
사람의 주관 판단이 필요한 A/B·before-after·model·prompt·UX·design 품질 비교 평가에서 가벼운 HITL 평가 루프를 설계하고 구현하도록 안내한다. 다음 상황에서 먼저 호출하세요: - "직접 봐야 판단 가능", "A/B 비교", "before/after", "품질 평가", "점수 남기기", "평가 대시보드", "HITL", "리포트 만들기"를 언급 - 자동화 테스트만으로는 품질 판단이 부족하고 사람의 주관이 필요한 경우 - AI 출력, 디자인 생성물, 프롬프트 변경, 파이프라인 변경, UX 후보를 비교할 때 - 평가 결과를 나중에 집계하거나 재개할 수 있어야 하는 경우 구현 스캐폴드가 아니라 가이드형이다. 코드를 바로 만들지 말고 먼저 설계를 확인한다.
Create focused thread-to-thread handoff packets (goal, locked decisions, relevant files, risks, acceptance criteria, and next-thread starter prompt). Use when switching tasks, moving from backend to frontend, splitting large work into phases, or asking another agent/thread to continue implementation without context loss.
Decide between Strategy, Template Method, or Hybrid for architecture and refactor tasks across frontend/backend systems. Use when evaluating branching logic, adding new variants/providers/modes, introducing shared execution pipelines, or debating runtime algorithm switching versus fixed skeleton workflows. Trigger especially for AI feature flows and payment/provider integrations where pattern choice affects extensibility, testability, and change risk.
Plan-first staged TDD orchestration for TypeScript web repositories. Use when a TypeScript, React, Next.js, or Node web task benefits from a short spec, explicit red/green/refactor/verify stages, deterministic validation, and keeping code changes behind explicit user approval. Best suited for small feature additions, policy changes, and bug fixes where tests should act as the primary oracle.
Install Codex skills into $CODEX_HOME/skills from a curated list or a GitHub repo path. Use when a user asks to list installable skills, install a curated skill, or install a skill from another repo (including private repos).
Guide for creating effective skills. This skill should be used when users want to create a new skill (or update an existing skill) that extends Codex's capabilities with specialized knowledge, workflows, or tool integrations.
Create or update grounded functional specifications for web, app, and backend features. Use when documenting an existing feature from the codebase into an AS-IS spec, turning a planned change into a TO-BE spec, standardizing team feature documentation, or reconciling UI, API, DB, and business rules into one decision-ready requirements document.
Use when extracting honest resume bullets, 1-minute interview stories, or likely interviewer questions from a defined work period (a project, week, sprint, or custom date range). Collects evidence from git, GitHub, and an Obsidian vault, filters strictly to verified self-attribution, refuses to invent impact metrics, and marks AI-assist where material. Triggers on "이력서 bullet", "주간 회고 이력서", "프로젝트 X 회고 이력서 톤", "포트폴리오 자료 뽑아줘".
Run Rabbit Scaffold dev verification flows (register, referral, notification preferences, password reset) and print API responses plus worker logs. Use when you need to quickly validate the dev docker-compose stack and event flow after a build or pull.
Split large git changes into readable, review-friendly, revert-safe commit stacks in Toolditor. Use when preparing a PR, reorganizing staged/unstaged changes, enforcing Korean `[type] message` commit convention, or producing commit history that is easy to review and selectively revert.