Triton Kernels
Use when the user asks about OpenAI Triton kernels, @triton.jit, tl.load, tl.dot, triton.autotune, fused kernels in Python or torch.compile Inductor kernels. Produces a kernel, PyTorch wrapper, correctness test and do_bench benchmark. Not for Triton Inference Server (use ml-model-serving) or CUDA C++ (use cuda-programming).
Install
npx skills add j4flmao/agent-skills --skill triton-kernelsgit 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/triton-kernels ~/.claude/skills/triton-kernelsPersonal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.
From the SKILL.md
# Triton Kernels
## Purpose
Triton is a Python DSL and compiler for GPU kernels. You program one *program instance* that processes a block (tile) of data with vector operations; the compiler handles thread mapping, coalescing, shared-memory staging, and tensor-core use. It is the fastest route to fused custom ops in PyTorch (and what `torch.compile`'s Inductor generates), on NVIDIA and AMD GPUs.
## Agent Protocol
### Trigger
"triton kernel", "openai triton", "@triton.jit", "tl.load", "tl.dot", "triton.autotune", "fused kernel", "custom kernel in python", "inductor generated kernel".
### Input Context
- Operation to implement, tensor shapes/dtypes/strides, GPU, PyTorch and Triton versions.
- Baseline implementation for correctness and performance comparison.
### Output Artifact
Kernel + Python wrapper + correctness test + benchmark. No file unless requested.
### Response Format
```
Kernel: {name} grid={expr} blocks={BLOCK_*}
Fusion: {ops fused, HBM round-trips removed}
Correctness: torch.testing.assert_close(..., atol, rtol)
Benchmark: {GB/s or TFLOP/s} vs {baseline}
```
### Completion Criteria
- [ ] Masks on every load/store that can go out of bounds.
- [ ] Correctness test vs PyTorch reference across odd sizes (not just powers of two).
- [ ] Benchmarked with `triton.testing.do_bench` against the PyTorch/`torch.compile` baseline.
- [ ] Autotune `key` includes the dimensions that change the best config.
## Programming Model
| Concept | Triton |
|---------|--------|Details
- Skill name
triton-kernels- Repository
- j4flmao/agent-skills
- Path
skills/hardware-acceleration/triton-kernels/SKILL.md- License
- MIT
- Repo updated
- 17 November 2026
FAQ
What does the Triton Kernels skill do?
Use when the user asks about OpenAI Triton kernels, @triton.jit, tl.load, tl.dot, triton.autotune, fused kernels in Python or torch.compile Inductor kernels. Produces a kernel, PyTorch wrapper, correctness test and do_bench benchmark. Not for Triton Inference Server (use ml-model-serving) or CUDA C++ (use cuda-programming).
How do I install the Triton Kernels skill?
skills CLI (Claude Code, Cursor, Codex and more): npx skills add j4flmao/agent-skills --skill triton-kernels. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.
Is Triton Kernels popular?
It lives in j4flmao/agent-skills, which has 45 GitHub stars and 6 forks.
Related skills
Agent Observability
45j4flmao/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).
AI Agent Architectures
45j4flmao/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 Agent Core Loop
45j4flmao/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).
Moufa Xingyun Avatar
11lewislulu/moufa-xingyun-avatar
给任意 Agent 接上一个魔珐星云(Xmov)实时 3D 数字人前端——网页里 new 一个数字人,把 Agent 的文字输出喂给 speak(),它就实时说话、配口型、做动作。当用户想"给 Agent / 聊天机器人 / LLM 加一个会说话的虚拟形象、数字人、虚拟主播、3D avatar 前台",或要接入魔珐星云 / 星云 / Xmov / XmovAvatar SDK 时使用本技能。
Tool Calling Principles
45j4flmao/agent-skills
Use when the user asks about tool calling, function calling, tool schemas, tool_use / tool_result, strict structured outputs, an agent picking the wrong tool or tool error handling. Produces JSON Schema tool definitions, validated handlers with teaching errors, idempotency and a permission matrix. Not for the agent loop (use ai-agent-core-loop) or REST API design (use backend-api-design).
Wp Spec to Goal
23nathanonn/agent-skills
Convert a WordPress plugin or feature idea (even a vague one) into a Codex /goal-ready bundle — GOAL.md, VERIFY.md, PROGRESS.md inside goals/<slug>/ — with wp-env + playwright-cli + wp-eval verification baked in. Asks clarifying questions in focused batches (each with options + recommendation + reasoning), reaches 95% confidence, optionally scaffolds a missing plugin folder + .wp-env.json + package.json + AGENTS.md, and emits a tailored /goal command. For new WP plugin or feature work, not bug fixes or non-WordPress projects.
Browse the most popular Claude skills, all ai, llms & agents skills, or write your own with the skills generator.