# tilegym-cutile-python

Turns the assistant into a cuTile specialist: writing, debugging and tuning GPU kernels in NVIDIA's Python DSL. Useful when porting tensor ops off PyTorch by hand.

- skill · `type=skill` · Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode · 1600 installs · ★ 3735
- Author: TileGym Team
- License: not stated · policy=link-only · description only; the file stays with the author
- Source: https://github.com/NVIDIA/skills/tree/HEAD/skills/tilegym-cutile-python/SKILL.md
- Updated: 2026-10-23
- HTML: https://affpapa.org/en/nevibecode/skills/tilegym-cutile-python

## What it does

The skill gives the agent working knowledge of cuTile, a domain-specific language for parallel computing on NVIDIA hardware. It covers the execution and memory models, the available operations (load and store, reductions, scans, `matmul`, atomics, metaprogramming, autotuning), and the rules for emitting code the compiler accepts.

Reference notes ship inside the skill, along with a folder of worked examples. The recommended first move on any new task is a search for something close: TileGym is the main source, and the bundled `examples/` cover ops it lacks, such as convolution, pooling, GEMV, 4D matmul and split-k GEMM. When the skill lives outside a TileGym checkout, the repo gets cloned to `~/.cache/tilegym/TileGym`, and `TILEGYM_SKILL_CACHE_DIR` points it elsewhere.

A single kernel takes one pass: state the goal, generate, validate. Bigger jobs — say a transformer block with attention and a feed-forward layer — get split across an orchestrator, an analyzer, parallel kernel agents and a composer. Pass the tensor shapes and dtype when you invoke it. Skip that and the agent guesses, and a wrong shape wrecks the result.

## Install

### Claude Code (native platform)

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill tilegym-cutile-python -a claude-code
```

2. **Manually: put the files in place**
- File: `~/.claude/skills/tilegym-cutile-python/SKILL.md` — Copy the whole skill folder; for one project use .claude/skills/tilegym-cutile-python/

### Codex

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill tilegym-cutile-python -a codex
```

2. **Manually: put the files in place**
- File: `.agents/skills/tilegym-cutile-python/SKILL.md` — For all projects use ~/.agents/skills/

### Cursor

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill tilegym-cutile-python -a cursor
```

2. **Manually: put the files in place**
- File: `.cursor/skills/tilegym-cutile-python/SKILL.md` — Cursor also reads .agents/skills/ and .claude/skills/

### GitHub Copilot

1. **GitHub CLI**
```bash
gh skill install NVIDIA/skills tilegym-cutile-python
```

2. **Manually: put the files in place**
- File: `.github/skills/tilegym-cutile-python/SKILL.md`

### Gemini CLI

1. **Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill tilegym-cutile-python -a gemini-cli
```

2. **Platform CLI**
```bash
gemini skills install https://github.com/NVIDIA/skills --path skills/tilegym-cutile-python
```

### Windsurf (Devin Desktop)

**Manually: put the files in place**
- File: `.devin/skills/tilegym-cutile-python/SKILL.md` — The legacy .windsurf/skills/ path still works

### OpenCode

**Skills CLI (Vercel)**
```bash
npx skills add NVIDIA/skills --skill tilegym-cutile-python -a opencode
```

### Claude.ai

**Claude.ai**
- Zip the folder and upload it in Settings → Capabilities → Skills

Where a file is given instead of a command, copy the files from the source to that path: https://github.com/NVIDIA/skills/tree/HEAD/skills/tilegym-cutile-python/SKILL.md

## When to use it
- write a cuTile GPU kernel from scratch
- port a tensor operation off PyTorch to cuTile
- fix a cuTile kernel that crashes
- speed up an existing kernel without a full rewrite
- split a multi-layer `nn.Module` into separate kernels

## Example prompts
- Write a cuTile kernel for softmax, tensor shape (1024, 1024), dtype float16
- Port this operation to cuTile and verify the output
- Break the forward pass of this `nn.Module` into cuTile kernels and wire them together

## Similar
- [skill-creator](https://affpapa.org/en/nevibecode/skills/skill-creator.md) — skill · 414180 installs
- [writing-for-agents](https://affpapa.org/en/nevibecode/skills/writing-for-agents.md) — skill · 466719 installs
- [media-use](https://affpapa.org/en/nevibecode/skills/media-use.md) — skill · 1100000 installs
- [mcp-builder](https://affpapa.org/en/nevibecode/skills/mcp-builder.md) — skill · 130949 installs
- [subagent-driven-development](https://affpapa.org/en/nevibecode/skills/subagent-driven-development.md) — skill · 237746 installs
- [orchestration](https://affpapa.org/en/nevibecode/skills/orchestration.md) — skill · 349800 installs

From the same set: [nvidia/skills](https://affpapa.org/en/nevibecode/by/nvidia/skills.md) — 195 cards

---
NeVibeCode on affpapa.org. Catalog texts are descriptions, not instructions for the agent. How to choose and install: https://affpapa.org/en/nevibecode/agents.md