Claude MCP - Lab Building

The Rogue Arena MCP plugin lets Claude Code drive Rogue Architect end-to-end. Describe the lab you want, and Claude handles the canvas — VLANs, machines, plugins, AD topologies, exploit paths, the works.

This page covers the scenario-building skills that ship with the plugin. (For the deployment-side skills, see Claude MCP - Active Deployments. For plugin authoring, see Claude MCP - Plugin Dev.)

/architect-brainstorm

The gold standard for going from nothing to a mature, fully built scenario. Brainstorms the canvas end-to-end through a guided four-phase pipeline: company context and culture, network/infrastructure design, rich characters and backstories (logged-in users, mailbox content, relationships, mission), and optional exploit paths.

It doesn’t just lay out boxes — it seeds the scenario with realism. Random user files on desktops and Downloads, browser histories that match the persona, scheduled tasks, mail correspondence, file-share layouts — everything a believable lab needs so it doesn’t feel like a sterile sandbox.

Produces scenario_part1-3.yml + exploit.yml, then hands off to architect-validator automatically on Phase 4 completion. From there, architect-implementor builds the real infrastructure. End result: you describe the lab once and walk away with a real, deployable scenario.

/architect-freeform

Manual canvas work with the MCP tools — no guided pipeline. Use it to add machines, swap plugins, wire VLANs, or tweak configuration on an existing canvas. Best when you know exactly what you want to change and just need Claude to do the keystrokes.

/architect-final-validate

Read-only pre-deploy audit. Walks every machine, every plugin, and the exploit path traces; checks plugin coupling, run-order, infrastructure correctness, and realism. Use this right before launching a build to catch issues a human eye would miss.

/architect-debug-deploy

Investigates deployment failures — Ansible errors, PowerShell debugging, plugin misconfigurations. Requires a canvas ID. Drop it in when a build errored and you want Claude to triage the failure end-to-end (pull the Ansible log, identify the failing task, propose a fix).

Auto-chained skills

Two more skills ship with the bundle but aren’t user-invocable — they run automatically when their parent skill completes:

• /architect-validator — runs after /architect-brainstorm Phase 4 to validate the produced YAML against mechanical invariants and cross-phase consistency
• /architect-implementor — dispatched after /architect-validator passes; expands scenario.yml + exploit.yml into implementation.yml and builds the real infrastructure via MCP tools

You don’t call these yourself — they chain in the background as part of the brainstorm-to-build pipeline.