Use this skill aggressively any time the work involves Rust code that touches GPU compute — including ferrotorch (and ferrotorch- crates), burn, candle, wgpu-rs, cubecl, cudarc, or any project involving CUDA, ROCm, Metal, Vulkan, WGPU, cuBLAS, PTX, WGSL, or kernel dispatch. Trigger on any mention of .cuda(), Device::Cuda(), .todevice(), kernel launches, tensor ops, autograd backward passes, JIT codegen, or "GPU acceleration" in a Rust context. Also trigger when the user names a Rust GPU/ML project (ferrotorch, ferray, burn, candle) or asks to implement, port, optimize, or debug compute kernels. This skill exists because the model otherwise systematically defaults to CPU implementations, stubs out GPU code paths, and reports "done" while the work secretly executes on CPU. Use this skill to enforce pre-flight planning, the forbidden-pattern checklist, mechanical verification, and honest reporting so that GPU work actually lands on the GPU.
Install with
Claude CodeCodexCursorGitHub CopilotGemini CLI
Supported by the Atlas install prompt. Skill-specific requirements vary.
Skill stats
Skills1CategoryEngineeringStars0
About this skill
Use this skill aggressively any time the work involves Rust code that touches GPU compute — including ferrotorch (and ferrotorch- crates), burn, candle, wgpu-rs, cubecl, cudarc, or any project involving CUDA, ROCm, Metal, Vulkan, WGPU, cuBLAS, PTX, WGSL, or kernel dispatch. Trigger on any mention of .cuda(), Device::Cuda(), .todevice(), kernel launches, tensor ops, autograd backward passes, JIT codegen, or "GPU acceleration" in a Rust context. Also trigger when the user names a Rust GPU/ML project (ferrotorch, ferray, burn, candle) or asks to implement, port, optimize, or debug compute kernels. This skill exists because the model otherwise systematically defaults to CPU implementations, stubs out GPU code paths, and reports "done" while the work secretly executes on CPU. Use this skill to enforce pre-flight planning, the forbidden-pattern checklist, mechanical verification, and honest reporting so that GPU work actually lands on the GPU.
Quick start
Copy the install prompt. Use the green button above this README.
Send it to your agent. Paste it into your coding agent’s chat and send.
Review the setup. Your agent checks the skill’s instructions, requirements, and installation scope.
Use this skill aggressively any time the work involves Rust code that touches GPU compute — including ferrotorch (and ferrotorch-* crates), burn, candle, wgpu-rs, cubecl, cudarc, or any project involving CUDA, ROCm, Metal, Vulkan, WGPU, cuBLAS, PTX, WGSL, or kernel dispatch. Trigger on any mention of `.cuda()`, `Device::Cuda(_)`, `.to_device()`, kernel launches, tensor ops, autograd backward passes, JIT codegen, or "GPU acceleration" in a Rust context. Also trigger when the user names a Rust GPU/ML project (ferrotorch, ferray, burn, candle) or asks to implement, port, optimize, or debug compute kernels. This skill exists because the model otherwise systematically defaults to CPU implementations, stubs out GPU code paths, and reports "done" while the work secretly executes on CPU. Use this skill to enforce pre-flight planning, the forbidden-pattern checklist, mechanical verification, and honest reporting so that GPU work actually lands on the GPU.
About this listing
Category
Engineering
Reported license
Apache-2.0
Listing language (estimated)
English
Review status
Not individually reviewed
Agent compatibility
Atlas installation supports Claude Code, Codex, and Cursor. Skill-specific dependencies and features are checked during setup.
Claude Code
Codex
Cursor
GitHub Copilot
Gemini CLI