What problem does it solve? AI-generated scientific paper drafts often contain unsupported claims, untraceable numbers, missing experiments, and fabricated references. This Skill diagnoses such drafts, runs only feasible experiments, rewrites the experiment section, and repairs the full LaTeX manuscript so every claim is backed by traceable evidence. ## Core Features & Use Cases - Diagnosis and experiment planning: Maps the paper's logic, identifies missing experiments and unsupported claims, and plans a minimal feasible experiment set with pre-run cost approval gates. - Traceable experiment execution: Runs experiments with dataset license checks, recomputes reported values from per-seed raw logs, and audits code-paper consistency across six mandatory reports. - Full manuscript repair: Edits the LaTeX source directly in ./paper/, verifies every reference, checks the rendered PDF layout, and syncs the repaired manuscript to Overleaf via Git. - Use Case: You receive an AI-drafted ML paper with blank result tables. The skill downloads the named datasets, runs the experiments, fills the tables with real numbers, weakens unsupported claims, verifies citations, and pushes the honest repaired manuscript to Overleaf. ## Quick Start Repair the AI-generated paper draft staged under ./input/draft/ by diagnosing it, running feasible experiments, and updating the manuscript in ./paper/ for claim-evidence consistency.
