Index an external documentation site into a markdown notebook the agent can browse like its own memory. Pages are stored as single frontmatter+body files under the agent's brain, and a master `index.md` lets the agent scan one-line hooks, pick candidates, and answer with citations — without any embedding model. Designed for customer-facing agents that have to answer questions from a knowledge base the owner has ingested.
외부 문서 사이트를 에이전트가 자신의 메모리처럼 탐색할 수 있는 마크다운 노트북으로 인덱싱합니다. 각 페이지는 YAML 프런트매터(설명, 카테고리, 질문 답변 등)와 정리된 본문이 포함된 하나의 파일이 되며, 마스터 index.md를 통해 에이전트가 임베딩 모델 없이도 한 줄의 후크를 스캔하여 올바른 페이지를 선택할 수 있습니다. 지식 베이스에 대한 질문에 답변해야 하는 고객 대면 CX 에이전트에게 유용합니다.
The user (owner) wants to ingest a documentation site so future conversations can answer from it with citations. Build or refresh the notebook.
A visitor in this conversation asks a question that should be answered from a knowledge base that has already been indexed. Scan the index, read the relevant page, answer with a markdown link.
The user wants to schedule periodic re-indexing of a docs site.
The owner is reviewing snapshot session transcripts and wants to know which docs topics are missing or under-described.
Not for ad-hoc one-off web reading — use http_request or browser for that. This skill is only worth loading when there is (or will be) a persistent notebook to consult.