Generate publication-grade peer reviews of academic papers (PDF) via a multi-stage neuro-symbolic pipeline: PDF→Markdown ingestion, five orthogonal analyses (story, presentation, evaluations, correctness, significance) with accumulated context, synthesis, self-critique, revision, and final quality check. Each analytical stage runs with dual grounding (original PDF for visual context + Markdown for symbolic structure) and stage-appropriate tool augmentation (code interpreter for evaluations/correctness, web search for significance). Preserves the closed-loop pattern `decompose → analyze → synthesize → self-verify → revise` that distinguishes self-improving inference from one-shot generation.
Use this skill whenever the user asks to: review a paper, write a peer review, critique a manuscript, generate an AI review, evaluate a submission for a conference/journal, check a paper's correctness or significance, assess a thesis or preprint, or any variant of "review this PDF" where the input is an academic paper. Tri
2026-05-02