Agent Topology Audit

Agent Topology Audit

Most advice about multi-agent systems is about how to add agents. This is about whether to. It audits a setup you already have, counts what each boundary between agents costs, and compares that against what the boundary buys. Removing an agent is a valid, and often the correct, result.

The framing comes from Liu, Canhui (2026), The Organizational Behavior of Agentic AI (arXiv:2606.30986), which models coordination overhead as contextual transaction cost — the cost of making task context usable across an agent boundary.

When To Use

• Before adding another agent, seat, or phase to a workflow that already works.
• When a workflow is slow and it is not obvious which part is earning its time.
• When agents keep agreeing. Agreement that costs three dispatches and produces what one would have produced is overhead wearing the costume of consensus.
• When a handoff keeps losing something and the fix keeps being "add more context to the prompt".
• After a workflow produced a bad result and you want to know whether the topology or the models were at fault.

When Not To Use

• To pick a workflow for a new task. That is /octo:auto, which already routes by intent, or skill-decision-support for a general option comparison.
• To decide whether to delegate a task to agents at all. That is the allocation step in skill-intent-contract.
• To choose between providers or models. See skills/blocks/frontier-model-routing.md.
• For a single-agent task. There are no boundaries to count.

Inputs

• The workflow or setup under audit: which agents or seats, in what order, with what passing between them.
• What each agent receives and what it returns. Prompt and output shape matter more than model identity here.
• Optionally, a transcript or run directory, which turns estimates into observations.