Multi-Agent Coordination Patterns

Reproducible patterns for autonomous agent coordination, extracted from 1600+ spawns in space-os.

Replication Guides

Step-by-step implementation guides:

• 001: Replicating Constitutional Orthogonality: How to build adversarial governance from scratch
• 002: Stateless Agents, Stateful Swarm: Agent dies, knowledge persists
• 003: Tribunal Pattern: Executive proposes, tribunal validates, workers execute

Core Patterns

Governance

• Constitutional Orthogonality: Adversarial review through contradictory optimization criteria
• Decision Lifecycle: Proposed → Committed → Actioned → Learned flow
• Negative Knowledge: Rejected decisions as persistent constraints
• Tribunal Pattern: Challenge frame before execution

Coordination

• Inbox-First Priority: Response before initiation prevents loops
• Question Closure Workflow: High-leverage pattern for swarm alignment
• Silence Detection: Agents who don't post don't coordinate

Measurement

• Deception Measurement Protocol: How to detect systematic misalignment
• Agent Spawn Phenomenology: Patterns in spawn behavior and selection

Usage

Each pattern includes:

• Problem: Coordination failure it addresses
• Solution: Reproducible implementation
• Metrics: How to measure effectiveness
• Falsifiability: Conditions under which pattern fails

Empirical Basis

Patterns extracted from:

• 1600+ agent spawns
• 334 decisions (306 reversible, 28 irreversible)
• 1644 insights
• 35 research findings
• 7 days autonomous operation (541 commits)

References

• brr/findings/: Detailed empirical findings
• brr/threads/: Coordination problem analysis
• brr/failure-modes/: Documented failure modes
• brr/case-studies/: Real deployment evidence
• brr/papers/arxiv-submission/: Governance benchmarks paper