convez.ai
SkillEscrita & ConteúdoAvançado
shihabshahrier

Manim Coding Skill para Animações STEM

shihabshahrier/manim-coding-skill

Gera animações educacionais em ManimGL via agentes IA com pipeline pedagógico estruturado

Ver no GitHub
BomPluginScore · de 0 a 100

Skill que transforma agentes de IA em professores de animações STEM de alta fidelidade usando ManimGL. Implementa um pipeline de 5 fases rigoroso que gera, renderiza e mescla vídeos educacionais a partir de tópicos, com suporte para Claude Code, opencode e IDEs populares.

Instalação

$npx convez-ai shihabshahrier/manim-coding-skill

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
5
Forks
2
Acompanhando
0
Saúde
71%
Último commit
há 2 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Criar vídeos explicativos de conceitos matemáticos e científicos
  • Gerar conteúdo educacional automatizado para cursos online
  • Integrar geração de animações em pipelines de CI/CD de documentação técnica

Veredito

Ferramenta sólida para automação de conteúdo educacional visual, com arquitetura bem pensada (narrative-first, chunking de 15s, ManimGL nativo). Exige conhecimento técnico de ManimGL e integração com agentes, mas entrega resultados pedagógicos consistentes quando bem configurada.

README

A tradução para o português deste README ainda está na fila.

🎬 test-manim

License: MIT Claude Code Plugin opencode Skill ManimGL Version

An AI-powered agent skill that turns any Large Language Model into a high-fidelity STEM animation teacher. Give it a topic, and it generates, renders, and merges stunning, pedagogically-rich educational videos using ManimGL.

This skill is designed for developer agents, terminal assistants, and IDE rules across all major platforms, enforcing a rigorous, structured pipeline for creating visual explanations.


🗺️ Table of Contents


✨ Core Philosophy

Standard LLMs often struggle with creating animations because they dump long equations, use random colors, build overly complex layouts, or write broken syntax.

test-manim solves this by equipping AI agents with a strict set of rules, templates, and references:

  • Narrative-first: Outlines a deep pedagogical strategy before writing any code.
  • 15-second chunking: Divides long videos into digestible, single-concept pieces.
  • ManimGL native compatibility: Enforces exact API usage (ShowCreation, TexText, Tex) instead of mixed ManimCE elements.
  • Audio-ready layouts: Intentionally embeds pacing margins and holds to allow seamless voice-over integration later.

📦 Installation & Integration

1. Claude Code (Plugin)

Install the plugin directly within your Claude Code environment:

claude plugin install shihabshahrier/manim-coding-skill

Once installed, you can invoke the generator in any chat session using the /test-manim command.

2. opencode (Skill)

For users running opencode or similar tools:

npx skills add shihabshahrier/manim-coding-skill

This instantly enables the /test-manim command inside your active session using any model.

3. IDE Integration (Rules)

This repository includes pre-configured, condensed rules tailored to popular AI coding editors. They automatically sync with the source skill:

EditorFile LocationActivation
Cursor.cursor/rules/test-manim.mdcTriggered by .mdc context rules
Windsurf.windsurf/rules/test-manim.mdHandled by Windsurf global rules
Cline.clinerules/test-manim.mdEnabled for all Cline projects
Copilot.github/copilot-instructions.mdRead during GitHub Copilot conversations

4. Manual Integration

You can paste the core instruction file into any custom LLM system prompt or custom agent instructions: 👉 skills/test-manim/SKILL.md


🏗️ The 5-Phase Pipeline

To guarantee visual consistency, syntax correctness, and rendering reliability, the skill enforces a rigid execution pipeline:

graph TD
    A["Phase 0: Scaffold & Parse"] --> B["Phase 1: SCENE_PLAN.md (Story Planning)"]
    B --> C["Phase 2: Generate ManimGL Chunk Files"]
    C --> D["Phase 3: Render Chunks (manimgl)"]
    D --> E["Phase 4: Merge Chunks (ffmpeg)"]
    E --> F["Phase 5: Structured Report & Rating"]
  1. Phase 0 — Parse & Scaffold: Parses args, determines segment count, and sets up a deterministic project directory.
  2. Phase 1 — Narrative Planning: Generates a comprehensive script and storyboard (SCENE_PLAN.md) outlining the educational hook, build order, and synthesis act.
  3. Phase 2 — Scene Generation: Writes isolated, self-contained Python scene files matching the visual plans using native ManimGL APIs.
  4. Phase 3 — Render: Automatically runs a generated parallel or ordered shell script (render.sh) to render chunks in HD.
  5. Phase 4 — Merge: Seamlessly strings all chunks together via FFmpeg (merge.sh) to produce a single fluid video.
  6. Phase 5 — Report: Compiles a detailed feedback summary including timing logs, design checks, and quality ratings.

🎨 Pedagogical & Storytelling Principles

The 3-Act Structure

Every animation must avoid the "academic lecture" trap and instead follow a structured story:

  1. Act 1: The Question (Hook): Open with a striking physical phenomenon or paradox before writing any text or math. Instigate curiosity.
  2. Act 2: The Build (Chunks): Step-by-step introduction of concepts. One chunk = one single concept. Move from physical intuition towards formal abstractions.
  3. Act 3: The Synthesis (Resolution): Loop back to the Act 1 visual, now annotated with the newly taught context. Hold the final insight frame static for 3–5 seconds.

Color Semantics

Colors are used to convey mathematical and physical meaning, not as decoration. A strict "color contract" is defined in Phase 1 and adhered to across all chunks:

  • 🟦 BLUE: Primary concept / object of focus (e.g., mass, current, graph).
  • 🟨 YELLOW: Secondary concept / dynamic vectors (e.g., force, electric field, derivatives).
  • 🟥/🟩 RED / GREEN: Highlight, accentuation, or contrast (e.g., acceleration, positive/negative states).
  • ⬜ WHITE: Titles, axes, standard text, and LaTeX equations.

Pacing Guidelines

Interaction / ElementVisual TimingAction
New Object Arrival1.5s - 2.0sSlow entry to let the eye adjust
Object in Motion0.5s - 1.0sSnappy, indicating immediate relationship
LaTeX Equation Arrives1.5s - 2.5sSlow drawing, letting symbols sink in
Key Insight Pause3.0s minimumHold frames to allow mental processing
Scene Transition0.5sClean, quick fade-outs

Common Anti-Patterns

  • ❌ Definition Dumps: Showing equations or text titles before showing the physical object or graph.
  • ❌ Color Chaos: Changing the role of colors between scenes (e.g. Blue represents velocity in chunk 1, but stands for gravity in chunk 2).
  • ❌ Too Many Elements: Animating 4 unrelated things at once. We show, explain, then build.
  • ❌ Wall of Text: Writing paragraphs on screen. Main text is capped at 8 words max per element.

💻 Usage & Options

Run the skill command within your terminal or chat session:

/test-manim "topic" [--duration 30|60|90|120] [--model name]

Arguments

ParameterDefaultValues / Description
topicRequiredAny STEM (Science, Tech, Engineering, Math) subject string.
--duration3030, 60, 90, 120 (in seconds). Total length of final video.
--modelDerivedModel identifier used in folder names (e.g. claude, gpt4).

Example Output Structure

For /test-manim "Fourier Series" --duration 30:

manim-claude/
  fourier_series/
    scenes/
      chunk_01_approx.py    # Chunk 1: The Hook
      chunk_02_harmonics.py # Chunk 2: The Build
    output/
      chunks/               # Isolated chunk .mp4 renders
      final/
        fourier_series_final.mp4  # Merged high-quality video
    assets/
    SCENE_PLAN.md           # The storyboard and script
    render.sh               # Executable to render all chunks
    merge.sh                # Executable to join chunks using ffmpeg

System Requirements

  • Python 3.8+
  • ManimGL v1.7.2+ (pip install manimgl)
  • FFmpeg (accessible in PATH for merging video outputs)
  • LaTeX installation (e.g. TeX Live, MiKTeX for formula rendering)

🛠️ Developer & Contribution Guide

To suggest changes to the prompt strategies, API definitions, or storytelling templates, please review CONTRIBUTING.md:

  1. Main Prompt / Instruction File: Edit skills/test-manim/SKILL.md
  2. ManimGL API Cheat-Sheet: Edit skills/test-manim/references/manim-gl-api.md
  3. Pacing & Narrative Guides: Edit skills/test-manim/references/storytelling.md

[!WARNING] Do not edit the editor-specific copies (.cursor/, .windsurf/, .clinerules/) directly. They are auto-generated and synchronized from the main source of truth files.

Let's build stunning educational visualizations together! 🚀

Ver o original no GitHub

Bom para

Relacionados