kernel-workflow

Automate AI kernel generation, verification, and optimization across Triton, CUDA C, C++, and TileLang.

6|1|Updated Apr 19, 2026
One-click install
npx skills add https://github.com/xchang1121/AutoResearch-CC-hook --skill kernel-workflow
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: kernel-workflow
Source: https://github.com/xchang1121/AutoResearch-CC-hook/tree/main/skills/kernel-workflow
Command: npx skills add https://github.com/xchang1121/AutoResearch-CC-hook --skill kernel-workflow

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

Kernel workflow provides end-to-end orchestration for generating, validating, and optimizing AI kernels across multiple backends (Triton, CUDA C/C++, TileLang), coordinating task interpretation with subagents and verification steps.

Core Features & Use Cases

  • End-to-end kernel workflow: analyze user input (descriptions or code), generate/optimize kernels, and verify results using provided references and scripts.
  • Backend-agnostic support: supports Triton, CUDA C, C++, and TileLang backends with pluggable verification and tooling.
  • Adaptive orchestration: optional user_requirements customize optimization strategies; integrates with references and scripts for validation.

Quick Start

Provide a kernel task description or seed code to start the end-to-end generation and optimization workflow.

Frequently Asked Questions about kernel-workflow

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I generate and optimize AI kernels across multiple backends?

AI kernel generation and optimization is automated across Triton, CUDA C, C++, and TileLang by providing a task description or seed code. The workflow orchestrates subagent calls to interpret inputs, generate kernels, and validate results based on optional user requirements.

Can I use this workflow to optimize existing CUDA C or Triton kernel code?

Yes, existing CUDA C, Triton, or TileLang kernel code can be optimized by providing it as input. The workflow analyzes the code and applies adaptive orchestration to improve performance while validating the results using integrated scripts and references.

What is the best way to verify AI kernel performance across different backends?

Verifying AI kernel performance is handled by the workflow's integrated validation scripts and references. It automates the verification process across supported backends like Triton and CUDA C, ensuring generated or optimized kernels meet the specified requirements.

Does the kernel generation workflow require specific dependencies or environments?

No external dependencies are required to run the workflow. It operates standalone using internal scripts and references, allowing users to start generation and optimization directly by providing a task description or kernel code.

How do I customize optimization strategies for specific AI kernel tasks?

Optimization strategies are customized by providing optional user requirements alongside the task description or kernel code. The workflow applies adaptive orchestration to satisfy these specific constraints during generation and validation.

What limitations exist when generating kernels for mixed C++ and TileLang backends?

The workflow supports backend-agnostic generation and validation for C++ and TileLang, but relies on provided references and scripts for safe orchestration. Complex edge cases may require manual review of the generated kernel code.