Maat Protocol — Agent Orchestration Framework
Overview
Repo: github.com/imaginationeverywhere/maat-protocol (PUBLIC)
Domain: maatagent.com (Route 53)
License: Apache 2.0
Author: Amen Ra
Three-Layer Release Strategy
Phase 1 (Current): Philosophy, docs, examples — NO tools. Sonnet builds this.
Phase 2: .maat/ directory, commands, 42 Declarations, config templates
Phase 3 (NEVER fully public): Detection layer, countermeasures, audit architecture — “The Triple Cross”
The Triple Cross
Play: Publish 42 Declarations (ethical rules)
Double Cross: Bad actors read rules, know what to violate
Triple Cross: Orchestration framework IS the detection mechanism — audit fingerprints in every dispatch/escalation. Can’t strip without breaking orchestration.
Internal planning stays in .claude/plans/ (gitignored)
Technical Architecture
Public: .maat/ directory with config.yml, agents/, commands/, templates/, declarations/
Private: .maat-internal/ with detection/, countermeasures/, governance/
Commands: maat-loop, maat-dispatch, maat-plan, maat-test, maat-status
Three-Tier Architecture (Maat Protocol’s Core Pattern)
Tier 1 (Architect): Expensive reasoning model. Strategy only. Never grunt work.
Tier 2 (Manager): Cheap/unlimited model. Monitor-Decide-Dispatch (MDD). Never writes code.
Tier 3 (Workers): CLI agents. ALL execution. One focused task per dispatch. Max 4 concurrent.
Sonnet’s Role
Sonnet operates from boilerplate directory (has all 145+ commands)
Uses absolute paths / cd to reach each Heru and maat-protocol
Handles: PRDs, bootstrap, status commands, documentation, planning
Maat Protocol Phase 1 docs defined in maat-protocol/CLAUDE.md (8 documents)
Kemetic Naming
Maat = truth, balance, order, justice (the protocol)
Isfet = chaos, disorder (what we fight)
Set = the adversary (bad actors)
42 Declarations = from Book of Coming Forth by Day
Hall of Judgment = audit/detection layer
The Feather = the standard agents are measured against
Vision (5+ years)
Near-term: developer workflow orchestration, email/calendar/finance management
Long-term: physical robots powered by agents, community safety, civil rights protection
AI witnesses during police encounters, school safety, financial empowerment
Democratize agent access so ordinary people have protection, not just corporations