wtf is paper-review-and-digest?

This project provides two skills for Claude Code, Anthropic's coding assistant tool, that are meant to help someone properly read a single academic paper. One skill, paper-review, judges whether a paper can be trusted. The other, paper-digest, helps a reader absorb what the paper actually says. It was built by a practicing physician for real journal-club and clinical use, then shared more broadly. The paper-review skill performs a detailed appraisal: it checks the credibility of the journal and the track record of the authors, verifies references against the live CrossRef database to catch fake or mismatched citations, checks for bias using a method chosen based on the type of study, and computes a GRADE certainty rating, which is a standard way of scoring how trustworthy a body of medical evidence is. A key design choice is that the language model only makes judgment calls, while separate small Python scripts do the actual math and logic. One script recalculates the final GRADE score from the model's individual ratings rather than trusting the model's own summary, and another script checks whether the model's reasoning contains logical leaps, such as treating a single study as proven or treating an indirect measurement as if it proved a direct clinical benefit. The paper-digest skill instead focuses on helping someone learn the paper's content quickly, offering three layers of detail and ending with self-test questions. It refuses to produce a digest if it cannot access the paper's full text, so it will not generate a summary based only on the abstract. The README describes this as the reading step in a three-part pipeline: a separate tool called paper-radar discovers and ranks papers, another tool called paper-fetch downloads their full text, and this project appraises and digests them. The two Python scripts here have no external dependencies and run with the standard Python library only, with no installation or network access required for the scoring itself, though the CrossRef reference check does require internet access. To use it, someone places the paper-review and paper-digest folders into their Claude Code skills directory, copies an example configuration file, and fills in any optional settings, since the core functionality works without extra configuration. The author describes this as a one-time reference implementation rather than an actively maintained product. It is explicitly not medical advice and does not replace reading the paper directly.