huiyuli-2000/gen-pseudocode-skill

gen-pseudocode-skill is a Claude custom skill (a set of instructions and reference files you install into your Claude-powered project) designed to help AI and machine learning researchers convert messy code or paper descriptions into clean, publication-ready algorithm pseudocode for academic papers. The problem it addresses is the gap between how algorithms look in actual source code, full of framework-specific API calls, data loading boilerplate, and implementation details, and how they should look in a journal or conference paper, where they are expressed in clean mathematical notation. For example, code like "optimizer.zero_grad(), loss.backward()" becomes concise LaTeX math notation describing the gradient update step. The skill guides Claude to take inputs such as a paper's methodology section, supplementary appendices, or source code, and produce compilable LaTeX using the algorithm2e package (a standard LaTeX package for typesetting algorithm boxes). It includes a notation mapping table that converts programming terms to their mathematical equivalents, and a style guide covering conventions for different publication venues like NeurIPS, ICML, AAAI, and medical informatics journals. The repository also includes a Python script that compiles the generated LaTeX into a PDF so you can immediately verify the output is correctly formatted. You install the skill by placing the provided folder in your project's Claude skills directory. The project is licensed under MIT.