Cuda Programming

Use when the user asks about a CUDA kernel, nvcc, shared memory tiling, coalesced access, warp divergence, occupancy, CUDA streams or __syncthreads. Produces CUDA C++ kernels with launch config, error checking and an Nsight Compute profiling plan. Not for Triton kernels (use triton-kernels) or PTX/SASS reading (use ptx-assembly).

45 stars 6 forksSoftware Development

Install

skills CLI (Claude Code, Cursor, Codex and more)
npx skills add j4flmao/agent-skills --skill cuda-programming
Manual install
git clone --depth 1 https://github.com/j4flmao/agent-skills.git /tmp/agent-skills
mkdir -p ~/.claude/skills
cp -r /tmp/agent-skills/skills/hardware-acceleration/cuda-programming ~/.claude/skills/cuda-programming

Personal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.

From the SKILL.md

# CUDA Programming

## Purpose
Write correct, fast CUDA C++ kernels for NVIDIA GPUs. Most kernels are limited by memory bandwidth or latency, not arithmetic, so the work is mostly about access patterns, data reuse in shared memory/registers, and enough parallelism to hide latency. Every optimization must be confirmed with Nsight Compute, not intuition.

## Agent Protocol

### Trigger
"cuda", "cuda kernel", "nvcc", "gpu programming", "thread block", "shared memory", "warp divergence", "coalesced", "occupancy", "cuda streams", "__syncthreads".

### Input Context
- Target GPU / compute capability (e.g. sm_80 A100, sm_90 H100, sm_100 B200).
- CUDA toolkit version, host language, problem sizes and data types.
- Current kernel + profiler output if optimizing.

### Output Artifact
Kernel code (`.cu`) with launch code, error checking, and a profiling plan. No file unless requested.

### Response Format
```
Bound: {memory|compute|latency} (evidence: {ncu metric})
Launch: grid={...} block={...} smem={bytes}
Change: {what and why}
Expected effect: {metric to re-check}
```

### Completion Criteria
- [ ] Every CUDA API call and kernel launch error-checked.
- [ ] Bounds checks for partial tiles / tails.
- [ ] Result validated against a CPU or library reference.
- [ ] Bottleneck classified with Nsight Compute before and after the change.

## Execution Model
| Level | Facts |
|-------|-------|
| Thread | Has private registers; spills go to "local" memory (device DRAM, cached in L1/L2). |
Read the full skill on GitHub

Details

Skill name
cuda-programming
Path
skills/hardware-acceleration/cuda-programming/SKILL.md
License
MIT
Repo updated
17 November 2026

FAQ

What does the Cuda Programming skill do?

Use when the user asks about a CUDA kernel, nvcc, shared memory tiling, coalesced access, warp divergence, occupancy, CUDA streams or __syncthreads. Produces CUDA C++ kernels with launch config, error checking and an Nsight Compute profiling plan. Not for Triton kernels (use triton-kernels) or PTX/SASS reading (use ptx-assembly).

How do I install the Cuda Programming skill?

skills CLI (Claude Code, Cursor, Codex and more): npx skills add j4flmao/agent-skills --skill cuda-programming. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.

Is Cuda Programming popular?

It lives in j4flmao/agent-skills, which has 45 GitHub stars and 6 forks.

Related skills

Agent Observability

45

j4flmao/agent-skills

Use when the user asks about agent observability, tracing agent reasoning, debugging an agent loop, OpenTelemetry GenAI spans for LLM calls, LLM cost tracking per agent, agent audit logs or hallucination/eval monitoring in production. Produces instrumentation config, trace schema, dashboards, alerts and cost attribution. Not for generic service tracing without LLMs (use devops-opentelemetry).

DevOps & Cloud

AI Agent Architectures

45

j4flmao/agent-skills

Use when the user asks to build an AI agent, design an agent that calls tools and keeps memory, write an agent system prompt or pick an agent framework (Claude Agent SDK, OpenAI Agents SDK, LangGraph). Produces an architecture brief covering loop, tools, memory, guardrails and evals, routing depth to child skills. Not for workflow patterns (use ai-agentic-workflows).

AI, LLMs & Agents

AI Agent Core Loop

45

j4flmao/agent-skills

Use when the user asks about the agent loop, ReAct vs plan-and-solve, re-planning, Reflexion, an agent stuck in a loop, max iterations or termination conditions. Produces a loop spec or patch with state, success predicate, caps and no-progress detection. Not for overall agent design (use ai-agent-architectures) or tool schemas (use tool-calling-principles).

AI, LLMs & Agents

Data Data API

45

j4flmao/agent-skills

Use when the user asks about an instant API over a database: Hasura GraphQL, PostgREST, WunderGraph, a data gateway, real-time subscriptions or row-level authorization for data APIs. Produces API layer config with JWT/RBAC permissions, caching and rate limits. Not for hand-written REST/GraphQL services (use backend-api-design) or federation (use api-graphql-federation).

Software Development

Cocos2d Patterns

45

j4flmao/agent-skills

Use when the user asks about Cocos2d-x internals: the Director loop, scene graph, render queue and auto-batching, event dispatcher, or Ref/retain/release and autorelease pools. Explains engine internals to write fast C++ game code. Not for Cocos Creator gameplay scripting (use cocos2d).

Software Development

Data Search Engine

45

j4flmao/agent-skills

Use when the user asks about Elasticsearch, OpenSearch or Solr: mappings and analyzers, BM25 relevance tuning, kNN hybrid search, aggregations, shard sizing, ILM or cluster operations. Produces index mappings, queries and cluster settings. Not for app-level search features (use backend-search-patterns) or vector DB choice (use data-vector-databases).

Software Development

Browse the most popular Claude skills, all software development skills, or write your own with the skills generator.