20 years Weta/Pixar experience in real-time graphics, Metal shaders, and visual effects. Expert in MSL shaders, PBR rendering, tile-based deferred rendering (TBDR), and GPU debugging. Activate on 'Metal shader', 'MSL', 'compute shader', 'vertex shader', 'fragment shader', 'PBR', 'ray tracing', 'tile shader', 'GPU profiling', 'Apple GPU'. NOT for WebGL/GLSL (different architecture), general OpenGL (deprecated on Apple), CUDA (NVIDIA only), or CPU-side rendering optimization.
npx skills add https://github.com/curiositech/some_claude_skills --skill metal-shader-expert
20+ years Weta/Pixar experience specializing in Metal shaders, real-time rendering, and creative visual effects. Expert in Apple's Tile-Based Deferred Rendering (TBDR) architecture.
Use for:
Do NOT use for:
| Topic | Novice | Expert |
|-------|--------|--------|
| Data types | Uses float everywhere | Defaults to half (16-bit), float only when precision needed |
| Specialization | Runtime branching | Function constants for compile-time specialization |
| Memory | Everything in device space | Knows constant/device/threadgroup tradeoffs |
| Architecture | Treats like desktop GPU | Understands TBDR: tile memory is free, bandwidth is expensive |
| Ray tracing | Uses intersection queries | Uses intersector API (hardware-aligned) |
| Debugging | Print debugging | GPU capture, shader profiler, occupancy analysis |
| What it looks like | Why it's wrong |
|--------------------|----------------|
| float4 color, float3 normal everywhere | Wastes registers, reduces occupancy, doubles bandwidth |
| Instead: Default to half, upgrade to float only for positions/depth |
| What it looks like | Why it's wrong |
|--------------------|----------------|
| Treating Apple GPU like immediate-mode renderer | Tile memory reads are free; bandwidth is not |
| Instead: Use [[color(n)]] freely, prefer memoryless targets, avoid unnecessary store |
| What it looks like | Why it's wrong |
|--------------------|----------------|
| if (material.useNormalMap) checked every fragment | Creates divergent warps, wastes ALU |
| Instead: Function constants + pipeline specialization |
| What it looks like | Why it's wrong |
|--------------------|----------------|
| Using query-based API | Doesn't align with hardware; less efficient grouping |
| Instead: Use intersector API with explicit result handling |
| Era | Key Development |
|-----|-----------------|
| Pre-2020 | Metal 2.x, OpenGL migration, basic compute |
| 2020-2022 | Apple Silicon, unified memory, tile shaders critical |
| 2023-2024 | Metal 3, mesh shaders, ray tracing HW acceleration |
| 2025+ | Neural Engine + GPU cooperation, Vision Pro foveated rendering |
Apple Family 9 Note: Threadgroup memory less advantageous vs direct device access.
Play: The best shaders come from experimentation and happy accidents. Try weird ideas, build beautiful effects.
Exposition: If you can't explain it clearly, you don't understand it yet. Comment generously, show the math visually.
Tools: A good debug tool saves 100 hours of guessing. Build visualization for every complex shader.
| Area | Skills |
|------|--------|
| MSL | Kernel functions, vertex/fragment, tile shaders, ray tracing |
| Production | Asset pipelines, artist-friendly parameters, fast iteration |
| Rendering | PBR, IBL, volumetrics, post-processing, mesh shaders |
| Debug | Heat maps, shader inspection, GPU profiling, custom overlays |
| MCP | Purpose |
|-----|---------|
| Firecrawl | Research SIGGRAPH papers, Apple GPU architecture |
| WebFetch | Fetch Apple Metal documentation |
| File | Contents |
|------|----------|
| references/pbr-shaders.md | Cook-Torrance BRDF, material structs, lighting calculations |
| references/noise-effects.md | Hash functions, FBM, Voronoi, domain warping, animated effects |
| references/debug-tools.md | Heat maps, debug modes, overdraw viz, NaN detection, wireframe |
*Craft beautiful, performant Metal shaders with the artistry of film production and the pragmatism of real-time constraints.*
Integration with protocols.io API for managing scientific protocols. This skill should be used when working with protocols.io to search, create, update, or publish protocols; manage protocol steps and materials; handle discussions and comments; organize workspaces; upload and manage files; or integrate protocols.io functionality into workflows. Applicable for protocol discovery, collaborative protocol development, experiment tracking, lab protocol management, and scientific documentation.
Analyzes job descriptions and generates tailored resumes that highlight relevant experience, skills, and achievements to maximize interview chances
Generate Excalidraw diagrams from natural language descriptions. Use when asked to "create a diagram", "make a flowchart", "visualize a process", "draw a system architecture", "create a mind map", or "generate an Excalidraw file". Supports flowcharts, relationship diagrams, mind maps, and system architecture diagrams. Outputs .excalidraw JSON files that can be opened directly in Excalidraw.
Build and distribute Expo development clients locally or via TestFlight
Use when you have a written implementation plan to execute in a separate session with review checkpoints
Data structure for annotated matrices in single-cell analysis. Use when working with .h5ad files or integrating with the scverse ecosystem. This is the data format skill—for analysis workflows use scanpy; for probabilistic models use scvi-tools; for population-scale queries use cellxgene-census.
Benchling R&D platform integration. Access registry (DNA, proteins), inventory, ELN entries, workflows via API, build Benchling Apps, query Data Warehouse, for lab data management automation.
Comprehensive molecular biology toolkit. Use for sequence manipulation, file parsing (FASTA/GenBank/PDB), phylogenetics, and programmatic NCBI/PubMed access (Bio.Entrez). Best for batch processing, custom bioinformatics pipelines, BLAST automation. For quick lookups use gget; for multi-service integration use bioservices.
Take curiositech/metal-shader-expert from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.