What problem does it solve? Authors submitting to CS conferences often discover reviewer-catchable weaknesses, overclaimed statements, and formatting risks only after submission. This Skill runs a pre-submission adversarial review that finds issues, adjudicates whether each one is valid, and applies only safe minimal LaTeX patches with author sign-off. ## Core Features & Use Cases - Courtroom review engine: N domain reviewers read the whole paper, contested issues go to a two-sided jury trial, and each verdict lands as fixable, author-required, or invalid, tracked in a durable JSON ledger. - Guarded LaTeX editing: Direct-edit mode drafts minimal patches (polish, de-AI, translate Chinese notes, compress, captions) with deterministic guards for anchors, cross-references, patching, and real LaTeX compilation. - Submission-readiness checks: Deterministic desk-reject screening plus compile-driven layout checks, degrading honestly when no LaTeX toolchain exists. - Use Case: Before submitting to CVPR or ACL, ask it to review your draft; it returns an issue ledger with evidence and verdicts, applies safe fixes you approve, and queues anything needing new experiments for your judgment. ## Quick Start Ask the assistant to review your paper before submission, for example by saying: review my LaTeX paper and flag weak claims, missing evidence, and formatting risks.
