Skip to content
Back to search

infra-skills

A collection of specialized agent skills for AI infrastructure development, enabling Claude Code to write, optimize, and debug high-performance systems.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/yzlnew--infra-skills/badge.svg)](/listings/yzlnew--infra-skills)

HTML

<a href="/listings/yzlnew--infra-skills"><img src="/api/listings/yzlnew--infra-skills/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Skills are installed by placing the skill directory in Claude's skills path:

**Natural Language:** Ask Claude Code directly: "Help me install skills from https://github.com/yzlnew/infra-skills"

**Personal (across all projects):**

# Clone and copy to personal skills directory
git clone https://github.com/yzlnew/infra-skills.git
mkdir -p ~/.claude/skills
cp -r infra-skills/tilelang-developer ~/.claude/skills/
cp -r infra-skills/megatron-memory-estimator ~/.claude/skills/
cp -r infra-skills/slime-user ~/.claude/skills/
cp -r infra-skills/tikz-flowchart ~/.claude/skills/
cp -r infra-skills/material-you-slides ~/.claude/skills/
cp -r infra-skills/anthropic-theme-flowchart ~/.claude/skills/
cp -r infra-skills/hf-architecture-tikz ~/.claude/skills/

**Project-level (for repository collaborators):**

# Clone and copy to project's skills directory
cd your-project
git clone https://github.com/yzlnew/infra-skills.git .claude/skills-repo
mkdir -p .claude/skills
cp -r .claude/skills-repo/tilelang-developer .claude/skills/
cp -r .claude/skills-repo/megatron-memory-estimator .claude/skills/
cp -r .claude/skills-repo/slime-user .claude/skills/
cp -r .claude/skills-repo/tikz-flowchart .claude/skills/
cp -r .claude/skills-repo/material-you-slides .claude/skills/
cp -r .claude/skills-repo/anthropic-theme-flowchart .claude/skills/
cp -r .claude/skills-repo/hf-architecture-tikz .claude/skills/

Skills automatically activate when relevant tasks are detected.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 356 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. # AI Infrastructure Agent Skills
  2. > **⚠️ WARNING**
  3. > This project is under active development and heavily generated by LLMs without strict proofreading. Use with caution and verify all code before production use.
  4. A collection of specialized agent skills for AI infrastructure engineers, covering both technical development (GPU kernels, distributed training, inference optimization) and soft skills (flowchart creation, presentation design).
  5. ## Overview
  6. This repository provides expert-level skills tailored for AI infrastructure engineering. Each skill packages domain knowledge, code examples, and best practices to transform Claude into a specialized assistant for specific frameworks and workflows—from writing high-performance CUDA kernels to creating professional technical presentations.
  7. ### Construction Methodology (Unless Otherwise Specified)
  8. 1. **Knowledge Gathering**: Use Gemini DeepResearch to collect comprehensive, up-to-date information on target frameworks
  9. 2. **Skill Development**: Transform research into structured skills using `skill-creator` in Claude Code
  10. 3. **Validation**: Test skill-generated code examples to ensure correctness
  11. 4. **Maintenance**: Regular updates based on latest official documentation
  12. ## Available Skills
  13. ### TileLang Developer
  14. Write high-performance GPU kernels using TileLang for NVIDIA, AMD, and Ascend hardware.
  15. **Capabilities:**
  16. - Matrix multiplication (GEMM) kernels
  17. - FlashAttention implementations
  18. - DeepSeek MLA operators
  19. - Performance optimization (swizzle layouts, pipelining, warp specialization)
  20. - Cross-platform kernel development
  21. **Status:** ✅ Complete
  22. ### Megatron Memory Estimator
  23. Estimate GPU memory usage for Megatron-based MoE and dense models. Built upon [megatron_memory_estimator](https://huggingface.co/spaces/ISEEKYAN/megatron_memory_estimator).
  24. **Capabilities:**
  25. - Estimate memory from HuggingFace configs
  26. - Support for MoE models (DeepSeek-V3, Qwen, etc.)
  27. - Parallelism strategy comparison (TP/PP/EP/CP)
  28. - Memory optimization recommendations
  29. **Status:** ✅ Complete
  30. ### SLIME User
  31. Guide for using SLIME (LLM post-training framework for RL Scaling). Built upon [THUDM/slime](https://github.com/THUDM/slime).
  32. **Capabilities:**
  33. - RL training setup and configuration (GRPO, GSPO, PPO, Reinforce++)
  34. - Multi-turn tool calling and agent workflows
  35. - Custom reward models and generation functions
  36. - Megatron and FSDP backend configuration
  37. - SGLang integration and optimization
  38. - Dynamic sampling and partial rollout
  39. - Multi-node distributed training
  40. **Status:** ✅ Complete
  41. Prompt to create this skill, with Sonnet 4.5:
  42. ```
  43. Use skill-creator to create a skill called slime-user at this repo. slime is an LLM
  44. post-training framework for RL Scaling. Its repo is https://github.com/THUDM/slime.
  45. Skill creation procedure:
  46. 1. Git clone the latest repo
  47. 2. Analyze `docs/en`, understand basic structure and write a doc navigation guide for user
  48. getting started or finding docs for advanced usage
  49. 3. Gather valuable examples from the docs and `examples` dir, write key ideas and script
  50. path down for quick reference
  51. 4. Checkout some important source code, for example `slime/slime/utils/arguments.py` and
  52. `slime/rollout/sglang_rollout.py`, provide its path and functions for a quick find.
  53. ```
  54. ### TikZ Flowchart
  55. Create professional flowcharts and architecture diagrams using LaTeX TikZ with standardized styles.
  56. **Capabilities:**
  57. - Professional flowcharts with Google Material-like color palette
  58. - Standardized node types (data, memory, operation, kernel boxes)
  59. - Architecture diagrams and process flows
  60. - Grouping and layout best practices
  61. - Clean orthogonal edges and relative positioning
  62. **Example Output:** [QAT Flowchart](gallery/qat_flowchart.pdf) | [Anthropic Theme](gallery/tikz_flowchart_anthropic_theme.png)
  63. ![TikZ Flowchart Anthropic Theme](gallery/tikz_flowchart_anthropic_theme.png)
  64. **Status:** ✅ Complete
  65. ### Material You Slides
  66. Create presentation slide decks using Material You (Material Design 3) design language.
  67. **Capabilities:**
  68. - Self-contained HTML slides (1280x720) with M3 color tokens
  69. - Roboto typography with multiple weight support
  70. - Professional slide types (title, section divider, content)
  71. - Component library (cards, flow diagrams, metric cards, code blocks)
  72. - Rounded shapes and generous whitespace
  73. - Surface hierarchy without drop shadows
  74. - Structured layouts (columns, tables, lists, tags/chips)
  75. **Example Output:** [SLIME RL Training Slides](gallery/slime_rl_slides.pdf)
  76. **Status:** ✅ Complete
  77. ### Anthropic Theme Flowchart
  78. Create polished standalone HTML/CSS flowcharts with Anthropic-inspired pastel styling, reliable geometry, and deterministic connector routing.
  79. **Capabilities:**
  80. - TS-first flowchart specs that generate standalone HTML artifacts
  81. - Deterministic node geometry and anchor-based connector routing
  82. - Transparent dashed grouping frames and pastel role-based node styling
  83. - Hollow `>` arrowheads and orthogonal bridge connectors
  84. - Gallery-ready exports generated from the same geometry source
  85. **Example Output:** [Review Demo PNG](gallery/anthropic_theme_flowchart_review_demo.png)
  86. ![Anthropic Theme Flowchart Review Demo](gallery/anthropic_theme_flowchart_review_demo.png)
  87. **Status:** ✅ Complete
  88. ### HF Architecture TikZ
  89. Generate Sebastian-Raschka-gallery-style TikZ architecture diagrams for any HuggingFace decoder-only LLM, with per-block parameter formulas and concrete numbers.
  90. **Capabilities:**
  91. - Extract architecture from HuggingFace configs into a structured spec
  92. - Render publication-quality vertical TikZ diagrams via Jinja templates
  93. - Annotate every sub-block with parameter formulas and concrete numbers
  94. - Support MHA, GQA, MLA, Hyper-Connections, sparse attention with learned indexer
  95. - Cover dense and MoE FFNs (incl. hash routing) and MTP heads
  96. - Models: DeepSeek-V4-Flash, Qwen, Llama, Mistral, gpt-oss, etc.
  97. **Example Output:** [DeepSeek-V4-Flash PNG](hf-architecture-tikz/examples/deepseek-v4-flash/deepseek-v4-flash.png) | [PDF](hf-architecture-tikz/examples/deepseek-v4-flash/deepseek-v4-flash.pdf)
  98. ![DeepSeek-V4-Flash Architecture](hf-architecture-tikz/examples/deepseek-v4-flash/deepseek-v4-flash.png)
  99. **Status:** ✅ Complete
  100. ### OpenAI Dotcom Viz
  101. Build figures in OpenAI's blog / research / system-card "dotcom" visual style, emitted as self-contained HTML/SVG with zero dependencies.
  102. **Capabilities:**
  103. - Monochrome bar charts (darker same-hue stroke, rounded corners, black y-axis with outward ticks, no gridlines, circle legend markers, value labels above bars, angled category labels)
  104. - Flow / process diagrams (rounded boxes, monospace uppercase pills, pink highlights, thin open-chevron connectors, dashed negative branches)
  105. - Real OpenAI Sans typography with zero external dependencies
  106. **Status:** ✅ Complete
  107. ## Planned Skills
  108. ### SGLang Developer
  109. Development skill for SGLang (Structured Generation Language) runtime and optimization.
  110. **Planned capabilities:**
  111. - SGLang runtime configuration
  112. - Custom sampling strategies
  113. - Performance tuning for LLM inference
  114. - Multi-GPU serving optimization
  115. **Status:** 🚧 Planned
  116. ### vLLM Developer
  117. Skill for vLLM engine development and deployment.
  118. **Planned capabilities:**
  119. - PagedAttention implementation
  120. - Custom scheduler development
  121. - Multi-LoRA serving
  122. - Quantization integration
  123. **Status:** 🚧 Planned
  124. ## Usage
  125. ### Installing Skills
  126. Skills are installed by placing the skill directory in Claude's skills path:
  127. **Natural Language:**
  128. Ask Claude Code directly: "Help me install skills from https://github.com/yzlnew/infra-skills"
  129. **Personal (across all projects):**
  130. ```bash
  131. # Clone and copy to personal skills directory
  132. git clone https://github.com/yzlnew/infra-skills.git
  133. mkdir -p ~/.claude/skills
  134. cp -r infra-skills/tilelang-developer ~/.claude/skills/
  135. cp -r infra-skills/megatron-memory-estimator ~/.claude/skills/
  136. cp -r infra-skills/slime-user ~/.claude/skills/
  137. cp -r infra-skills/tikz-flowchart ~/.claude/skills/
  138. cp -r infra-skills/material-you-slides ~/.claude/skills/
  139. cp -r infra-skills/anthropic-theme-flowchart ~/.claude/skills/
  140. cp -r infra-skills/hf-architecture-tikz ~/.claude/skills/
  141. ```
  142. **Project-level (for repository collaborators):**
  143. ```bash
  144. # Clone and copy to project's skills directory
  145. cd your-project
  146. git clone https://github.com/yzlnew/infra-skills.git .claude/skills-repo
  147. mkdir -p .claude/skills
  148. cp -r .claude/skills-repo/tilelang-developer .claude/skills/
  149. cp -r .claude/skills-repo/megatron-memory-estimator .claude/skills/
  150. cp -r .claude/skills-repo/slime-user .claude/skills/
  151. cp -r .claude/skills-repo/tikz-flowchart .claude/skills/
  152. cp -r .claude/skills-repo/material-you-slides .claude/skills/
  153. cp -r .claude/skills-repo/anthropic-theme-flowchart .claude/skills/
  154. cp -r .claude/skills-repo/hf-architecture-tikz .claude/skills/
  155. ```
  156. Skills automatically activate when relevant tasks are detected.
  157. ### Examples
  158. **TileLang Kernel Development:**
  159. ```bash
  160. # User request:
  161. "Write a FP16 matrix multiplication kernel optimized for A100"
  162. # Claude loads tilelang-developer skill and generates:
  163. # - Complete TileLang kernel code
  164. # - Performance optimizations (swizzle, pipelining)
  165. # - Testing code
  166. # - Hardware-specific tuning recommendations
  167. ```
  168. **Megatron Memory Estimation:**
  169. ```bash
  170. # User request:
  171. "Estimate memory for DeepSeek-V3 with TP=8, PP=4, EP=8"
  172. # Claude loads megatron-memory-estimator skill and provides:
  173. # - Detailed memory breakdown (model, optimizer, activations)
  174. # - Comparison across different parallelism strategies
  175. # - Memory optimization recommendations
  176. # - Hardware configuration suggestions
  177. ```
  178. **SLIME RL Training Setup:**
  179. ```bash
  180. # User request:
  181. "Help me set up GRPO training for Qwen3-4B with multi-turn tool calling"
  182. # Claude loads slime-user skill and provides:
  183. # - Environment setup instructions
  184. # - Custom generation function for tool calling
  185. # - Training script configuration
  186. # - Multi-node scaling guidance
  187. ```
  188. **TikZ Flowchart Creation:**
  189. ```bash
  190. # User request:
  191. "Create a flowchart showing the FlashAttention-2 algorithm flow"
  192. # Claude loads tikz-flowchart skill and generates:
  193. # - Professional LaTeX TikZ diagram with standardized colors
  194. # - Data nodes (green), operation nodes (blue), memory nodes (orange)
  195. # - Clean layout with orthogonal edges
  196. # - Grouped kernel phases with proper styling
  197. ```
  198. **Material You Slides Creation:**
  199. ```bash
  200. # User request:
  201. "Create a presentation deck about our AI infrastructure architecture"
  202. # Claude loads material-you-slides skill and generates:
  203. # - Self-contained HTML file with Material Design 3 styling
  204. # - Title slide with gradient background and branding
  205. # - Section dividers with large translucent numbers
  206. # - Content slides with cards, flow diagrams, and metric displays
  207. # - Responsive 1280x720 slides ready for presentation
  208. ```
  209. ## Development
  210. ### Testing Skills
  211. Validate code examples in skills:
  212. ```bash
  213. # Run all tests from project root
  214. pytest
  215. # Run tests for specific skill
  216. pytest tests/tilelang-developer/
  217. # Run specific test file
  218. pytest tests/tilelang-developer/test_gemm.py
  219. ```
  220. ### Updating Skills
  221. When frameworks release major updates:
  222. 1. Update skill source files (SKILL.md, references/) with latest information
  223. 2. Run validation tests to ensure examples are correct
  224. 3. Commit and tag new version
  225. ## Quality Standards
  226. All skills must meet these criteria:
  227. - ✅ **Accurate**: Code examples must be tested and correct
  228. - ✅ **Concise**: Follow progressive disclosure (SKILL.md < 500 lines)
  229. - ✅ **Complete**: Include workflow, API reference, examples, and debugging
  230. - ✅ **Current**: Based on latest stable framework version
  231. - ✅ **Clear**: Explicit triggers in description for automatic activation
  232. ## Contributing
  233. ### Skill Requests
  234. Open an issue with:
  235. - Framework/tool name
  236. - Use cases and scenarios
  237. - Link to official documentation
  238. ### Skill Improvements
  239. 1. Fork the repository
  240. 2. Update skill source files
  241. 3. Run validation tests
  242. 4. Submit PR with changelog
  243. ## Roadmap
  244. - [x] TileLang developer skill
  245. - [x] Megatron memory estimator skill
  246. - [x] SLIME user skill
  247. - [x] TikZ flowchart skill
  248. - [x] Material You slides skill
  249. - [x] Anthropic theme flowchart skill
  250. - [x] HF architecture TikZ skill
  251. - [ ] SGLang developer skill
  252. - [ ] vLLM developer skill
  253. - [ ] Automated testing pipeline
  254. - [ ] Documentation update monitoring
  255. - [ ] Skill versioning system
  256. ## Resources
  257. - [Claude Agent Skills Documentation](https://docs.anthropic.com/en/docs/build-with-claude/agent-skills)
  258. - [skill-creator Guide](https://github.com/anthropics/claude-code)
  259. - [TileLang Repository](https://github.com/tile-ai/tilelang)
  260. - [SLIME Repository](https://github.com/THUDM/slime)
  261. ## License
  262. Skills are provided as-is for development purposes. Generated code follows the license terms of the underlying frameworks.
  263. ---
  264. **Note**: This is a specialized repository for AI infrastructure developers. Skills contain advanced technical content and assume familiarity with GPU programming, compiler design, and deep learning systems.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.