メニュー
Start with deep research before planning. Use when the user asks for research-first behavior, wants detailed investigation before a plan, or when the request clearly needs substantial discovery, comparison, or validation before implementation.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Prepare and create a GitHub pull request from the current branch. Use when the user asks to open a PR, create a pull request, or requests `/pr`, including an optional base branch argument.
Orchestrate staged repository work through delegated preflight, single-writer implementation, independent validation, and a final closure gate. Use for multi-stage refactors, migrations, phased rollouts, or other work with clean stage boundaries. The parent coordinates; subagents perform all repository observation and mutation.
Resolve XY problems — surface questions that mask a deeper goal. Use when the user invokes /xy, when a request is vague or mismatched to its context, or when a narrow how-to looks like it's hiding the real objective.
Always use pnpm/pnpx for Node.js work unless the project packageManager field specifies another package manager. Use when Node.js, JavaScript, TypeScript, npm scripts, package installation, or Node CLI tools are involved.
YouTube 영상 URL에서 자막을 추출하고 한국어로 상세 요약을 작성합니다. Use this skill whenever the user provides a YouTube URL and asks to summarize, recap, extract a transcript, analyze video contents, or create Korean notes from a video, including Korean prompts such as "유튜브 요약", "영상 정리", "자막 추출", or "한국어로 요약".
Use when the user wants to rapidly learn, memorize, quiz, review, or prepare for exams from Markdown course notes or weekly lecture files. Runs an adaptive active-recall loop with questions, grading, error patches, summaries, and next-action selection until the user stops.
| name | research-first |
| description | Start with deep research before planning. Use when the user asks for research-first behavior, wants detailed investigation before a plan, or when the request clearly needs substantial discovery, comparison, or validation before implementation. |
Ask detailed questions first, grill-me style, to define exactly what to research, which competing areas or approaches matter, what constraints and unknowns exist, and what evidence would make the research sufficient.
Then split the work into 2-3 execution-relevant tracks. If team workflow is available, use a leader plus 2-3 agents; otherwise use 2-3 subagents. Assign each one a distinct area or competing approach, and have them judge practicality, tradeoffs, and readiness for execution.
Synthesize the results and report the research in detail. Then use AskUserQuestion to decide whether to do more research or save the research as markdown and enter plan mode.
In plan mode, keep asking grill-me-style clarification questions until the requirements are sharp, then produce the plan and continue with normal plan-mode execution.
