REA
Reverse engineer anything from your terminal or agent with one CLI and MCP server.
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.
如果你是作者,可把这段代码贴到 README。徽章会链接回本页。
[](https://mcp.popnx.com/en/skills/viettranx-3dviz-pro-max)在同一工作中把 AI 连到服务的 MCP
Reverse engineer anything from your terminal or agent with one CLI and MCP server.
Codebase intelligence for TypeScript and JavaScript, exposed as structured MCP analysis tools.
MCP server for terminal commands, file operations, and process management
同类工作,排序与 Skill 列表相同
| 本项 3dviz-pro-max | rea | andrej-karpathy-skills | graphify | MoneyPrinterTurbo | |
|---|---|---|---|---|---|
| 作者 | viettranx | morluto | multica-ai | Graphify-Labs | harry0703 |
| Skill 数 | 1 | 1 | 1 | 1 | 1 |
| GitHub 星标 | 1,423 | 468,733 | 234,332 | 150,287 | 144,566 |
| 最近代码更新 | 2026年9月11日 | 2026年11月19日 | 2026年4月20日 | 2026年11月18日 | 2026年11月18日 |
| 许可证 | MIT | MIT | – | Apache-2.0 | MIT |
viettranx 提供的 1 个 Skill(指令集),AI 读取后照做 · 不是 MCP server
MCP = 手
把 AI 连到外部服务或数据,例如读取 Notion 或开发票。需要有程序或服务器在运行。
Skill = 手册
教 AI 怎么做某项工作的文本文件,不连接任何服务。
有些会重叠:少数 Skill 带有调用工具的脚本,少数 MCP 只提供知识。请从你想完成的工作出发。
Claude CodeClaude (เว็บ/แอป)Codex
依据开发者 README 和 repo 中的文件夹
GitHub account viettranx (person), 13 years old
★ 1,423 stars · code last changed 2026年9月11日
67 scripts that can run on a computer: scripts/blender_locate.py, scripts/capture.py, scripts/capture_view_quality.py, scripts/catalog.py and 63 more
Could not check every script (some are too large); read them first
No hooks or writes to machine settings found
Checked automatically from the repo files at commit d077e0e (read only, nothing was run). Not a safety guarantee: read it and decide for yourself
第一步:你用哪个 AI?
选择你用的应用,本页只显示该应用的安装方法
This skill is over 5 MB or 300 files, so we do not build a ZIP. Use "Let the AI install it" instead
复制这句话,粘贴到 Claude Code 的对话中
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
重新打开 Claude Code,输入 /3dviz-pro-max,或直接提出相关任务,Claude 会自动选用
Checked against official docs 2026年11月12日 · Source: code.claude.com
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
/plugin marketplace add viettranx/3dviz-pro-max
/plugin install 3dviz-pro-max@3dviz-pro-max摘自开发者 README,会下载并运行开发者的代码
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
git clone --depth 1 https://github.com/viettranx/3dviz-pro-max.git /tmp/popnx-skill-viettranx-3dviz-pro-max-d077e0e && mkdir -p ~/.claude/skills && cp -R '/tmp/popnx-skill-viettranx-3dviz-pro-max-d077e0e/skills/3dviz-pro-max' ~/.claude/skills/3dviz-pro-maxChatGPT 桌面应用内置 Codex,Codex 从 ~/.agents/skills 文件夹读取 Skill
在 ChatGPT 桌面应用中打开 Codex,把这句话粘贴到对话中
The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
把 3dviz-pro-max 文件夹放到 ~/.agents/skills/3dviz-pro-max
~ 是你的主文件夹($HOME)
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
git clone --depth 1 https://github.com/viettranx/3dviz-pro-max.git /tmp/popnx-skill-viettranx-3dviz-pro-max-d077e0e && mkdir -p ~/.agents/skills && cp -R '/tmp/popnx-skill-viettranx-3dviz-pro-max-d077e0e/skills/3dviz-pro-max' ~/.agents/skills/3dviz-pro-maxChecked against official docs 2026年11月15日 · Source: learn.chatgpt.com
不能使用你自己安装的 Skill,只能使用目录中插件自带的 Skill
想在 ChatGPT 中使用这个 Skill,请按上面的方法使用 ChatGPT 桌面应用
Checked against official docs 2026年11月15日 · Source: learn.chatgpt.com
复制这句话,粘贴到 Codex 的对话中 · 适用于 ChatGPT 桌面应用中的 Codex、Codex CLI 和 IDE 扩展
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
Codex 会自动发现新安装的 Skill;如果没有出现,请重启 Codex · 使用方法:在 ChatGPT 中输入 @ 并选择 Skill · 在 Codex CLI 或 IDE 扩展中运行 /skills,或输入 $ 加 Skill 名称
Checked against official docs 2026年11月15日 · Source: learn.chatgpt.com
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
git clone --depth 1 https://github.com/viettranx/3dviz-pro-max.git /tmp/popnx-skill-viettranx-3dviz-pro-max-d077e0e && mkdir -p ~/.agents/skills && cp -R '/tmp/popnx-skill-viettranx-3dviz-pro-max-d077e0e/skills/3dviz-pro-max' ~/.agents/skills/3dviz-pro-max复制这句话,粘贴到 Cursor 的对话中
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
在 Cursor 的 Agent 聊天中输入 / 并选择 3dviz-pro-max
Checked against official docs 2026年11月12日 · Source: cursor.com
Skill 是 AI 会照做的指令。安装前先阅读文件,只安装可信开发者的 Skill。
Mac:按 Command+空格输入 Terminal · Windows:下方命令在 PowerShell 中无法使用,请改用上方的“让 AI 帮你安装”
这条命令把 repo 下载到临时文件夹,只复制含 SKILL.md 的文件夹
git clone --depth 1 https://github.com/viettranx/3dviz-pro-max.git /tmp/popnx-skill-viettranx-3dviz-pro-max-d077e0e && mkdir -p ~/.cursor/skills && cp -R '/tmp/popnx-skill-viettranx-3dviz-pro-max-d077e0e/skills/3dviz-pro-max' ~/.cursor/skills/3dviz-pro-max适用于其他能读取 Skill 的应用。这句话列出了 Claude Code、Codex 和 Cursor 的路径,AI 会按你的应用选择 · Copy this sentence into the app’s chat
For one project, paste it into a chat that is open in that project’s folder · The AI always shows you the path before installing
The AI tells you what it will do first and asks your permission at every step
Checked against official docs 2026年11月12日 · Source: code.claude.com, learn.chatgpt.com, cursor.com
其他支持 Agent Skills 的应用(如 Gemini CLI、GitHub Copilot)使用不同文件夹,请查看该应用的指南,再把含 SKILL.md 的文件夹复制过去。
已于 2026年11月11日 对照 Claude Code、Codex 和 Cursor 官方指南核实 skill 文件夹
✓ 命令已检查 2026年11月19日
以下文字来自开发者的 repo,不是 PopNx MCP 的指令
显示前 8,000 个字符 · 在 GitHub 打开完整文件
---
name: 3dviz-pro-max
description: Design and build expressive 3D scenes, explainers and interactive models with grounded subject knowledge. Use for 3D creation or edits, including scientific visualization and spatial puzzles.
---
# 3Dviz Pro Max
Use the agent's spatial reasoning, geometry, composition and programming abilities to realize the
user's idea. The catalog, kits, rigs and scripts accelerate decisions and construction; they do not
set the boundaries of what you can make. A runnable scene still needs visual judgment.
## Workflow
For a new scene, work through the decisions below. For an existing artifact, inspect it first and
apply only the steps affected by the change. User constraints override suggested tooling and records.
1. **Intent and visual direction.** Identify what the viewer should see, feel, understand or do.
Choose a distinctive visual idea, useful references and the intended viewing distance. Decide how
shape, material, light, camera and motion work together. Time of day matters for some environments;
an anatomical study or mathematical object may need a studio, diagrammatic or invented setting.
2. **Reason about the important objects.** Before selecting a template, identify the silhouette,
proportions, parts, connections, surface identity and behavior that make each important object
convincing. Use [object reasoning](references/object-reasoning.md) and
[object craft](references/object-craft.md). Solve the hardest or most distinctive object early.
3. **Ground the representation.** Choose illustration, discrete state, simulation or playback; a
scene can combine them if their boundaries are clear. For established academic, scientific or
technical claims, research the actual passages and fact-check claim → model → rendered result
using [research and truth](references/research-and-truth.md). A decorative mesh, imported anatomy
or physics engine does not certify a scientific claim. Identify the authoritative state.
4. **Choose construction per object.** Use the routing table below. Search for the missing decision
when useful; do not force a subject into the closest lexical match. Recipes, numerical defaults
and look profiles are candidates to adapt, not instructions to copy every value.
At important joins, decide whether the object needs a continuous skin, a layered covering or
separate articulated parts. Use [surface continuity](references/object-craft.md#choose-continuity-at-the-join)
when attached volumes look accidental; mesh merging is not a universal cleanup prescription.
```sh
python3 scripts/search.py "<subject and unresolved decision>"
python3 scripts/resolve.py <record-id>
python3 scripts/design-context.py "<objective>" --blueprints --look <style-id>
```
5. **Set quality targets and discover tools.** State what the object must withstand at overview,
inspection and close-up distances, including factual fidelity where relevant. Discover available
tools without treating tool availability as artistic quality:
```sh
python3 scripts/host-probe.py --out host.json
python3 scripts/design-context.py "<objective>" --blueprints --host host.json \
--delivery web-desktop --hero <blueprint-id>
```
T0–T4 describe the existing kit pipelines and their declared proof status. The legacy
`quality_ceiling` field is a shipped-kit pipeline recommendation, not a limit on custom geometry,
external assets, procedural materials or the agent's creativity. A missing kit tier does not
prohibit another construction method. Blender availability does not make an object high quality;
Three.js-only construction does not make it low quality.
6. **Build a convincing first view.** Keep the user's stack. Reuse shared scripts, renderer setup,
[rigs and scaffold](templates/README.md), material utilities and geometry operations where they
help. For a large scene or collection, inspect one representative demanding object or scene before
multiplying the approach. A kit-shaped prototype should not silently become the finished hero.
Author custom geometry, shaders, materials, textures, lighting and interactions as needed. Use
Blender for suitable modeling, sculpting, rigging, UV or baking work; use Three.js for suitable
procedural geometry, interactive state and rendering. Choose their handoff deliberately.
For a distinctive finish, develop [palette and material direction](references/object-craft.md#develop-a-material-direction)
from useful references: translate them into part assignments, surface scale and light response.
Material refinement can strengthen a good form; keep evaluating shape and attachments separately.
When adding or changing directional, practical, day/night or rescaled lighting, read
[lighting direction and scale](references/lighting-direction-and-scale.md) before implementing the
rig. Identify emitter/receiver ownership and final world scale; choose the artistic treatment freely.
7. **Make behavior belong to the model.** Keep transforms, joints, contacts and controls tied to
their authoritative state. Inspect surrounding objects too: a stand needs support, a light ray
must stop at an opaque obstacle in its modeled scene, a creature needs clearance. An illustration
may simplify physics, but should not accidentally imply a simulation it does not perform.
Allocate reusable geometry once where practical; update transforms, uniforms or buffers during
motion. Park static scenes and respect reduced-motion preferences. Do not remove identity-defining
detail merely to meet an unmeasured budget; optimize what actually costs work.
Use [subject clarity and meaningful motion](references/subject-clarity-and-motion.md) when
identity, mechanism or control effects are unclear. Explain what changes and stays fixed; keep
landmarks, color meanings and readouts synchronized with the visible state. Check local deformation
and attachments through relevant extremes, not just the resting silhouette.
8. **Run and inspect the real output.** Build through the project's existing commands, then view
the artifact in an available authorized browser. Inspect overview, meaningful close-ups and the
important interactions. If automated capture is allowed, the optional shared command is:
```sh
pnpm build
python3 scripts/capture.py --dir dist --all-views --click "#control" --out captures
```
Add `--motion-check 2000 --expect-motion` only when the tested state should visibly move.
`--expect-usable` can flag gross visibility failures; it does not judge art. A visible-browser
inspection is valid when headless execution is excluded. Never describe an unobserved frame.
For lighting changes, inspect source isolation at fixed exposure, an informative direction change,
and any output/reset controls. Check affected receivers and occluders, not just glowing bulbs.
Preserve the user's motion and camera choices across lighting modes unless a control promises reset.
9. **Refine by subject and evidence.** Use [visual review](templates/checklists/visual-anti-slop.md)
and [interaction/frame inspection](templates/checklists/inspection-per-frame.md) as conditional
prompts. Correct factual errors, broken controls and misleading relationships, then improve the
most consequential visual weakness. A failed shape may require a different construction method,
not more texture or a higher kit tier. For pinching, detached attachments or repeated pose fixes,
use [regional deformation diagnosis](references/object-craft.md#diagnose-deformation-at-the-region-that-fails):
solver accuracy, surface deformation and visual quality need separate evidence. Revisit ownership
and rig assumptions after changing geometry. Continue useful refinement; reassess an approach that keeps
failing rather than applying a fixed number of cosmetic passes.
10. **Report honestly.** Separate what was built, observe提问、分享技巧或报告无法使用。按“有帮助”数量排序。