- name
SKILL.md
Le fichier du skill reste dans le dépôt de l'éditeur.
Collez la commande copiée dans un terminal. Elle installe ce SKILL.md dans cet agent.
npx skills add agentsope/SkillAlchemy --skill agentsop-agent-topology-selection -a cursor -yCollez la commande copiée dans un terminal. Elle installe ce SKILL.md dans cet agent.
npx skills add agentsope/SkillAlchemy --skill agentsop-agent-topology-selection -a claude-code -yCollez la commande copiée dans un terminal. Elle installe ce SKILL.md dans cet agent.
npx skills add agentsope/SkillAlchemy --skill agentsop-agent-topology-selection -a codex -yCollez la commande copiée dans un terminal. Elle installe ce SKILL.md dans cet agent.
npx skills add agentsope/SkillAlchemy --skill agentsop-agent-topology-selection -a grok -yEnvoyez le message copié dans un chat Grok Bot. Le Bot récupère ce SKILL.md sur GitHub et l'ajoute aux skills que partagent tous vos Bots.
Install this skill: https://github.com/agentsope/SkillAlchemy/tree/HEAD/skills/agentsop-agent-topology-selectionDe l'éditeur
Overlay posture: this skill decides whether and which topology. It does not teach the API — descend to
[[crewai]]or[[agentsop-langgraph]]for that. Every load-bearing claim carries an inline source tag resolving inreferences/R1-source-evidence.md.
Activate when any of the following fire:
Agent(...) × N,
LangGraph supervisor/swarm, OpenAI Swarm handoffs) and has not yet justified
why a single agent with tools is insufficient.Process.sequential vs Process.hierarchical,
or LangGraph supervisor vs swarm vs hierarchical-teams, and wants the rubric,
not the syntax.Do not activate for: a single LLM call, a one-shot RAG query, or a fixed tool-call pipeline with no role separation. Those are the single-agent baseline this skill defends.
Mental check: "An agent needs agency, otherwise it's just another script." — João Moura, CrewAI founder
[[crewai · §1.3]]. If you can write the control flow inif/else, you do not need multiple agents — you need one agent (or a graph) with explicit edges.
Most "multi-agent" problems are single-agent + tools. Add agents only when context isolation or parallel expertise genuinely demands it.
"Single-agent is right for approximately 80% of cases; the trap is reaching for multi-agent because it sounds more capable."
[[crewai · DC-1]]
Two — and only two — forces justify a second agent:
[[crewai · DC-1]].If neither force is present, a single agent with the union of tools wins — fewer hops, fewer tokens, no handoff failures. This is the baseline the rubric must beat, not the default to escape.
Q0 Is single-agent + tools enough?
(no context-isolation need, no parallel-expertise need)
YES → single-agent + tools. STOP. Do not add agents.
NO → ↓
Q1 Do the agents need to KNOW ABOUT EACH OTHER (peer handoff)?
NO → one funnels through a coordinator → SUPERVISOR
(or static order → SEQUENTIAL, if order is fixed)
YES → ↓
Q2 Must the OUTPUT speak with ONE VOICE / single audit funnel?
YES → SUPERVISOR (single user-facing persona, one funnel)
NO → SWARM (dynamic peer handoff, last-active agent remembered)
Scaling override: ≥6 specialists that group into teams → HIERARCHICAL
(supervisor-of-supervisors). Use only for grouping, not for routing.
The two questions that actually separate the patterns: (a) can sub-agents know
each other, (b) is one user-facing voice mandated. Everything else is tuning
[[langgraph · Case 2]].
Walk top-down. Each gate can send you back down the ladder — collapsing agents is as valid an answer as adding them.
Ask Q0. Enumerate the would-be roles. For each, ask: would merging it into one
agent's prompt + toolset actually degrade output? If you cannot point to a
concrete failure mode (style drift, missed checklist, context bloat, parallel
latency), the honest answer is single-agent + tools. Exit here ~80% of the time
[[crewai · DC-1]].
[[crewai · §2.3]]. In CrewAI this is Process.sequential; in LangGraph it is
static edges A→B→C.Ask Q1 then Q2.
[[langgraph · Step 4]].[[langgraph · §SOP Step 4]].LangChain's own benchmark found swarm "slightly outperformed supervisor across
all scenarios" and supervisor "consistently uses more tokens than swarm" — yet
they still ship supervisor as the recommended default [[langgraph · Step 4]].
Why the nuance matters:
[[langgraph · Step 4]].[[langgraph · Step 4]].Any topology with runtime handoff (swarm, hierarchical, CrewAI delegation) can
loop. Bound it before shipping — cross-link [[agentsop-bounded-loop]]. Concretely:
default allow_delegation=False on workers, set per-agent max_iter, wrap an
outer timeout, and bake an explicit exit counter into state rather than trusting
the LLM to stop [[crewai · DC-5]] [[agentsop-bounded-loop]].
≥6 specialists → group into HIERARCHICAL teams purely for navigability, not
to get free routing (see Anti-patterns). At >5 agents CrewAI starts hitting
coordination failure [[crewai · §6.1]]; that is a signal to group or collapse,
not to add more.
Format: Trigger → Action → Output → Evidence.
[[crewai · DC-1]] "single-agent right for ~80% of cases".[[crewai · §2.2]], [[crewai · DC-1]].context=[...]), do not rely on implicit transfer.Process.sequential or LangGraph static
edges.[[crewai · §2.3]] (sequential = 1× tokens, low debug cost).[[langgraph · Step 4]] decision tree.[[langgraph · Case 2]] (compliance ⇒ supervisor; UX continuity ⇒ swarm).[[langgraph · Step 4]] — swarm beats supervisor on bench, yet
supervisor remains the shipped default for third-party safety.[[langgraph · Step 4]]. Switch paradigm only if still over budget.[[langgraph · Case 2]].allow_delegation=False on workers, per-agent max_iter, outer
timeout, explicit state-based exit counter.[[crewai · DC-5]] (delegation ping-pong), [[agentsop-bounded-loop]].[[langgraph · Step 4]]. A coder
agent copying the default would leave accuracy and tokens on the table.[[langgraph · Step 4]].[[langgraph · Case 2]].Process.hierarchical looks like it
"should auto-route". In practice the manager executes all tasks and the last
task's output overwrites the rest — it does not skip on triage [[crewai · DC-2]].
Query: "Why is my laptop overheating?" (pure technical)
Expected: triage → technical_agent → done
Hierarchical reality: triage → technical → billing → … → last output wins
manager_llm is unreliable; latency/cost
matter.[[crewai · DC-2]].@router
/ @listen) or a LangGraph conditional edge — explicit branch, then call
one small crew / single agent per branch [[crewai · DC-2]].manager_agent carrying an explicit branching backstory — never
the bare default manager_llm [[crewai · OP-4]].allow_delegation=False, outer timeout [[agentsop-bounded-loop]].d-query-routing-skill for the routing-specific rubric.[[crewai · DC-1]].Process.hierarchical (or a
supervisor) to get free if/else routing. It runs everything; routing must be
explicit control flow (Flow / conditional edge) [[crewai · DC-2]].[[langgraph · Step 4]].[[crewai · §6.1]].allow_delegation=True on every agent ⇒ ping-pong
loops. Default off on workers; bound with max_iter + timeout [[crewai · DC-5]],
[[agentsop-bounded-loop]].[[langgraph · Case 2]].Hard boundaries (this rubric does NOT decide):
[[crewai]] vs [[agentsop-langgraph]] ecosystem sections.d-query-routing-skill.[[agentsop-bounded-loop]].[[langgraph · 反模式]].Notes:
hierarchical is coordination,
not routing — the default manager_llm runs all tasks [[crewai · DC-2]].[[langgraph · Step 4, Case 2]].[[langgraph · 生态对照]].Pick the smallest topology that fits the two questions; promote upward only when a named force demands it, and collapse back down when the force disappears.
Inline tags resolve to source-skill sections in references/R1-source-evidence.md:
[[crewai]] = /Users/5imp1ex/Desktop/Skill-Workplace/output/crewai-sop-skill/SKILL.md[[agentsop-langgraph]] = /Users/5imp1ex/Desktop/Skill-Workplace/output/langgraph-sop-skill/SKILL.mdwww.langchain.com/blog/benchmarking-multi-agent-architectures), surfaced via
[[langgraph · Step 4 / Case 2]].>-
---name: agentsop-agent-topology-selectiondescription: >----# skills/agentsop-agent-topology-selection/SKILL.mdFAQ
Téléchargez skills/agentsop-agent-topology-selection/SKILL.md auprès de l'éditeur, ou copiez le chemin et ouvrez le dépôt. PlugMCP répertorie le skill et n'héberge pas le fichier.
L'éditeur le maintient sur https://github.com/agentsope/SkillAlchemy.
Claude, Cursor, Codex et Grok Build peuvent lire un SKILL.md que vous ajoutez depuis le dépôt de l'éditeur. Grok Bot l'installe lorsque vous envoyez dans un chat le lien du dossier GitHub du skill.