academic-paper-reviewer

Provide the manuscript and request a full review. The skill simulates an Editor-in-Chief, three reviewers, and a Devil's Advocate, each producing evidence-based reports with scores across five dimensions and a final editorial decision letter.

Citation verification spot-checks 5-10 random references per review using the CrossRef API for DOIs, Semantic Scholar for title search, and PubMed or arXiv for domain-specific papers. References that cannot be found are flagged as unverifiable in the report.

Yes, re-review mode loads the original review and revised manuscript, then checks every original concern as addressed, partial, or unaddressed. It produces a traceability matrix and re-scores the manuscript, flagging regressions greater than three points.

No, the skill is strictly read-only and produces review reports without modifying the manuscript. Every criticism includes what is wrong, where it is located, and a proposed fix for the author to implement.

Scores are ordinally meaningful but not cardinally interpretable without calibration against a gold set of 5-20 papers with known outcomes. Calibration mode reports false negative rate, false positive rate, and balanced accuracy to contextualize subsequent reviews.