context-engineering

SKILL.md

The discipline of curating the smallest high-signal token set that achieves desired outcomes.

Context engineering goes beyond prompt engineering. While prompts focus on what you ask, context engineering focuses on everything the model sees—system instructions, tool definitions, documents, message history, and tool outputs.

Key Insight: Context windows are constrained not by raw token capacity but by attention mechanics. As context grows, models experience degradation.

에이전트 시스템 프롬프트를 디자인할 때, RAG 검색을 최적화할 때 또는 컨텍스트가 너무 비싸거나 느린 경우에 사용합니다. 전략적 포지셔닝과 주의를 기울이는 디자인을 통해 품질을 유지하면서 토큰을 줄입니다. 출처: yonatangross/skillforge-claude-plugin.

원본 보기