code-memory

Local-first memory layer (semantic vectors + call/import graph + episodes + bi-temporal user claims) wired in via MCP. Use to orient on unfamiliar code, answer topology questions (who calls X, who imports Y), recall prior work, refresh the index after writes, AND to remember durable user assertions — preferences ("I love / prefer / want"), tech-stack decisions ("we use X"), rejections ("don't ship Y"), ownership, location facts. Load this skill whenever the user states an opinion, preference, choice, or correction worth keeping across sessions.

Orient on the codebase via the code-memory MCP index before grep/read/shell, and drive the backend manually (retrieve / record / reingest / ingest).

Local-first memory layer (semantic vectors + call/import graph + episodes) wired in via MCP and the OpenCode plugin. Use to orient on unfamiliar code, answer topology questions (who calls X, who imports Y), recall prior work, and refresh the index after writes.