Claude Code skill pack for AI infrastructure: GPU kernels, Megatron memory estimation, SLIME RL training, and technical diagram/slide generation. Install via git clone + copy into skills path; heavily LLM-generated, verify-before-use.
DSH integration
Ecosystem-related
Author-claimed
Safety audit
Unaudited
Last verified
2026-08-24
License
Not specified by the author
01What can it help you accomplish?
Write and optimize high-performance GPU kernels for AI infrastructure work inside Claude Code
Complete TileLang / CUDA kernel code, performance optimizations (swizzle, pipelining), tests and hardware-specific tuning recommendations
AI infrastructure engineers who use Claude Code and need expert help writing GPU kernels, distributed training and inference optimization
Produce technical diagrams and presentation decks from Claude Code
LaTeX TikZ flowcharts/architecture diagrams, Material You HTML slide decks, OpenAI-dotcom-style figures and HF architecture TikZ diagrams
AI infrastructure engineers and researchers who need professional flowcharts, architecture diagrams and slides without leaving the coding agent
02How to install into DeepSeek Harness
Prerequisites
- Claude Code (or a compatible agent that reads Claude's skills path) installed
- git available to clone the repository
Installation steps
- 01
Clone the repo: `git clone https://github.com/yzlnew/infra-skills.git`
$ git clone https://github.com/yzlnew/infra-skills.git
- 02
Personal install: `mkdir -p ~/.claude/skills` then `cp -r infra-skills/<skill> ~/.claude/skills/` for each skill (tilelang-developer, megatron-memory-estimator, slime-user, tikz-flowchart, material-you-slides, anthropic-theme-flowchart, hf-architecture-tikz)
- 03
Project install: clone into `.claude/skills-repo` inside your project and copy the skill folders into `.claude/skills/`
Verify the integration
Not specified by the author
03DSH integration and capability boundaries
Claude Code agent-skill pack listed under the dsh-plugin ecosystem; skills install into Claude's skills path and auto-activate. Not a DeepSeek Harness native runtime.
TileLang Developer
kernel task description (e.g. "FP16 GEMM optimized for A100")→complete TileLang kernel code, swizzle/pipelining optimizations, tests, hardware tuning recommendations
Megatron Memory Estimator
HuggingFace model config + parallelism strategy→GPU memory breakdown, parallelism comparison (TP/PP/EP/CP), optimization recommendations
SLIME User
RL post-training request (GRPO/GSPO/PPO, multi-turn tool calling)→environment setup, custom reward/generation functions, Megatron/FSDP backend config, multi-node scaling guidance
TikZ / Material You / Anthropic / OpenAI / HF architecture diagrams
diagram or slide description→LaTeX TikZ flowcharts, Material You HTML decks, Anthropic-theme HTML flowcharts, OpenAI-dotcom figures, HF architecture TikZ diagrams
04Who is it for? When not to use it?
Good for
- AI infrastructure engineers who use Claude Code and need expert help writing GPU kernels, distributed training and inference optimization
- AI infrastructure engineers and researchers who need professional flowcharts, architecture diagrams and slides without leaving the coding agent
Not for
- The maintainers warn the project is under active development and heavily LLM-generated without strict proofreading; verify all generated code before production use.
- No SPDX license is declared in the README; skills are provided as-is and generated code follows the license of the underlying frameworks, not this repo.
- Skills target Claude Code's skills path (~/.claude/skills or .claude/skills); they are not a DeepSeek Harness native runtime and rely on Claude's auto-activation.
05Compatibility, maintenance and safety notes
- The maintainers warn the project is under active development and heavily LLM-generated without strict proofreading; verify all generated code before production use.
- No SPDX license is declared in the README; skills are provided as-is and generated code follows the license of the underlying frameworks, not this repo.
- Skills target Claude Code's skills path (~/.claude/skills or .claude/skills); they are not a DeepSeek Harness native runtime and rely on Claude's auto-activation.
No SPDX license declared; no tagged release; project self-describes as under active LLM-generated development
06Frequently asked questions
Is infra-skills a DeepSeek Harness (dsh) plugin?
Not natively. It is a Claude Code agent-skill pack listed under the dsh-plugin ecosystem. Skills copy into Claude's skills path and auto-activate; there is no dsh CLI runtime or native integration described in the README.
How do I install a skill from this repo?
Clone `https://github.com/yzlnew/infra-skills.git`, then `cp -r infra-skills/<skill> ~/.claude/skills/` for a personal install, or copy into your project's `.claude/skills/` for collaborators. Skills activate automatically when a matching task is detected.
Which skills are complete versus planned?
Complete: TileLang Developer, Megatron Memory Estimator, SLIME User, TikZ Flowchart, Material You Slides, Anthropic Theme Flowchart, HF Architecture TikZ, OpenAI Dotcom Viz. Planned only: SGLang Developer and vLLM Developer.
Is it safe to use the generated code in production?
The README explicitly warns the project is under active development and heavily generated by LLMs without strict proofreading. Verify all generated code and examples before production use.
What license does this repository use?
No SPDX license is declared in the README. Skills are provided as-is for development, and generated code follows the license terms of the underlying frameworks rather than this repository.
07Related DSH workflows
archify
by tt-a1i
Agent skill for beautiful, verifiable architecture, workflow, sequence, data-flow, and lifecycle diagrams—self-contained HTML with motion and crisp export.
openviking
by volcengine
Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.
nocobase
by nocobase
NocoBase is an open-source AI + no-code platform for building business systems fast. Instead of generating everything from scratch, AI works on top of production-proven infrastructure and a WYSIWYG no-code interface, so you get both speed and reliability.
learn-harness-engineering
by walkinglabs
Harness engineering beginner tutorial, from 0 to 1
08Data and sources
A collection of specialized agent skills for AI infrastructure engineers, covering both technical development (GPU kerne…
Skills automatically activate when relevant tasks are detected.
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-08-24. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.