Pillar I · 도구 모음

Context Engineering Toolset 3종

시연에서 본 도구들 — Slash Commands · Sub-agents/Skills · MCP — 세 가지가 함께 작동할 때 agent는 결정론적으로 움직인다
도구 1
Slash Commands
특화 트리거
"Specialized triggers for precise navigation and intent signaling."정확한 navigation과 의도 신호를 위한 특화 트리거
  • 워크플로우를 codify하는 1차 도구
  • 반복 5단계를 한 줄 명령으로 압축
  • agent에게 "이 작업은 이 워크플로우다" 명시
/refactor-module
도구 2
🧩
Sub-agents · Skills
모듈형 Agent
"Modular agents assigned to narrow domains to keep local context windows clean."좁은 도메인의 모듈형 에이전트 — 로컬 컨텍스트를 깨끗하게
  • 특정 작업을 격리된 context로 위임
  • narrow domain에만 책임 → clean local context
  • 무거운 탐색은 위임, 압축된 요약만 회수
.claude/agents/
도구 3
🔌
MCP
표준화된 인터페이스
"Standardized interfaces for external data retrieval."외부 데이터 retrieval을 위한 표준화된 인터페이스
  • Jira · Notion · Linear · 사내 wiki를 agent의 도구
  • 각 시스템별 인터페이스 학습 ✕ → MCP 한 번
  • Second Brain의 외부 확장판
Model Context Protocol
🔗 Toolset의 통합
개별 사용 ✕ — 연결된 그래프로   CLAUDE.md → /slash-command → sub-agent → MCP