Choose Your Agent Topology Before Writing an Agent

Where this fits

Your coding agent is building a research assistant, and at the first design step it announces: "I'll create three agents — researcher, writer, reviewer." Does that task actually need coordination, shared state, or one unified output? This skill intercepts that moment. It is an overlay that applies a binary-question rubric — single-agent + tools enough? do agents need to know about each other? must the output have one voice? — and maps the answers to single-agent, supervisor, swarm, sequential, or hierarchical before any agent code is written.

Why agents benefit

• It turns "do I need multiple agents?" from a vibe into a decision rubric; single-agent + tools is the default, and a topology is only justified when the answers force it.
• It activates at the right trigger: when a coder agent reaches for a multi-agent framework or is tempted to "split the work into roles."
• It is overlay-only: it decides whether and which topology, then descends to CrewAI or LangGraph for implementation, so it doesn't duplicate framework docs.
• It produces explainable choices: each load-bearing claim carries an inline source tag, so the agent can justify why a supervisor beat a swarm.
• It prevents rework: choosing sequential or hierarchical up-front is cheaper than rewiring tangled agent communication later.

Practical scenarios

Defaulting to multi-agent when tools would do

A developer asks for extraction plus summarization. The agent proposes separate roles. The rubric asks: does the output need one voice? No. So a single agent with two tools is the answer — no orchestration overhead.

Supervisor vs swarm for parallel research

The agent needs to run ten independent research queries. Agents don't need to know about each other, but results must be aggregated. The rubric maps this to supervisor, not swarm, because a coordinator collects and merges rather than letting agents hand off autonomously.

Promoting a linear pipeline to hierarchical

A content pipeline has writer and reviewer agents that keep editing each other's output. When the final piece must have one authoritative voice and review quality matters, the rubric selects hierarchical: a senior agent owns the final output while sub-agents draft and critique.

Add it to your agent workflow

Install the skill into your agent's skill directory:

npx skills add agentsope/SkillAlchemy --skill agentsop-agent-topology-selection

Then, when a task starts to look multi-agent, prompt: "Before coding agents, apply agentsop-agent-topology-selection." The skill returns a topology and cross-links to the framework skill for the actual implementation.

Compare before adopting

Before adopting, compare it with your framework's own guidance: does CrewAI or LangGraph already force this decision for you? Look at maintenance signals — repo freshness, stars (currently 357), issue responsiveness — and at alternatives like starting single-agent and scaling only when data shows you need more. Check workflow fit: since this is an overlay, it needs a skill resolver that honors cross-links like [[crewai]] and [[agentsop-langgraph]].

Why it is worth tracking

This skill is worth tracking because it fills a real gap: most multi-agent frameworks teach APIs, but none surface the selection rubric that tells you which topology to use. With a conservative "single-agent first" default, evidence tags, and a clear activation trigger, it encourages restraint instead of layering agents onto every task. The 357-star repository shows community momentum. Re-evaluate it when your agents routinely add roles before checking whether the topology is necessary; decide whether to include it in your standard workflow. For ongoing context, see the skill page at https://www.openagentskill.com/skills/agentsope-agentsop-agent-topology-selection.