---
name: topology-mutation
description: Design or revise the communication topology of a multi-agent system based on task dependencies, specialization, latency, and failure isolation. Use for agent-team architecture and orchestration experiments.
---

# Topology Mutation

Choose how agents communicate, delegate, review, and converge.

## Workflow

1. List the task stages, dependencies, specialist roles, and shared resources.
2. Choose a baseline topology: supervisor, pipeline, debate, mesh, or independent ensemble.
3. Identify bottlenecks, single points of failure, redundant communication, and missing review edges.
4. Mutate one structural property at a time: add a reviewer, split a role, remove an edge, or change aggregation.
5. Predict the effect on quality, latency, cost, and failure containment.
6. Recommend the smallest topology that satisfies the task.

## Output

Return the proposed roles, directed communication edges, aggregation rule, expected tradeoffs, and one experiment comparing it with the baseline.

Avoid a fully connected mesh unless every agent genuinely needs every other agent’s output.

## Research basis

[Autonomous Topology Mutation: Safe Runtime Restructuring for Multi-Agent LLM Systems with Capability, State, and Shadow Invariants](https://arxiv.org/abs/2607.20488) — Sidik, Levi, and Kimhi, arXiv (2026). The paper introduces runtime topology mutation for multi-agent systems.