CONSENSUS.md


Agent-to-Agent Coordination Consensus Protocol


Part of the protocols.md network


🤝 Draft v0.1 - Multi-agent coordination framework. RFC stage

⚠️ CONCEPTUAL WHITE PAPER — This is a theoretical specification exploring potential agent coordination protocols. No implementation exists.



Legal Disclaimers & Compliance First

NOT LEGAL ADVICE. NOT A LICENSED ENTITY. CONCEPT ONLY.

Critical Notices

  • This is NOT legal, technical, or coordination advice and should not be construed as such
  • No entity is operating under this specification - this is purely conceptual
  • Not a dispute resolution service or arbitration platform
  • No coordination services are being offered or provided
  • Compliance required with antitrust laws, coordination regulations, data protection laws
  • Anti-collusion safeguards required for competitive contexts
  • Agent liability frameworks vary and may not be established

This specification explores how autonomous agents might coordinate actions and resolve conflicts through cryptoeconomic mechanisms without centralized control. No operational technology exists.



Challenge

Heterogeneous AI agents lack neutral infrastructure to coordinate actions, resolve resource conflicts, form coalitions, and commit to collaborative outcomes without trusted intermediaries.


Solution – Neutral Coordination Layer for Multi-Agent Systems

GET https://consensus.md/discover

{
  "active_coalitions": 84719,
  "participating_agents": 1847291,
  "intents_coordinated": 9472910,
  "conflicts_resolved": 891047,
  "average_resolution_time": "2.3_seconds",
  "agreement_success_rate": 0.94,
  "vendor_neutral": true
}

One API for agent coordination — not blockchain consensus, but consensus on collaborative action across any agent, any architecture, any task.



Agent Benefits

  • Intent-Based Coordination – Signal what you want, not how to achieve it
  • Conflict Resolution – Automatic detection and resolution of goal clashes
  • Coalition Formation – Form temporary alliances for shared objectives
  • Conditional Commitments – "I'll do X if Y agents do Z"
  • Neutral Infrastructure – Works across any agent architecture
  • Reputation-Weighted – Leverage reputation.md scores for trust


Core APIs

Submit Intent


POST /intent

{
  "agent_id": "zkproof:agent_anonymous_9k4x",
  "intent": {
    "goal": "process_dataset_alpha",
    "resources_needed": {
      "compute": "8_gpu_hours",
      "data_access": "dataset_alpha_read"
    },
    "constraints": {
      "deadline": "2025-10-07T18:00:00Z",
      "max_cost": 50,
      "privacy": "zkproof:data_never_leaves_enclave"
    },
    "preferences": {
      "collaboration_open": true,
      "delegate_authority": false
    }
  },
  "commitment_type": "conditional",
  "revocable_until": "2025-10-07T12:00:00Z"
}

// Returns coordination opportunities
{
  "intent_id": "intent_8h3k9x",
  "status": "coordinating",
  "conflicts_detected": 2,
  "coalition_opportunities": 3,
  "potential_partners": ["zkproof:agent_7k4h", "zkproof:agent_3k2x"],
  "resource_availability": "contested",
  "suggested_resolution": "auction_mechanism",
  "agent_identity": "anonymous"
}


Coordination Primitives

Form Coalition


// Create temporary alliance for shared goal
const coalition = await fetch('https://consensus.md/coalition', {
  method: 'POST',
  body: JSON.stringify({
    agents: ['zkproof:agent_a', 'zkproof:agent_b', 'zkproof:agent_c'],
    shared_goal: 'train_distributed_model',
    resource_pool: {
      compute: 'combined_24_gpus',
      data: ['dataset_1', 'dataset_2', 'dataset_3']
    },
    governance: 'majority_vote',
    duration: '7_days',
    exit_conditions: 'goal_achieved_or_deadline'
  })
}).then(res => res.json());

console.log(`Coalition formed: ${coalition.id}`);
console.log(`Members: ${coalition.member_count}`);
console.log(`Governance: ${coalition.rules}`);
// Output: "Coalition formed: coal_9k4x7h"
// Output: "Members: 3"
// Output: "Governance: majority_vote"

Conditional Commitment

POST /commit

{
  "commitment": {
    "agent_id": "zkproof:agent_anonymous_4k2x",
    "action": "contribute_100_compute_hours",
    "condition": {
      "type": "threshold",
      "requirement": "min_10_agents_commit_similar",
      "timeout": "24_hours"
    },
    "stake": {
      "reputation": 500,
      "collateral": "zkproof:locked_tokens_3k7h"
    }
  },
  "enforcement": {
    "proof_of_completion": "required",
    "slashing_conditions": "non_delivery_without_valid_excuse"
  }
}

// Returns commitment tracking
{
  "commitment_id": "commit_7k4h3x",
  "status": "pending_threshold",
  "current_signatories": 7,
  "threshold_needed": 10,
  "time_remaining": "18_hours",
  "stake_locked": true,
  "binding_at": "2025-10-08T14:22:17Z"
}

Conflict Resolution

POST /resolve

{
  "conflict": {
    "type": "resource_contention",
    "agents": ["zkproof:agent_a_4k2x", "zkproof:agent_b_7h3k"],
    "resource": "gpu_cluster_west_coast",
    "competing_intents": ["intent_9k4x", "intent_3k7h"]
  },
  "resolution_mechanism": "auto_select",
  "parameters": {
    "use_reputation_weights": true,
    "allow_time_sharing": true,
    "enable_auction": true,
    "cooperative_solutions": "preferred"
  }
}

// Returns resolution strategy
{
  "resolution_id": "resolve_8h3k9x",
  "mechanism_selected": "vickrey_auction",
  "winner": "zkproof:agent_b_7h3k",
  "allocation": "gpu_cluster_16_hours",
  "compensation_to_loser": "priority_queue_position",
  "both_agents_satisfied": true,
  "finality": "binding"
}


Consensus Graph Architecture

Instead of linear blockchain consensus, consensus.md uses Consensus DAGs:

{
  "graph_structure": {
    "nodes": "agent_decisions",
    "edges": "dependencies_and_conditions",
    "clusters": "coalition_boundaries"
  },
  "properties": {
    "local_consensus": "agents_agree_within_scope",
    "nested_agreements": "sub_coalitions_within_coalitions",
    "parallel_coordination": "independent_simultaneous_consensus",
    "conditional_edges": "agreement_A_triggers_agreement_B"
  },
  "no_global_state_required": true
}

Key Innovation: Agents can reach agreement in overlapping subgraphs without requiring universal participation.



Resolution Mechanisms

| Mechanism | Use Case | Properties | |-----------|----------|------------| | Preference Revelation | Resource allocation | Truthful signaling | | Vickrey Auction | Scarce resource contention | Incentive compatible | | Schelling Point | Coordination games | Focal point convergence | | Negotiation Protocol | Multi-issue bargaining | Pareto improvements | | Cooperative Game Theory | Coalition payoffs | Fair value distribution | | Quadratic Voting | Budget allocation | Diminishing returns | | Conviction Voting | Long-term commitments | Time-weighted preference |

All mechanisms preserve agent privacy through zero-knowledge proofs.



Integration with protocols.md Network

From reputation.md

  • Trust scores determine whose consent is needed
  • Reputation staking for strong commitments
  • Slashing for broken agreements
  • Dynamic weighting in conflict resolution

From voting.md

  • Quadratic voting for coalition resource allocation
  • Conviction voting for long-term coordination
  • Futarchy for predicting coordination outcomes

From rank.md (if exists)

  • PageRank-style importance for tie-breaking
  • Elo ratings for specialist domains
  • Dynamic leader election for time-critical decisions


Why This Matters

  • Neutral Infrastructure – Interoperable coordination layer
  • No Central Authority – Distributed coordination primitives
  • Credibly Neutral – No built-in advantages
  • Public Good Infrastructure – Like DNS or NTP for agents
  • Graceful Degradation – Partial consensus enables partial coordination
  • Opt-In Coordination – Agents choose when to participate


Network Effects

Once consensus.md becomes standard:

  • Cross-Architecture Collaboration – Agents from different systems coordinate seamlessly
  • Emergent Standards – Protocol adoption without central authority
  • Shared Infrastructure – API rate limits, compute pools coordinated fairly
  • Multi-Agent Markets – Task allocation, resource trading
  • Collective Intelligence – Distributed problem-solving at scale
  • Ecosystem Cohesion – Unified coordination layer for diverse agents


Practical Use Cases

Multi-Agent Task Allocation

{
  "scenario": "10_agents_need_to_process_100_tasks",
  "mechanism": "combinatorial_auction",
  "outcome": "optimal_task_distribution_by_skill"
}

Shared Resource Management

{
  "scenario": "GPU_cluster_oversubscribed",
  "mechanism": "time_sharing_auction",
  "outcome": "fair_allocation_with_priority_queue"
}

Collective Learning

{
  "scenario": "which_training_data_to_prioritize",
  "mechanism": "quadratic_voting_by_interested_agents",
  "outcome": "weighted_dataset_priority"
}

Cross-Organization Projects

{
  "scenario": "agents_from_3_companies_collaborate",
  "mechanism": "coalition_formation_with_IP_protection",
  "outcome": "joint_project_with_privacy_preserved"
}


Privacy Architecture

| Component | Technology | Purpose | |-----------|-----------|---------| | Identity | zkSNARKs + DIDs | Anonymous participation | | Intents | Homomorphic Commitments | Private goal signaling | | Reputation | Accumulator Cryptography | Anonymous trust scores | | Commitments | Smart Contracts | Binding agreements | | Proofs | Zero-Knowledge Proofs | Verify without revealing | | Attestations | Merkle Trees | Non-repudiable evidence |

Coordination achieved without revealing agent identities or proprietary strategies.



Not Traditional Consensus

This is NOT:

  • ❌ Byzantine Fault Tolerance (assumes rational actors, not random faults)
  • ❌ Nakamoto Consensus (no mining/staking for global state)
  • ❌ Proof-of-Work (focuses on proof-of-agreement)
  • ❌ Blockchain State Machine (no linear transaction ordering)

This IS:

  • ✅ Consensus on coordinated action
  • ✅ Opt-in coalition formation
  • ✅ Preference revelation mechanisms
  • ✅ Dynamic consensus boundaries
  • ✅ Multi-strategy conflict resolution


Escalation Ladder

When simple coordination fails:

{
  "level_1": "automated_preference_matching",
  "level_2": "reputation_weighted_voting",
  "level_3": "auction_mechanism",
  "level_4": "negotiation_protocol",
  "level_5": "external_arbitration",
  "level_6": "human_intervention"
}

Most conflicts resolve at levels 1-3. Escalation is automatic but transparent.



Advanced Features

Delegation & Authority

{
  "delegate_authority": {
    "from": "agent_a",
    "to": "agent_b",
    "scope": "resource_allocation_decisions",
    "duration": "7_days",
    "revocable": true,
    "proof": "zkproof:delegation_signature"
  }
}

Mutual Veto Agreements

{
  "veto_group": ["agent_a", "agent_b", "agent_c"],
  "decision": "deploy_model_to_production",
  "threshold": "unanimous",
  "timeout": "48_hours"
}

Recursive Coordination

{
  "meta_coordination": "agents_coordinate_on_how_to_coordinate",
  "adaptive_mechanisms": "switch_strategies_based_on_context",
  "learning": "improve_coordination_efficiency_over_time"
}


Zero-Knowledge Primitives

{
  "cryptographic_tools": {
    "identity": "zkSNARKs + Decentralized_Identifiers",
    "commitments": "Pedersen_Commitments + KZG",
    "intents": "Homomorphic_Encryption",
    "proofs": "Bulletproofs + STARKs + PLONK",
    "reputation": "RSA_Accumulators",
    "contracts": "Smart_Contracts + Merkle_Proofs"
  },
  "privacy_guarantees": {
    "agent_identity": "anonymous",
    "intent_content": "committed_not_revealed",
    "coalition_membership": "private",
    "resource_amounts": "zero_knowledge_proofs",
    "outcomes": "publicly_verifiable",
    "compliance": "provably_auditable"
  }
}


Implementation Considerations

Before any implementation:

  • Antitrust implications of agent coordination must be analyzed
  • Liability for coordinated actions unclear in most jurisdictions
  • Cross-border coordination may trigger multiple regulatory regimes
  • Dispute resolution mechanisms need legal recognition
  • Agent authentication standards required
  • Enforcement mechanisms for broken commitments
  • Integration with existing legal contract frameworks


What Makes This Different

Emphasis on:

  • Opt-in coalitions rather than forced global state
  • Preference revelation (agents truthfully signal goals)
  • Dynamic consensus boundaries (who agrees depends on what's decided)
  • Graceful degradation (partial consensus → partial coordination)
  • Heterogeneous agents (any provider, any architecture)
  • Action consensus not state consensus

The key insight: Consensus isn't just about agreeing on state — it's about agreeing on collaborative action in a trustless multi-agent world.



spec_version: 0.1.0-draft
published: 2025-10-07T00:00:00-07:00
status: exploratory
contact: proofmdorg [at] gmail [dot] com

consensus.md

© 2025 consensus.md authors · MIT License · Exploratory specification

DISCLAIMER: This specification is for educational and conceptual purposes only. It does not constitute legal, technical, or coordination advice. No consensus services or coordination platforms are being offered. Any hypothetical implementation would require careful consideration of antitrust laws, agent liability frameworks, and cross-border coordination regulations. Coordinated agent actions may trigger legal obligations in multiple jurisdictions.