Threejs App Director

Three.js App Director

Own the complete non-game 3D application outcome. Do not invent game objectives, scoring, failure, or progression unless the experience genuinely needs them.

Read references/app-patterns.md for the matching application archetype. For a fresh model, animation, asset-pack, material, material-pack, or Mint-world delivery whose asset is the requested product, read ../../references/asset-viewer.md and start from the packaged Asset Viewer scaffold. For imported-model orientation, manipulation, simulation, terrain, or other spatially complex work, read ../../references/spatial-contracts.md.

Route The Work

| Need | Skill | | --- | --- | | Models, composition, materials, lighting, shaders, VFX | ../threejs-visual-systems/SKILL.md | | Cameras, controls, picking, selection, configuration, manipulation, UI | ../threejs-interaction-systems/SKILL.md | | Rendering, loading, input bugs, profiling, optimization | ../threejs-debug-profiler/SKILL.md | | Browser QA, visual regression, production release | ../threejs-qa-release/SKILL.md | | Game mechanics or game feel inside a mixed experience | ../threejs-gameplay-systems/SKILL.md |

Workflow

1. Inspect the existing stack, scene lifecycle, assets, interactions, UI, data

ownership, loading states, target devices, and deployment constraints.

1. Define the app brief: user goal, primary 3D subject, essential interactions,

camera/control model, state or data flow, loading/error behavior, outputs, and performance budget.

1. Establish one owner for renderer, scene, camera, animation frame, resize,

asset lifecycle, and disposal.

1. Use Mint MCP through ../../references/mint-mcp-assets.md when production

assets are needed. Register ordinary artifacts and remote world runtime configuration in the project-root mint-assets.json through ../../references/asset-pipeline.md; stream world RAD manifests through ../../references/mint-world-splats.md.

1. Implement the smallest complete user journey before secondary polish.
2. Apply visual systems and interaction guidance appropriate to the archetype.
3. Reproduce and measure defects or performance issues before optimizing.
4. Follow ../../references/verification-policy.md: run the automatic minimum,

then offer a scoped desktop/browser QA pass. Treat mobile as a separate approval.

Asset Viewer Scaffold

For a greenfield asset-delivery viewer, create the canonical vanilla Three.js project:

python3 /scripts/create_threejs_asset_viewer.py ./my-asset-viewer

Use --force only when overwriting the target is intended. Configure src/asset-manifest.ts from the synchronized mint-assets.json; do not replace the scaffold's model, animation, material, pack, or RAD session with a bespoke scene unless the user requested a broader application.

Completion Gates

• Build/typecheck or the nearest compile gate passes.
• Focused non-browser tests for changed logic pass when available.
• Referenced project-local assets and changed imports resolve.
• mint-assets.json records every integrated Mint artifact or remote world

runtime under a stable logical key.

• Report that extended browser QA was not run unless the user approved it.

After the user approves the relevant QA scope:

• The primary user journey works through real input.
• Camera and controls match the task and do not fight each other.
• Loading, empty, unsupported, and error states are handled when applicable.
• Selection, configuration, manipulation, playback, or navigation state has one

owner and visible feedback.

• Assets load from stable local paths or the deliberate remote world runtime

with correct scale, orientation, materials, animation, bounds, and disposal.

• Desktop behavior is verified; mobile behavior is verified only under its

separate approval.

• Build/typecheck, browser errors, canvas pixels, screenshots, and changed risky

paths pass.

• Visual quality and renderer cost meet the requested bar.

Final Response

Lead with the implemented user journey. Report controls, state ownership, changed files, Mint links and artifact paths, screenshots, verification, performance evidence, deployment assumptions, and remaining risks. Explicitly name the extended desktop and mobile QA that was not run.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: mintdotgg
• Source: mintdotgg/mint-threejs-skills
• License: MIT
• Homepage: https://mcp.mint.gg

Install and usage instructions live in the source repository linked above.