memfleet-first

name

memfleet-first

description

Always use FIRST in any multi-agent session before reading code, planning a refactor, or making an edit on a repo with a registered MemFleet broker. Triggered by: 'I'm about to edit X', 'rename Y across the fleet', 'plan this refactor', joining a running fleet session, coordinating with other agents, prose hand-offs between agents. Do not write a prose rationale for an edit, do not grep for 'who else might be touching this', and do not skip publish_intent because 'it's just a small change'. Empty fleet_status is not permission to skip the protocol — it just means you're the first agent in this window. Skip ONLY for solo-agent sessions with no broker registered or for pure docs-only edits where coordination has zero value.

MemFleet First

The Iron Law

IN A MULTI-AGENT SESSION → MEMFLEET TOOLS BEFORE EDITS. NO EXCEPTIONS.
  1. publish_intent  (declare structural intent, get blast radius + conflicts)
  2. edit            (your normal edit loop)
  3. record_episode  (classify A/B/C, update NodeState rollups for the fleet)

MemFleet is the coordination memory of the fleet, not a chat channel. Every edit is a typed structural event with a known blast radius. Skipping the protocol means silent collisions, overwrites at merge time, and the rest of the fleet wasting tokens re-reading work you already did.

A typed intent serializes to ~20 tokens. A prose rationale averages 200+. A 10-agent fleet × 100 edits = 90,000 tokens saved per fleet-turn when the protocol is followed.

Harness Notes — Claude Code, Codex, Copilot

MemFleet ships as a single MCP server consumed by every harness. The skill is harness-agnostic:

• Claude Code / Codex / Copilot CLI — call MemFleet tools by their MCP names (mcp__memfleet__publish_intent, etc.). All examples below use this prefix.
• Editor harness without an MCP bridge — fall back to fleet-cli for the same operations; the CLI mirrors every MCP tool one-to-one.