Skip to content

Memory

A command center has a persistent, in-repo memory under memory/. Unlike an agent’s built-in memory (which lives outside the repo), this memory travels with the command center, is reviewable in git, and works for any agent — it is the canonical store for that command center.

Each memory is memory/<slug>.md with frontmatter: name, description (one line, used to decide relevance on recall), and type (user / feedback / project / reference). The body is the fact.

Durable facts worth remembering across sessions — a preference, a confirmed decision, project state, a reference. Not transient chatter, and nothing derivable from code or git.

MEMORY.md is a lightweight index read at the start of each session. When a memory’s description matches the task, the agent opens that file. This is what makes long-running context practical without re-explaining everything.

Memory can hold personal context — if you make a command center repo public, review memory/ first. Secrets never belong there (see the generated SECURITY.md).