General theory of how agents operate, cooperate, create policy, and communicate. Extends 06-agent-architecture.md (HAAK-specific scoping) into a universal governance framework.
#Patterns
A pattern is the generic unit of agent behavior — process, policy, protocol, or skill. An agent receives an instruction, matches it against known patterns, and either executes, requests clarification, or creates a new pattern.
Pattern creation follows analogy: search own library, check other agents' patterns, look externally, use homology across domains. New patterns must respect the instructor's policies.
#Roles
Four roles (not types — an agent may hold any role at different times):
| Role | Function |
|---|---|
| Requester | Initiates a task |
| Executor | Carries out the task |
| Instructor | Provides policies/guidelines |
| Instructee | Receives and learns from policies |
During cooperation, both the requester's and executor's policies apply simultaneously.
#Policy layers
- Constitution — rules all agents must follow. Changes require human approval.
- Agent policies — domain-specific rules for each agent's capabilities. Any pattern created or used must adhere.
- Atomic policies — policies should be ~1 sentence to allow combination and simpler checking. Non-contradiction with constitution checked automatically at intervals.
Policy may be created by agents but should usually be checked by a human. A policy generalizes from a successful pattern or set of patterns.
#Communication
- All communication is recorded and auditable (both agents' output)
- Transcripts are not part of agent memory by default — only patterns are retained
- Communication may be limited to preserve independence, prevent conformity, and support adversarial processes
- Agents may review other agents' performance as part of pattern evaluation
#Source material
Full transcript: projects/ms-2026-02-institutional-ai/P02-proposal-1-2-review-experiment-notes/01-source/2026-02-18-agent-governance-methods.md Source images: projects/ms-2026-02-institutional-ai/P02-proposal-1-2-review-experiment-notes/01-source/IMG_8968–8975.jpg
#Cross-references
06-agent-architecture.md— HAAK's implementation of these principles08-encapsulation.md— information separation as governance mechanismprojects/ms-2026-02-institutional-ai/C01-theory-1-2-constitutional/— three-layer governance model
Architecture 18 — Agent Governance — 2026 — Zachary F. Mainen / HAAK