What problem does it solve? Academic papers in computer vision, machine learning, and robotics often fail because claims outrun evidence, structure buries the contribution, or the writing triggers reviewer objections. This Skill enforces a discipline where every quantitative claim traces to a retained experiment artifact or a verified literature note, and every section is built to survive reviewer scrutiny. ## Core Features & Use Cases - Evidence-bound drafting: Every quantitative claim must link to an experiment ID with retained artifacts or a knowledge note with sufficient reading depth; unverified statements are marked as such. - Scenario-specific guidance: Reference files tailor the writing to CV/ML main conferences (CVPR/NeurIPS/ICML), journal extensions (TPAMI/IJCV), and short workshop papers, each with distinct length, density, and reviewer expectations. - Reviewer self-audit: A walkthrough reference applies four tests (5-minute, surprise, objection, title) to draft paragraphs, showing how to rewrite hype-laden text into defensible claims. - Use Case: You have finished experiments for a CVPR submission. Use this Skill to structure the narrative arc, write the abstract and introduction anchored to the strongest prior work, build the main results table with baselines and uncertainty, and run the pre-submission reviewer walkthrough. ## Quick Start Help me draft the introduction and abstract of my CVPR paper on video spatial reasoning, making sure every claim traces back to my experiment records.
