paper-replication
Deep Learning Paper Replication Skill
读取深度学习论文(PDF 或文本),将论文中的数学描述、网络结构和实现细节转化为可运行、可验证、可维护的 PyTorch 代码与配套文档。
Core Objective
- 先理解论文的核心问题、创新点、数学公式和超参数。
- 再用 Mermaid 可视化关键数据流和张量形状。
- 最后输出工业级 PyTorch 实现,并在需要时生成模块级技术文档。
Workflow
根据用户需求和论文复杂度,选择合适的执行模式:
- 完整模式(默认):执行全部 4 个 Phase,适用于复杂模型的完整复现。
- 快速模式:用户明确只需要代码时,跳过 Phase 4(文档生成)。
- 代码优先模式:用户已理解论文时,简化 Phase 1 的解构,重点执行 Phase 3。
对于简单模型(如 MLP、单层网络),自动简化:
More from bahayonghang/my-claude-code-settings
tech-design-doc
Generate technical design documents with proper structure, diagrams, and implementation details. Use when designing a new feature, documenting architecture decisions, or planning refactoring work. Default language is English.
106bidwriter
智能标书编写专家,专精工程咨询、建筑设计、市政工程领域的投标文件编写。
70article-cover
Generate article cover images as SVG. Use when creating cover/banner/header graphics for blogs, documentation, social media cards, or OG images. Trigger this skill whenever the user mentions article covers, blog banners, post thumbnails, social sharing images, or any visual header for written content — even if they don't explicitly say 'SVG'. Do not use it for full-scene illustrations, photo editing, or slide theme work.
63tech-blog
Write technical blog posts with source code analysis OR doc-driven research. Use when user wants to explain system internals, architecture, implementation details, or technical concepts with citations.
53git-commit-cn
Safely orchestrate Chinese Conventional Commits for staged Git changes, or for all working-tree changes when the user explicitly asks to include everything. Use when the user asks to write a commit message, split staged changes, split all changes, commit everything regardless of stage state, include untracked files in the commit set, organize a messy index before committing, prepare a Chinese commit, draft a Conventional Commit, or generate structured commit text without pushing by default.
47gemini-image
>-
46