Agent Knowledge Management
Agent Knowledge Management
Overview
이 맵은 GitHub + Markdown + Obsidian + 에이전트 규칙 기반 개인/에이전트 지식관리 시스템의 내비게이션 문서입니다.
Core Concepts
- [Personal Knowledge Management](/notes/30-concepts__Personal Knowledge Management/)
- [Second Brain](/notes/30-concepts__Second Brain/)
- RAG
- Obsidian
- [Knowledge Graph](/notes/30-concepts__Knowledge Graph/)
- [Agent Memory](/notes/30-concepts__Agent Memory/)
- [Agent Harness](/notes/30-concepts__Agent Harness/)
- [Ephemeral Agent Sandbox](/notes/30-concepts__Ephemeral Agent Sandbox/)
- [Agent Skill](/notes/30-concepts__Agent Skill/)
- [Human Approval Boundary](/notes/30-concepts__Human Approval Boundary/)
Key Sources
- [Use GitHub Markdown as Knowledge Source of Truth](/notes/50-decisions__Use GitHub Markdown as Knowledge Source of Truth/)
- [OpenAI Build Hour - Agents SDK](/notes/10-sources__OpenAI Build Hour - Agents SDK/)
Important Insights
- [Production agents need recoverable failure boundaries](/notes/20-insights__Production agents need recoverable failure boundaries/)
- [Skills turn organizational know-how into executable agent assets](/notes/20-insights__Skills turn organizational know-how into executable agent assets/)
Current Position
Markdown Git repo를 지식의 원장으로 두고, Obsidian은 인간용 그래프/탐색 UI, RAG/vector DB는 Markdown에서 재생성 가능한 검색 캐시로 둔다.
에이전트 운영 관점에서는 대화 히스토리만이 아니라 [Agent Harness](/notes/30-concepts__Agent Harness/), [Ephemeral Agent Sandbox](/notes/30-concepts__Ephemeral Agent Sandbox/), [Agent Skill](/notes/30-concepts__Agent Skill/), [Human Approval Boundary](/notes/30-concepts__Human Approval Boundary/)를 함께 설계해야 장시간 작업을 안전하게 운영할 수 있다.
Decisions
- [Use GitHub Markdown as Knowledge Source of Truth](/notes/50-decisions__Use GitHub Markdown as Knowledge Source of Truth/)
Open Questions
- 대화에서 생긴 인사이트를 어느 기준으로 저장할 것인가?
- concept note 자동 갱신은 어느 정도까지 허용할 것인가?
- RAG index는 언제 붙일 것인가?
- 샌드박스 스냅샷과 skill 변경 이력은 지식관리 그래프에서 어느 수준까지 추적할 것인가?