What is rw-research-skill?

RW Research Skill is a collection of 20 separate skills for AI agents like Codex or Claude, aimed at helping academic researchers move a project forward one step at a time, from an early idea through literature review, evidence checking, study design, writing, and finally submitting to a journal. Rather than trying to do everything in one pass, it picks a single relevant skill for whatever stage you are currently stuck on. You can hand it many different kinds of material and it will respond differently depending on what you give it. A folder of notes and saved research material gets scanned to figure out what you have already used, written about, or only briefly touched. A vague research interest gets turned into a clearly defined, testable question. A pile of papers gets searched, checked for reliability, and organized so the relationships between studies, including where they agree, conflict, or leave gaps, are visible. A research plan gets checked for weaknesses in its design, sample, measurements, and possible alternative explanations for its results. A manuscript with statistics gets checked to see if the reported numbers match the actual analysis. A full draft gets organized and checked for whether claims are backed by the cited sources. Reviewer comments get split into individual tasks so only the approved changes get made, with a record of what was changed and why. Installation happens through a command line tool. You can install just the entry point skill, a single skill you need right now, or all 20 at once. The project also tracks its own accuracy: a cross model test comparing agents using these skills against agents without them, on a fixed set of synthetic tasks, found a large gap in favor of using the skills, though the authors are careful to note this result applies only to that specific test set and not as a general claim about improved research outcomes. No personal papers, private data, or real review records are included in the public repository, only synthetic examples. The project is released under the Apache 2.0 license.