Architecture advice, no generation

This skill advises. It does not create files. For generation, use /trabuco:new-project, /trabuco:design-system, or /trabuco:add-module.

Flow

1. Delegate to the specialist if possible: invoke the trabuco-architect subagent. It loads the trabuco_expert MCP prompt and has the full reasoning framework. Only handle in-skill if the subagent isn't available.

1. Gather the requirement if missing: 2–3 tight questions, not a survey. Focus on what differentiates architectures:

• Data shape (relational / document / cache / mix)
• Traffic shape (sync REST, async events, scheduled jobs, long-running tasks, real-time streams)
• AI agent involvement (will users interact with an LLM-backed agent?)
• Scale posture (small internal tool / B2B service / large-scale SaaS)

1. Get the recommendation: call mcp__trabuco__suggest_architecture with the requirement. It returns ranked patterns + a recommended module config.

1. Present as reasoned advice, not a list dump:

• Top recommendation with rationale grounded in the returned patterns
• What tradeoff you're making (e.g., "Kafka gives durability but costs operational complexity vs RabbitMQ")
• What you'd change if the requirements shifted (e.g., "if latency matters more than throughput, I'd drop EventConsumer and stick with REST + async jobs")

1. Cite sources: when you make a claim about what a module does, reference trabuco://modules or list_modules output. When you cite a pattern, reference trabuco://patterns. Don't hallucinate.

1. Offer to act: "If you want me to generate this, use /trabuco:new-project" — but only if they ask. Don't push.

Rules

• No generation from this skill. Period. The user picks when to commit.
• Be honest about fit. If Trabuco isn't the right tool (read trabuco://limitations), say so — "for a React frontend, Trabuco won't help; pair it with a Next.js or Vaadin service."
• Don't over-architect. If a single-module service is enough, say that. Resist the urge to recommend 8 modules when 3 suffice.