Paper Fraud Auditor

Use this skill when the user wants to审查论文 PDF、识别论文造假、检查论文图片/表格/实验数据异常、使用耿同学方法论、本福特定律、MinerU PDF 转 Markdown，或要求对论文做严格多轮审查。

Non-negotiables

• Do not claim a paper is fraudulent from one signal. Say "异常", "疑点", "需要人工复核", or "高风险证据链".
• Every finding must cite evidence location: page, figure/table number, markdown line or content block, image path, table row/column, and original value.
• Benford's Law is gated by applicability. If sample size, range, or data type is unsuitable, say it is not applicable and use terminal-digit/roundness checks instead.
• Treat API tokens as secrets. Never write MinerU tokens into SKILL.md, scripts, reports, command history, or examples. Read MINERU_API_TOKEN from the environment.
• Prefer deterministic scripts for extraction/statistical checks, then use agents for interpretation, visual review, and adversarial explanation testing.
• Evidence images must be traceable. Prefer deterministic annotated table renders from the original XLSX/PDF extraction over decorative AI images. Use generative image tools only for presentation/overview graphics, not as primary evidence.

Required References

Read only what is needed: