3dviz-pro-max: an agent skill for Three.js and Blender scenes
Agent skill for creative 3D visualization: turn an idea into a Three.js/Blender scene worth exploring. Claude Code + Codex plugin, 223 recipes, 440 knowledge records, 22 proved kits, 37 runnable studies.
At a glance
- What is it?
- 3dviz-pro-max is a Claude Code and Codex skill that turns a one-sentence idea into a Three.js or Blender scene, backed by 223 recipes and 37 runnable studies. It is useful when you want the agent to inspect what it rendered, and awkward when you already know exactly which renderer you want.
- Who is it for?
- Adopt 3dviz-pro-max if you want an agent to pick a representation, build a Three.js or Blender scene, and capture frames so it can check its own output, and you are willing to let it choose the renderer. Skip it if you already have a fixed pipeline, a locked art direction, or a host that is not Claude Code or Codex, because the skill is distributed as a plugin and skill folder for those two hosts.
- Can I use it commercially?
- Yes. MIT is a permissive licence: you can use, modify and sell software built on it, as long as you keep its copyright and licence notices.
- Is it still maintained?
- Yes. The repository last received commits 7 days ago.
- What is it written in?
- Mainly JavaScript, according to GitHub's language statistics.
Answers come from the project's GitHub data, last synced on September 16, 2026, and from our analysis. They are not legal advice.
DEEP OPEN-SOURCE ANALYSIS
The problem 3dviz-pro-max targets: agents that describe scenes instead of rendering them
Ask a coding agent for a 3D scene and you usually get plausible prose about a scene. Geometry, camera, lighting and motion stay unverified, because the agent never looks at a frame. 3dviz-pro-max is built around that gap. The README states the goal plainly: "shape the world, model its rules, make it interactive, and check what actually appears on screen." The last clause is the differentiator. The skill ships a capture helper so the agent inspects the frames it rendered rather than narrating them.
The audience is narrow on purpose. It is for people working inside Claude Code or Codex who want an explorable scene, not a static diagram. The README's own opening prompt asks for "a small fantasy village I can explore" with moving creatures and an inspection step. The repository also carries directions for anatomy, operations and logistics, architecture and products, so the intended use is broader than fantasy: cutaways, assembly sequences, sorter models, patrol routes. Factual explanations in the project are expected to carry sources; the README is explicit that "animated decoration does not become a simulation just by moving." That sentence sets the tone of the whole project.
How the skill works: a ten-step workflow over a searchable catalog
The mechanism is a skill folder plus data. Under skills/3dviz-pro-max/ the repository holds SKILL.md, which describes a ten-step workflow, and a data directory containing catalog-summary.json. The counts in the README are read from that file: 223 recipes across 24 registered directions, 440 knowledge records in 18 populated families, 22 kit blueprints, 37 style profiles, 24 directions, 37 example studies and 21 captured frames.
The retrieval layer is the interesting part. Recipes and knowledge records are searchable, and 69 knowledge records carry authored numeric defaults, as do 33 of the 37 style profiles. So when the agent needs a camera angle, a gear ratio or a lighting setup, it can pull a recorded number instead of inventing one. Kit blueprints are described as a record plus a module plus four real captures, re-checked by SHA-256, which means a kit is only valid if its captures match.
Data flow runs in one direction: prompt, retrieval from the catalog, scene generation in Three.js or Blender, headless capture, then inspection of the resulting frames. The README says the agent "should use the existing project stack, choose an appropriate representation, and refine the scene from observed results." You are not asked to pick a renderer or write a specification first. That is convenient and also the main source of unpredictability, since the representation is chosen for you.
Installing 3dviz-pro-max into Claude Code or Codex
The README points installation at docs/installation.md, which has separate anchors for Claude Code and Codex: the badges link to docs/installation.md#claude-code and docs/installation.md#codex. The repository layout confirms two host directories at the top level, .claude-plugin/ and .codex-plugin/. The install commands themselves are not reproduced in the README, so read that file before running anything; do not guess a plugin marketplace name or a CLI flag.
The first real use is a prompt, not a command. The README gives this example verbatim:
Use 3dviz-pro-max to build a small fantasy village I can explore. Choose a distinctive art direction, add a few moving creatures, and inspect the result.What you should see is a scene you can navigate, plus captured frames the agent examined. The repository ships 37 runnable example studies across five chapters, listed in examples/README.md, with directories such as examples/cottage/, examples/gears/, examples/heart/ and examples/lantern/. Running one of those first is a cheaper way to confirm the capture path works than starting from a blank prompt. There is also examples/.env.example, which suggests the studies expect some environment configuration; the README does not say which keys it contains, so open the file rather than inventing variables.
Where 3dviz-pro-max breaks down
The strongest limitation is stated by the project itself, in the caption under its own hero image. The animated village GIF is labelled "Visual inspiration, not a benchmark," an author-supplied project "recorded before this skill existed," with a UI containing Vietnamese. It is historical footage, not a demo or a runtime test of the skill. Anyone who reads that GIF as evidence of what the skill outputs has misread the page. The project deserves credit for saying so, but it also means the headline visual is not a product demo.
The demo gallery is the honest counterweight: 21 frames captured headlessly in Chromium on a GPU at 1920x1080, with host, tiers and hashes recorded in docs/demos/README.md. The README also notes that the village frames come from a retired pilot revision at mixed tiers, and that no frame claims a tier its record calls missing. Read that page before treating any image as representative.
Two other constraints matter. First, host lock-in: this is distributed as a Claude Code plugin and a Codex skill folder, so a different agent host is out of scope. Second, the no-specification design is a trade-off. Letting the agent choose the representation helps when you have a vague idea, and hurts when you have a fixed pipeline, a brand palette, or an existing Three.js codebase you must not restructure. The README also lists no releases, so there is no versioned changelog to pin against; CHANGELOG.md exists at the top level, but the README does not document a rollback path.
Alternatives and the actual difference in approach
The obvious comparison is writing Three.js by hand with the three.js documentation open. That gives you total control over the scene graph, the render loop and the asset budget, and it costs you the retrieval layer and the capture loop. With 3dviz-pro-max you trade control for a catalog: 223 recipes, 440 knowledge records and 69 records with authored numeric defaults that the agent can draw on. If your scene is one canvas with a known camera and a known model, hand-written Three.js is faster and more predictable.
The second comparison is Blender scripting without an agent. Blender's own Python API is the tool for baked assets, and the README's demo gallery includes a "Blender-baked hero" cottage at tier T3, so the project does use Blender output. The difference is that 3dviz-pro-max treats Blender as one option among several and decides between it and a runtime Three.js scene based on the request. If you already know the asset must be baked, you are paying for a decision layer you do not need.
The third comparison is a general-purpose coding agent with no skill installed. It can write Three.js code, but it has no recipe catalog, no kit blueprints with SHA-256-checked captures, and no capture helper. The gap is verification, not code generation.
Maintenance, licence and what upgrading costs you
The repository is not archived, and the last push was on 2026-09-11. It is MIT licensed, which is permissive: you can use, modify and redistribute it, including commercially, provided the licence and copyright notice are kept. That is a statement about the licence text, not legal advice, and the repository also carries a SECURITY.md and a CONTRIBUTING.md at the top level if you plan to send changes back.
The upgrade cost is unusual because the project tries to make it low. The README states that the landing page recomputes every published count from skills/, docs/, evals/ and evidence/ at build time, "so a number cannot drift from the data behind it." In practice that means a catalog update changes the numbers automatically rather than leaving stale figures in the README. The catch is that the numbers are the interface: if a recipe is retired or a knowledge record loses its authored defaults, the agent's retrieval changes behaviour without any version bump you can see. No releases are listed, so pinning to a tag is not an option described here. Pin to a commit instead, and re-run one example study after any pull to confirm the capture path still works.
Editorial conclusion
Adopt 3dviz-pro-max if you want an agent to pick a representation, build a Three.js or Blender scene, and capture frames so it can check its own output, and you are willing to let it choose the renderer. Skip it if you already have a fixed pipeline, a locked art direction, or a host that is not Claude Code or Codex, because the skill is distributed as a plugin and skill folder for those two hosts. Before relying on it, open docs/installation.md, confirm which host you are installing into, and run one of the 37 example studies under examples/ to see whether the capture helper works on your machine.
Frequently asked questions
What is 3dviz-pro-max and who is it for?
It is an agent skill for creative 3D visualization that turns an idea into a Three.js or Blender scene worth exploring. It targets people working inside Claude Code or Codex who want an explorable scene rather than a static diagram.
How do I install 3dviz-pro-max?
The README points to docs/installation.md, which has separate sections for Claude Code and Codex, and the repository carries .claude-plugin/ and .codex-plugin/ directories at the top level. The install commands are not reproduced in the README, so read that file rather than guessing a flag.
Does 3dviz-pro-max work with renderers other than Three.js and Blender?
The README describes the skill as producing Three.js templates and kits and using Blender for baked assets such as the timber cottage at tier T3. No other renderer is mentioned.
Community notes