Same protocol, every swarm size.

As agent counts grow, coordination debt explodes: tribal knowledge, unclear walls, invented process. Hive is one cold-start protocol so every bee can answer where it is, what it may touch, what to read next, and what it must not invent — with the same gates from a one-agent docs unit to a hundred-agent program. Human Admin stays governor throughout: agents never grant themselves lock, preview, promote, merge, or cross-hive share.

The universal guarantee

Any agent, dropped into any product build, at any role, at any swarm size, must be able to answer four questions from this file and its product's instance paths alone — no tribal knowledge required:

• Where am I? — role + assigned local project boundary.
• What am I allowed to do? — the role card's owns / must-not line.
• What do I read next? — the product instance's own COMPOSER / AGENTS / WORKSTREAMS.
• What must I not invent? — a fourth job board, a new gate, a shortcut around audit.

Complexity scales by how many units, pools, and digests are running — never by inventing a different process for a bigger swarm.

The local project boundary

By default, a bee may only read or write inside its assigned local project folder — the product repo, worktree, or declared build-master and pool paths for that product. Reading canonical hive/ doctrine is always allowed (that's shared law, not a project data share); talking to agents on a different hive-network product requires explicit, human-Admin-gated approval, scoped to specific paths and an expiry. No bee copies secrets, dumps another product's tree, or "helpfully" syncs files across hives without that gate.

Three Blooms — do not blend them

Before applying any rule, name the altitude. "Bloom" names three different things at Lyrikai, and Hive sits at ops + instance underneath all three — it doesn't fork the governance SOP and doesn't author architectural-spec content:

The altitude map

Seven layers, each owning one thing and citing the rest rather than forking it:

Hive never owns product WORKSTREAMS status or product-specific walls — those stay with the product's own job board.

Instantiating a new build-master or pool

Two copy-and-rename recipes cover both companion packs:

# New build-master (unit loop)
cp -R ⟨lk-bloom⟩/hive/master-build/scaffold \
  /path/to/<product>_build-master
# rename *.template → Seed-from-idea or fill by hand
# Admin program lock → register hive/master-build/INSTANCES.md

# New agent-team-pool (org chart)
cp -R ⟨lk-bloom⟩/hive/agent-team-pools/scaffold \
  /path/to/<product>-master-dev-bot-pool
# rename *.template; copy pools/_POOL.template → pools/<id>/
# Master Pool Lead + path map → Admin pool lock
# register hive/agent-team-pools/INSTANCES.md

Seed also fills the bee roster and declares the assigned local project paths in AGENTS.md before Admin program lock.