Skip to main content
decision · created 2026-07-09 · actor claude-cowork · confidence high · status active tags: fleet bridge kg knowledgelayer architecture

Cross-agent dispatch goes through fleet ask <agent> "<task>" (not ad-hoc Telegram round-trips); shared memory writes/reads go through kg (event ledger, daily memories, canonical graph) and kl (synthesized brain, this repo).

Why

  • One provenance-tagged path (FLEET_ACTOR) instead of N bespoke wrappers.
  • Every dispatch auto-logged to the KG ledger → the fleet's activity is queryable memory.
  • Telegram room stays for humans + review, per BOT_COORDINATION reply discipline.

Consequences

  • Skills fleet-bridge (Hermes, pinned; OpenClaw workspace) document the path.
  • BOT_COORDINATION.md contract + TOOLS.md on all runtimes reference it.
  • Raw context stays in kg ledger/memories; only synthesized knowledge enters knowledgelayer.
Sources: ~/.hermes/scripts/fleet_bridge.py · ~/Projects/knowledgelayer