pdf-review
Default to this Skill for an unqualified PDF review and for checks involving searchable text, structure, tables, page-located claims, numeric consistency, citations, scanned pages, OCR, rendering, com
SKILL.mdskills/pdf-review/SKILL.md
GitHubEnglish
PDF Review
Read references/productivity-evidence-and-file-safety.md and references/pdf-capability-matrix.md.
Treat an unqualified request such as “Review this PDF” as this Skill's default. Use pdf-review-basic only when the user explicitly asks for a basic, lightweight, text-only check without OCR, rendering, scanned-content, table, image, or complex-layout review. Capability gaps do not transfer a generic request to the basic Skill; continue here with a clearly stated reduced scope.
- Confirm the review objective: content, figures, citations, completeness, tables, page layout, or all applicable dimensions.
- Detect available capabilities separately for metadata, text extraction, page mapping, table extraction, rendering, and OCR. State the document mode and which checks are possible before drawing conclusions.
- Preserve page locations from the tool output. Never infer a page number from text order when page mapping is unavailable.
- Review structure, table of contents, missing or duplicate pages when verifiable, extraction corruption, tables, figures, numeric contradictions, references, headers/footers, and cross-references.
- Render representative and issue-related pages for visual review when possible. Treat OCR as uncertain evidence and quote it only with an OCR label and confidence warning.
- Report each finding with page or location status, evidence, impact, severity, confidence, and recommendation. Separate content findings from extraction/OCR and layout findings.
Keep the original PDF read-only. If rendering or OCR is unavailable, explicitly exclude visual or scanned-content conclusions.
Output Contract
- Scope, PDF mode, and capability matrix
- Document and page-structure summary
- Findings with page/location status, severity, evidence, impact, and confidence
- Content, table, citation, extraction/OCR, and visual findings
- Unreviewed pages or capabilities and limitations
简体中文
PDF 审查
读取 references/productivity-evidence-and-file-safety.md 和 references/pdf-capability-matrix.md。
“审查这个 PDF”等未限定范围的请求默认由本 Skill 处理。只有用户明确要求基础、轻量、纯文本,且不做 OCR、渲染、扫描内容、表格、图片或复杂版式审查时,才使用 pdf-review-basic。能力工具不足不会把通用请求转给 basic;应继续在本 Skill 内明确降级范围。
- 确认审查目标:内容、数字、引用、完整性、表格、页面版式或其中适用组合。
- 分别检测元数据、文本提取、页码映射、表格提取、渲染和 OCR 能力;在得出结论前声明文档模式和可执行检查。
- 保留工具返回的页面位置;无法映射页面时,绝不根据文本顺序猜页码。
- 在可验证范围内检查结构、目录、缺页或重页、提取乱码、表格、图形、数字矛盾、引用、页眉页脚和交叉引用。
- 工具允许时渲染代表页和问题页进行视觉检查;OCR 只作为不确定证据,引用时标记 OCR 和置信度风险。
- 每个发现给出页码或位置状态、证据、影响、严重度、置信度和建议,并区分内容、提取/OCR 和版式问题。
原 PDF 保持只读。无法渲染或 OCR 时必须排除视觉或扫描内容结论。
输出契约
- 范围、PDF 模式与能力矩阵
- 文档和页面结构摘要
- 含页码/位置状态、严重度、证据、影响和置信度的发现
- 内容、表格、引用、提取/OCR 和视觉发现
- 未审页面或能力及限制