Manim Codex Plugin

This repo contains a local Codex plugin that helps create mathematical animations and still-frame visualizations with Manim Community Edition.

What is included

• plugins/manim/.codex-plugin/plugin.json
  • The plugin manifest and UI metadata.
• plugins/manim/skills/manim-visualizer/SKILL.md
  • The skill Codex should use for Manim-based math visualization work.
• plugins/manim/skills/transformer-visualizer/SKILL.md
  • A transformer-focused skill for step-by-step LLM inference visuals.
• plugins/manim/scripts/init_manim_project.sh
  • Checks native prerequisites and creates a standalone uv-managed Manim project.
• plugins/manim/scripts/render_manim.sh
  • Renders a specific scene from a local Manim project.
• plugins/manim/scripts/render_manim_docker.sh
  • Renders a scene inside a pinned official Manim Docker image.
• plugins/manim/docker/Dockerfile
  • Pins the repo's container path to the official manimcommunity/manim:v0.20.1 image.
• .devcontainer/devcontainer.json
  • Opens this repo inside the pinned Manim container.
• .agents/plugins/marketplace.json
  • A repo-local marketplace entry for the plugin.

Design choices

• The plugin is skill-first instead of MCP-heavy.
• Setup is explicit: missing Cairo, Pango, or LaTeX dependencies are surfaced directly.
• Rendering stays workspace-local so generated scenes and media live with the user's project, not inside the plugin.
• The isolated path is container-first: use Docker when you want LaTeX support and reproducible renders without modifying the host.

Transformer inference examples

The repo now includes reusable transformer visualization primitives plus one scene per major inference stage under manim/scenes/transformer_inference/.

Example steps:

• EmbeddingAndPositionStep
• QKVProjectionStep
• AttentionScoringStep
• AttentionAggregationStep
• ResidualMLPStep
• LogitSamplingStep

Render one step at a time:

export PATH="$HOME/Library/TinyTeX/bin/universal-darwin:$PATH"
./plugins/manim/scripts/render_manim.sh \
  --project-dir ./manim \
  --scene-file ./manim/scenes/transformer_inference/attention_scoring_step.py \
  --scene-name AttentionScoringStep \
  --quality low

Docker usage

Render a workspace-local project with the pinned official Manim container:

./plugins/manim/scripts/render_manim_docker.sh \
  --project-dir /path/to/workspace/manim \
  --scene-file /path/to/workspace/manim/scenes/example.py \
  --scene-name ExampleScene \
  --quality low

Open this repo in a devcontainer:

devcontainer open .

Notes

• The setup flow in this plugin was verified against the current stable Manim docs on April 6, 2026.
• At that time, the latest GitHub release was v0.20.1 dated February 27, 2026.
• The Docker guidance in this repo was verified against the current stable Docker docs on April 6, 2026.