Gems system prompts
│ ├── GPTs-Prompt-Generator.md # For ChatGPT GPTs
│ └── Gems-Prompt-Generator.md # For Gemini Gems
│
└── examples/ # Usage examples
├── gpt-5.2-examples.md
├── claude-4.5-examples.md
└── image-generation-examples.md
macOS / Linux:
git clone https://github.com/tofukyung/prompt-engineering-skills.git /tmp/pes && \
mkdir -p ~/.claude/skills ~/.claude/commands && \
cp /tmp/pes/skills/*.md ~/.claude/skills/ && \
cp /tmp/pes/commands/*.md ~/.claude/commands/ && \
rm -rf /tmp/pes && \
echo "✅ Done! Now you can use /prompt in any project"Windows PowerShell:
git clone https://github.com/tofukyung/prompt-engineering-skills.git $env:TEMP\pes; `
New-Item -ItemType Directory -Force -Path "$env:USERPROFILE\.claude\skills", "$env:USERPROFILE\.claude\commands" | Out-Null; `
Copy-Item "$env:TEMP\pes\skills\*.md" "$env:USERPROFILE\.claude\skills\"; `
Copy-Item "$env:TEMP\pes\commands\*.md" "$env:USERPROFILE\.claude\commands\"; `
Remove-Item -Recurse -Force "$env:TEMP\pes"; `
Write-Host "✅ Done! Now you can use /prompt in any project" -ForegroundColor Green| Install Location | Scope | When to Use |
|---|---|---|
~/.claude/ (global) |
All projects | Personal tools, always available |
.claude/ (project folder) |
This project only | Team sharing, project-specific |
Note: Forking/cloning the repo only installs to that project. Run the global install command above to use in all projects.
# Global install (use in all projects)
cp skills/*.md ~/.claude/skills/
cp commands/*.md ~/.claude/commands/Then use the /prompt command in Claude Code.
To update to the latest version after installation:
/prompt-sync
Or one-click sync:
macOS / Linux:
git clone --depth 1 https://github.com/tofukyung/prompt-engineering-skills.git /tmp/pes-sync && \
cp /tmp/pes-sync/skills/*.md ~/.claude/skills/ && \
cp /tmp/pes-sync/commands/*.md ~/.claude/commands/ && \
rm -rf /tmp/pes-sync && echo "✅ Sync complete!"Windows PowerShell:
git clone --depth 1 https://github.com/tofukyung/prompt-engineering-skills.git $env:TEMP\pes-sync; `
Copy-Item "$env:TEMP\pes-sync\skills\*.md" "$env:USERPROFILE\.claude\skills\"; `
Copy-Item "$env:TEMP\pes-sync\commands\*.md" "$env:USERPROFILE\.claude\commands\"; `
Remove-Item -Recurse -Force "$env:TEMP\pes-sync"; `
Write-Host "✅ Sync complete!" -ForegroundColor Green- Go to GPT Editor
- Upload files from
skills/to Knowledge - Paste
instructions/GPTs-Prompt-Generator.mdinto Instructions
- Go to Gems Settings
- Upload files from
skills/to Knowledge files - Paste
instructions/Gems-Prompt-Generator.mdinto Instructions
The main skill file covering:
- Model-specific prompt strategies
- Required XML blocks for each model
- Context Engineering principles
- Purpose-specific templates
- Quality checklists
Comprehensive image generation guide based on @specal1849's work:
- Signal and Faithful concepts
- Prompt format comparison (JSON/XML/Markdown/Natural language)
- Style templates (Product, Food, Fashion, Character, etc.)
- Lighting and camera techniques
- Infographic creation guide
Context Engineering principles based on Agent Skills for Context Engineering:
- Attention budget management
- Context degradation prevention
- Multi-agent patterns
- Tool design principles
| Feature | Pattern |
|---|---|
| Verbosity Control | <output_verbosity_spec> |
| Scope Constraints | <design_and_scope_constraints> |
| Uncertainty Handling | <uncertainty_and_ambiguity> |
| Tool Usage | <tool_usage_rules> |
Codex Key Principle: "Less is More" - minimal prompting works best
| Feature | Pattern |
|---|---|
| Action by Default | <default_to_action> |
| Explicit Instructions | Required for all behaviors |
| Extended Thinking | Enable for complex reasoning |
| Parallel Tool Calls | <use_parallel_tool_calls> |
| Feature | Pattern |
|---|---|
| Constraints First | Place constraints at the top |
| Temperature | 1.0 recommended |
| Multimodal | Native image/audio context |
| Veo Audio | Dialogue, sound effects, ambient |
Based on materials by 공냥이 (@specal1849):
Based on Agent Skills for Context Engineering by Muratcan Koylan.
Official docs from OpenAI, Anthropic, Google
MIT License - see LICENSE
Contributions welcome! Please read the contributing guidelines before submitting PRs.
- Current: 1.8.0
- Last Updated: 2026-01-01
- Changes v1.8.0:
- Unified video prompt JSON structure: storyboard format (scenes array) for all videos
- GPTs default model: Sora 2, Gems default model: Veo 3.1
- Added storyboard required elements checklist (sequence, duration, description, camera, audio)
- Changes v1.7.0:
- Storyboard step now mandatory for video generation (cannot skip)
- Detailed storyboard JSON structure (lighting, character action, camera motion, etc.)
- Changes v1.6.0:
- Explicit element expansion: expand implicit elements from user input
- Added Agent mode prompt support
- Changes v1.5.0:
- Unified all channels (GPTs, Gems, Skills) to version 1.5.0
- Video prompts now use JSON+natural language format (same as images)
- Standardized gpt-image model naming (DALL-E/GPT Image 1.5 → gpt-image)
- Added
/prompt-synccommand: automatic update synchronization - Added one-click install/sync scripts
- Added progress indicators (🔍→🧠→✅)
- Added $ARGUMENTS handling rules
- Changes v1.4.0:
- README version info synchronized
- GPTs-Prompt-Generator: v5.2.0 (workflow overhaul, all-task restrictions applied)
- Gems-Prompt-Generator: v4.3.0 (native image generation call specified)
- commands/prompt.md: v4.1.0 (restrictions strengthened, improvement options UI)
- Changes v1.3.0:
- Added search/research model recommendations (Search Arena)
- Fact-check: GPT-5.2 Thinking fixed recommendation
- Added research-prompt-guide.md skill file
- Changes v1.2.0:
- Removed fast presets → Replaced with LMArena-based recommendations
- Added LMArena source attribution to all documentation