Turns an arXiv paper into runnable Python code with citations linking every implementation choice back to the source. You paste an arXiv URL, it fetches the PDF, extracts structure, identifies the core contribution, then generates minimal working code in PyTorch, JAX, or NumPy. The key thing here is the ambiguity audit: it flags every detail the paper leaves unspecified and documents all decisions in REPRODUCTION_NOTES.md instead of quietly inventing hyperparameters. Comes with a walkthrough notebook that maps paper sections to code blocks. Three modes: minimal (just the contribution), full (training and eval), or educational (extra comments and theory). Honest about what it can't infer.
npx -y skills add prathamlearnstocode/paper2code --skill paper2code --agent claude-codeInstalls into .claude/skills of the current project.
Select a file.
juliusbrussee/caveman
mattpocock/skills
shadcn/improve
obra/superpowers
forrestchang/andrej-karpathy-skills
vercel-labs/skills