---name: mint-threejs-skillsdescription: Build, revise, debug, and verify browser-based Three.js apps, games, asset viewers, model and asset-pack deliveries, material and material-pack deliveries, animated-model viewers, and explicitly requested Gaussian-splat worlds with Mint MCP as the production asset pipeline.---

Mint Three.js Skills

Mint's agent skill suite for browser-based Three.js work. It combines and adapts open source Three.js game, application, and graphics-agent workflows, generalized for 3D apps while retaining deep game workflows.

Choose A Route

• Fresh request whose primary deliverable is one generated model, an animated model, a coherent model asset pack, one material, a material pack, or one explicitly requested Mint world: read references/asset-viewer.md, then use the app director. The asset is the product; the viewer is its inspection and download shell.
• General 3D app, viewer, configurator, simulation, walkthrough, editor, or interactive experience: read skills/threejs-app-director/SKILL.md.
• Game or game-like request with objectives, challenge, scoring, failure, progression, or game feel: read skills/threejs-game-director/SKILL.md.
• Mixed experience: start with the app director, then add only the relevant game specialists.

Both routes share visual systems, interaction, debugging, QA, and references/mint-mcp-assets.md. Before the first Mint MCP write for a codebase, read references/mint-project-workspaces.md to resolve one Mint Project for the work. Projects that use Mint assets also use the durable registry in references/asset-pipeline.md.

Existing app or game context wins over asset delivery. Generate and integrate a requested asset into that project instead of scaffolding a separate viewer.

Invariants

• Existing project architecture wins. For greenfield work, default to TypeScript, Vite, and Three.js modules.
• Vanilla Three.js is the default; support React Three Fiber or another Three.js-based stack when the project or user chooses it.
• Mint MCP is the only generated-asset production pipeline. Keep MCP calls out of browser runtime code.
• Distinguish the local code project (the folder or repository being edited) from its Mint Project (the remote workspace that owns Mint chats and assets). Resolve or create the Mint Project once, persist it in mint-assets.json, and reuse it for every independent generation started for that codebase.
• For every project that imports Mint files or remote world configuration, maintain a project-root mint-assets.json through scripts/sync-mint-assets.mjs. Reuse stable logical keys and preserve the existing project's asset-root conventions.
• For every Three.js path that loads Mint-generated GLBs, read references/gltf-runtime-compatibility.md and use a Draco-capable shared loader. Mint-optimized GLBs are not compatible with a bare GLTFLoader.
• Prefer discrete generated models and compose them in Three.js. Generate a Mint world only when the user explicitly chooses a generated environment; then read references/mint-world-splats.md.
• Use procedural or user-provided assets when they are the right design choice or Mint MCP lacks the required capability. Never create a competing procedural version of a subject that Mint generated successfully.
• Before verification, read references/verification-policy.md. Run its automatic minimum, ask before extended desktop/browser QA, and require a separate secondary approval for mobile QA. Its approval boundary overrides broader specialist completion gates.
• Do not force game concepts such as objectives, pressure, rewards, or failure onto general 3D apps.

User-Owned UI

• Treat the delivered app as the user's product, not as a demo of the asset-generation pipeline.
• Keep provider names, branding, badges, generation links, asset IDs, and provenance out of runtime UI unless the user explicitly asks for them.
• Mention generation provenance and handoff links only in the final response or developer documentation.
• Default to the minimum UI required for the experience: loading and error status, essential controls, and explicitly requested actions.
• Do not add headers, title bars, navigation, marketing copy, attribution, or decorative application chrome unless requested.
• For the canonical asset viewer, keep the canvas dominant with a compact centered details dialog opened from an info button plus bottom-centered inspection controls. Do not reserve permanent sidebar space. For other simple viewers and walkthroughs, use a compact bottom-centered control group. Place loading, ready, or error status directly above it using the same compact visual language.

When the user asks for a reusable prompt, use references/request-templates.md.