← ClaudeAtlas

engramlisted

Networked PARA knowledge brain — manages documents by PARA (Projects, Areas, Resources, Archives) AND weaves them into one connected knowledge graph through bi-directional links, MOC hubs, and integrity linting: a logical link layer over physical folders, so the brain grows like a network instead of a filing cabinet. The brain is normally a shared Postgres-backed STORE that is searched rather than grepped, with a revision history per document; a local file brain is used for repos the store does not admit. Use to manage, organize, search or save documents; save meeting notes; archive or migrate a project; connect notes; find orphans; fix broken links; update MOCs; raise neural density; review what the brain gained this session; read a document's history; or point this machine at a store. Matches intent in any language — e.g. "문서 정리", "회의록 저장", "노트 연결", "링크 점검", "고아 문서", "MOC 업데이트", "브레인 리뷰", "브레인 검색", "저장소에서 찾아줘", "이 문서 이력", "누가 언제 바꿨어", "organize docs", "save this as a note", "connect notes", "find orphans",
poorants/engram · ★ 0 · API & Backend · score 68
Install: claude install-skill poorants/engram
# engram — networked PARA knowledge brain Manage documents with the PARA method while weaving them into **one connected knowledge network** through bi-directional links, MOC hubs, and integrity linting. The core idea is *Networked PARA*: a logical link layer on top of physical classification. Two layers: - **Management (PARA)** — create, move, archive, migrate, review. Folders own governance. - **Connection (network)** — link documents by context, clear orphans, weave lonely spokes into a mesh rather than a star, catch broken links. Links own context. The detailed rules are in [references/linking-rules.md](references/linking-rules.md); load it before any linking work and follow it. The thing being defended is simple: **a session dies, knowledge should not.** The measure of success is not investigating the same question twice. ## The store — resolve this first The brain is normally **not a tree of files**. It is a Postgres-backed service, and there is **one client for it**: the `engram` binary. Everything below is a surface over that one client — the address, the token, the path rules and the author byline live there, once. 1. **MCP tools (prefer these in a session)** — `brain_search`, `brain_get`, `brain_revisions`, `brain_integrity`, `brain_put`, `brain_move`. Same endpoints, same single ranking, no permission churn, and available before this skill even loads. One difference: an MCP server is spawned once per session and cannot see which checkout