What problem does it solve? Scientific drafts often contain unreproducible numbers, overstated claims, inconsistent citations, and weak figures that a single proofread misses. This Skill runs a structured critique-grade-revise loop that pushes an existing draft past a measurable quality bar while protecting against score inflation and fabricated fixes. ## Core Features & Use Cases - Five specialist judges: figures, scientific content, style, formatting, and code reviewers each emit schema-validated findings with concrete, executable fixes. - Independent peer reviewer: a fresh grader each iteration re-derives scores on the same axes, spot-checks numbers and citations against the dataset, enforces hard gates, and guards against surface compliance. - Sandboxed revision loop: the writer fixes code first, regenerates figures by running the user's plot command inside a sandbox, grounds new citations via the literature-search toolchain, and never touches the original files. - Use Case: You have a draft paper with a dataset, figures, and a plotting script. The loop recomputes the headline statistic, catches a bug inflating r from 0.62 to 0.98, fixes the code, regenerates the figure, de-causalizes the claims, and iterates until the peer-review score clears 85. ## Quick Start Ask the assistant to run the scientific-writer loop on your draft, providing the draft path, dataset files, figures directory, and the command that regenerates your figures.
