What problem does it solve? Research projects drift out of sync with their papers: it becomes unclear which code implements which equation, where the implementation diverges from the paper, and how the architecture fits together. This Skill regenerates an on-demand orientation map connecting paper claims to code locations. ## Core Features & Use Cases - Concept-to-Code Mapping: Maps paper equations, algorithms, and concepts to exact code locations (e.g., "Eq. 7 ELBO → losses.py:elbo"). - Architecture and Dataflow Diagrams: Produces small mermaid diagrams of module dependencies and data pipelines. - Divergence Detection: Lists evidence-backed paper-code divergences with file:line and paper-section citations. - Three Output Sinks: Emits to chat for orientation, stamps a README Overview block as the shareable snapshot, or writes a full self-contained notes/main.html with MathJax. - Use Case: Onboarding onto an unfamiliar ML research repo, ask for a project map to see which files implement the paper's training algorithm and where the code silently differs from the published method. ## Quick Start Map this research project by linking each paper equation and algorithm to its code location and listing any divergences.