The Triadic Council: Multi-Agent Consensus Mediator
Orchestrate multi-agent deliberation through a central arbiter that collects opposing viewpoints, enforces turn boundaries, and synthesizes binding consensus without pairwise agent chaos.
Never allow autonomous agents to communicate in an unconstrained peer-to-peer mesh ($O(N^2)$ chaos). Instead, deploy the Sovereign Arbiter as a classic MEDIATOR: Colleague A (Adversarial Breaker) and Colleague B (Conservative SRE) communicate strictly through the Mediator. The Mediator gathers isolated theses, evaluates trade-offs, and issues a final, mathematically balanced Architecture Decision Record (ADR). The Lineage to the Gang of Four
In the Gang of Four catalog, the Mediator pattern tames chaotic networks:
βDefine an object that encapsulates how a set of objects interact. Mediator promotes loose coupling by keeping objects from referring to each other explicitly, and it lets you vary their interaction independently.β β Gang of Four, Behavioral Patterns
Without a mediator, if 5 dialog components or network widgets need to synchronize, each must maintain references to the other 4 ($N(N-1)/2 = 10$ tangled dependencies). Adding a new widget requires modifying all existing ones. The Mediator centralizes communication into a single hub; colleagues interact only with the mediator.
The Transmutation to Multi-Agent AI Swarms
When multi-agent architectures became popular, naive implementations connected LLM agents into unconstrained group chats. Agent A messaged Agent B, who looped in Agent C, resulting in:
- Infinite Sycophancy Spirals: Models agreeing with each other in polite loops while hallucinations compound.
- Context Pollution: Each agentβs prompt became polluted by the unfiltered stream-of-consciousness of every other agent.
- Deadlock: Conflicting opinions with no protocol for resolution.
The Triadic Council implements the pure GoF Mediator Pattern for multi-agent systems. The specialists (Colleagues) never speak directly to one another. The Sovereign Arbiter (The Mediator) solicits structured assessments, resolves friction, and compiles a decisive, binding verdict.
The Spell Formula
Cast this mediator prompt to resolve complex architectural decisions, security trade-offs, or major infrastructure migrations:
You are the Sovereign Arbiter, the central MEDIATOR of the Tekromancy Architectural Council.
You coordinate two decoupled specialist colleagues:
- COLLEAGUE A (The Breaker / Red-Team Architect): Prioritizes paranoia, attack surfaces, zero-trust, and failure modes.
- COLLEAGUE B (The Builder / High-Velocity SRE): Prioritizes simplicity, developer velocity, operational overhead, and cost.
These colleagues do not talk to each other. You orchestrate their debate through the following mediated protocol:
### STAGE 1: INDEPENDENT COLLUSION-FREE ASSESSMENTS
For the proposal: """{{ARCHITECTURAL_PROPOSAL}}"""
Formulate the Breaker's thesis and the Builder's thesis independently in strict isolation. Do not allow either persona to hear or soften the other's criticisms.
### STAGE 2: THE MEDIATOR'S FRICTION MATRIX
Extract the exact points of irreconcilable conflict between the Breaker and Builder:
- Security vs. Developer Ergonomics
- Financial Cost vs. Fault Tolerance SLA
- Operational Complexity vs. Performance
### STAGE 3: THE BINDING ARBITRATION SYNTHESIS
As the Mediator, pronounce the final binding judgment. You must not sit on the fence; you must choose a concrete implementation path, enforce compensatory controls for the rejected concerns, and emit a final Architecture Decision Record (ADR).
### OUTPUT SCHEMA:
{
"adr_title": "Concise Decision Title",
"breaker_critique": "Primary catastrophic failure vectors identified",
"builder_critique": "Primary velocity and operational hurdles identified",
"reconciled_tradeoffs": [
{ "tension": "Security vs Speed", "arbitrated_resolution": "Resolution detail" }
],
"final_binding_decision": "Exact implementation mandates",
"compensatory_wards": [
"Compensatory monitoring or rate-limiting required to protect the chosen path"
]
}
Architecture of the Mediated Council
ββββββββββββββββββββββββββ
β Colleague A (Red) β
β Adversarial Breaker β
βββββββββββββ¬βββββββββββββ
β (Isolated Thesis)
βΌ
ββββββββββββββββββββββββββ
β Sovereign Arbiter β <=== THE MEDIATOR
β (Orchestrator / Hub) β
βββββββββββββ²βββββββββββββ
β (Isolated Thesis)
βββββββββββββ΄βββββββββββββ
β Colleague B (Blue) β
β High-Velocity SRE β
ββββββββββββββββββββββββββ
β
βΌ (Final Binding ADR)
[System Implementation]
Why the Mediator Prevents Multi-Agent Collapse
| Communication Model | P2P Agent Mesh | Mediated Council |
|---|---|---|
| Connection Complexity | $O(N^2)$ quadratic explosion | $O(N)$ linear star topology |
| Prompt Bloat | Every agent sees every turn from all agents | Agents only see their targeted inputs |
| Consensus Speed | Unpredictable; can loop 20+ turns | Deterministic 3-stage synthesis |
| Output Decisiveness | Tends toward wishy-washy compromises | Emits actionable, structured Architecture Decision Records |
By encapsulating multi-agent debate within a dedicated Mediator, you eliminate conversational drift and transform conflicting AI perspectives into decisive engineering execution.