Memory ownership

Brigade’s memory model has one rule: many harnesses may write handoff notes, but exactly one workspace owns the durable memory they get promoted into.

One canonical owner

When several harness workspaces are present, Brigade picks the canonical owner by priority:

openclaw > hermes > claude > codex > this-repo

Override the automatic choice with --owner on commands that route memory.

The owner workspace holds MEMORY.md, memory/cards/, TOOLS.md, USER.md, and rules/. Writer harnesses never edit those files directly; they drop notes into their own per-harness inboxes (.claude/memory-handoffs/, .codex/memory-handoffs/, .opencode/memory-handoffs/) and brigade ingest routes them.

Multiple agent homes? Use a hub.

Treat the owner workspace as the hub. Remote or secondary workspaces write handoffs into their own per-harness inboxes, and a trusted sync pulls those files into a staging inbox on the owner. brigade repos ingest can sweep every registered fleet repo, route each repo’s handoffs into the owner’s memory, and archive the processed handoffs back in the source repo.

Agents stay informed without ever creating a second canonical memory.

Why this matters

Two canonical memories drift. Drifted memory is worse than no memory, because agents trust it. Keeping promotion behind one reviewed pipeline into one owner is what keeps shared memory from becoming a junk drawer.