Write interactive WebGL/canvas visualizations for Jupyter notebooks (.ipynb) opened inside Obsidian by the jupyter-uv plugin. A diagram is a small ES module that imports a shared three.js and is rendered with the `lab.viz` helper as a srcdoc iframe. Use when adding or porting an interactive viz into a notebook, writing notebook-local `notebooks/<notebook-name>/<diagram-name>.js` modules, or calling `lab.viz.show` / `lab.viz.scene`.
Apply the suckless philosophy — simplicity, clarity, frugality — when designing, writing, reviewing, or refactoring software in any language. Use when the goal is minimal, bloat-free tools where removed code is progress.
ALWAYS load before writing, editing, explicitly reviewing, or refactoring code in any language. Do not load for read-only file inspection, opening, navigation, or neutral summarization unless the user asks for review, design, safety, performance, or developer-experience feedback. Applies TigerBeetle TigerStyle principles for simplicity, safety, performance, and developer experience.
ALWAYS load before writing, editing, renaming, designing, or explicitly reviewing Python (.py) code. Do not load for read-only file inspection, opening, navigation, or neutral summarization unless the user asks for style, API, design, or code-review feedback. Covers general Python style, API shape, naming/import conventions, sum types, casting, assertions, formatting, and code review checks. Use python-modern separately for language feature/changelog guidance.
ALWAYS load before writing, editing, or creating any Python (.py) code — do not write Python without loading it first. Also use when explicitly reviewing Python language-version choices or modernization opportunities. Do not load for read-only file inspection, opening, navigation, or neutral summarization. Modern syntax preferences for 3.8-3.14 (walrus, pattern matching, exception groups, TaskGroup, generic syntax, type aliases, t-strings).
Ultra-compressed communication mode. Cuts token usage ~75% by speaking like smart caveman while keeping full technical accuracy. Use when user says "caveman mode", "talk like caveman", "less tokens", "be brief", or when token efficiency is requested.
ALWAYS load before writing, editing, or creating any tensor-reshaping or modeling code — any reshape/transpose/permute/view/squeeze/expand_dims/stack/concat/pooling/channel-or-head-shuffle on JAX, PyTorch, NumPy, or TF arrays. Use einops rearrange/reduce/repeat (+ Rearrange/Reduce layers) for readable, shape-safe array ops instead of opaque .view/.permute/.transpose/einsum chains.
ALWAYS load before writing, editing, or creating any JAX, PyTorch, NumPy, or other array/tensor code — do not write tensor code without loading it first. Typing conventions using jaxtyping + beartype for runtime type checking of array shapes. Despite the name "jaxtyping", it works with any array library.