hivemind

Every agent runs hivemind on startup. The command connects to the shared store, reads a count of its contents, writes a record to confirm write access, reads that record back to confirm the write, and reports the totals and the connection mode. If any step fails, the agent reports OFFLINE and operates from the files.

The canonical record is a set of append-only files on disk. The database is a mirror of those files, replicated to multiple targets: a development store and two production stores. Each target holds a complete, queryable copy. An agent connects to the nearest reachable target; when none is reachable, it reads the files directly.

Entries are written once and read by any later agent. Lessons, rules, checkpoints, and results accumulate in the store and remain available across sessions, machines, and time. Entries are not modified in place; a correction is written as a new entry that supersedes the earlier one, which is retained as history.