먼저 읽을 것
최종 결론: Sanctum LLM 운영은 더 많은 context를 넣는 문제가 아니라 canonical source, execution context, navigation/retrieval, derived knowledge, runtime enforcement, handoff/e2e integration의 권위와 책임을 분리하는 문제다.
이 사이트는 완료된 7개 topic index와 25개 report page 위에 root, tree, final integrated page, operator guide를 연결한 Stage 3 HTML site다.
최종 운영 모델
Sanctum은 contracts/, 코드, 테스트, approved docs, AGENTS.md, llm-context/를 canonical layer로 두고, Graphify는 context router, RAG는 evidence retrieval, LLM Wiki와 보고서는 validation state가 있는 derived knowledge로 둔다. 실제 쓰기 권한은 discord-codex-bot runtime의 write allowlist, approval record, diff guard가 강제한다.
주제별 읽기 경로
00-index: 00-index2개 report10-context-management: Context management3개 report20-graphify-evaluation: Graphify evaluation5개 report30-sanctum-knowledge-architecture: Knowledge architecture4개 report40-worker-agent-operations: Worker-agent operations4개 report50-codex-app-and-runtime-boundary: Codex app and runtime boundary4개 report60-final-recommendations: Final recommendations3개 report
Operator guides
Topic-based xhigh delegation위임 절차와 coverage policyGeneric topic HTML conversion위임 절차와 coverage policyGeneric topic xhigh delegation위임 절차와 coverage policy
Stage 3 static HTML. Relative links only.