菜单
Use this skill when describing, summarizing, extracting structured data from, or answering questions about PDF documents, or when planning PDF layout and structure for reports, contracts, and publications. Not for editing binary PDF files directly. Not for converting PDFs to other formats programmatically.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use this skill when designing, cleaning, deduplicating, or documenting datasets for model training and evaluation including schema design, class imbalance handling, and train/val/test splits. Not for running model training or hyperparameter tuning. Not for real-time data pipeline engineering.
Use this skill when building evaluation frameworks to measure LLM quality, safety, accuracy, or alignment including test suites, human eval rubrics, automated evals, and metrics design. Not for training or fine-tuning models. Not for dataset curation or benchmark comparison across publicly available models.
Use this skill when comparing AI or LLM models on benchmarks, capability, cost, latency, context window, or task-specific fit to help teams select the right model for their use case and budget. Not for training or fine-tuning models. Not for building eval frameworks from scratch.
Use this skill when crafting, iterating, or optimizing prompts for LLMs including zero-shot, few-shot, chain-of-thought, role prompting, structured output, and prompt chaining. Not for fine-tuning or training models. Not for evaluating model quality across benchmarks.
Use this skill when designing or improving observability stacks with Prometheus, Grafana, Loki, Jaeger, or configuring alerting rules and SLIs/SLOs. Not for application code debugging or infrastructure provisioning. Not for log analysis of already-collected logs.
Use this skill when generating, planning, or describing structured Word-compatible documents from outlines, raw content, or templates including heading hierarchy, table structures, and formatting conventions. Not for editing existing DOCX files programmatically. Not for PDF layout or presentation slide design.
| name | |
| description | Use this skill when describing, summarizing, extracting structured data from, or answering questions about PDF documents, or when planning PDF layout and structure for reports, contracts, and publications. Not for editing binary PDF files directly. Not for converting PDFs to other formats programmatically. |
| version | 1.0.0 |
| author | community |
| tags | ["files","pdf","documents","extraction","summarization"] |
| license | MIT |
This skill covers working with PDF documents in two primary modes: (1) comprehension and extraction — reading, summarizing, and pulling structured data from PDF content; and (2) layout planning — designing the structure, sections, and visual hierarchy of a new PDF document for reports, contracts, research papers, or publications. For comprehension tasks, it applies structured extraction, hierarchical summarization, and question-answering techniques. For layout tasks, it describes the page structure, section organization, typography, and visual design that tools like LaTeX, ReportLab, WeasyPrint, or Adobe InDesign should implement.
pdf2docx, LibreOffice)| Task | Approach |
|---|---|
| Summarize a PDF | Hierarchical summary: document summary → section summaries → key findings |
| Extract a table | Identify column headers, data types per column, output as Markdown or JSON |
| Find specific data | Quote the exact text with page reference; flag if not found explicitly |
| Identify key clauses | Scan for section headings; extract clause text, obligation, and party responsible |
| Plan PDF layout | Define: page size, margins, header/footer, section hierarchy, font stack, color palette |
| Generate PDF programmatically | Use ReportLab (Python) for data-driven PDFs; WeasyPrint for HTML-to-PDF |
| Compare two PDFs | Identify additions, deletions, and changed clauses between versions |
Understand the document type — Identify what kind of PDF you are working with: research paper, financial report, legal contract, government form, technical manual, or marketing brochure. Each type has different structural conventions and extraction priorities.
Navigate the document structure — Review the table of contents or section headings to build a mental map of the document. For extraction tasks, identify which sections contain the target data before diving in. For long documents (100+ pages), focus on the executive summary, section introductions, and conclusion.
For summarization — apply hierarchical structure — Create a three-level summary:
For data extraction — identify target entities — Define what you are looking for before reading: specific data types (dates, monetary amounts, percentages, names), specific tables (revenue by quarter, list of parties, schedule of deliverables), or specific clauses (termination, liability, payment terms). Then systematically search and extract.
Present extracted data in structured format — Always convert extracted data to a clean format:
Flag ambiguities and gaps — When data is implied but not explicitly stated, flag it. When a section is unclear or contradictory, note it. When the document appears to be missing pages or the extraction may be incomplete, say so explicitly.
For PDF layout planning — define the grid — Specify: page size (A4 or Letter), margins (top/bottom/left/right), number of columns (single, two-column academic style, etc.), gutter width between columns, and baseline grid for text alignment.
Define the typography system — Specify: primary font (serif for formal, sans-serif for modern), heading hierarchy sizes, body text size and line height, caption style, and code/monospace font if needed. Define a limited color palette (typically 2–3 colors: primary, secondary, accent).
Plan the page components — Describe: running headers (chapter title, document title), page numbers (position and format), footer content (confidentiality, date), and any watermarks or security markings.
For programmatic generation — provide the data schema — If the PDF is data-driven (a report generated from a database), define the data schema that feeds the template: field names, types, formatting rules, and conditional logic (e.g., show a warning box if value exceeds threshold).
Input: A 120-page annual report PDF for a publicly traded company. Extract: revenue, net income, EPS, segment breakdown, and key risk factors.
Output:
Document summary:
Acme Corp 2024 Annual Report (120 pages) covers fiscal year ended December 31, 2024.
The company reported record revenue driven by North America growth, offset by EMEA softness.
Key risks include regulatory changes in the EU and supply chain concentration in Southeast Asia.
Financial highlights (extracted from pages 8–12, Consolidated Statements of Operations):
{
"fiscal_year": 2024,
"currency": "USD millions",
"revenue": {
"total": 4823,
"yoy_change_pct": 12.4
},
"gross_profit": {
"amount": 2156,
"margin_pct": 44.7
},
"operating_income": {
"amount": 891,
"margin_pct": 18.5
},
"net_income": {
"amount": 672,
"margin_pct": 13.9
},
"earnings_per_share": {
"basic": 4.21,
"diluted": 4.18
}
}
Segment breakdown (from page 15, Segment Revenue table):
| Segment | Revenue ($M) | YoY Growth | % of Total |
|---|---|---|---|
| North America | 2,841 | +18.2% | 58.9% |
| EMEA | 1,124 | -2.1% | 23.3% |
| Asia Pacific | 858 | +14.7% | 17.8% |
Key risk factors (from pages 34–41, Risk Factors section):
1. Regulatory Risk (EU AI Act) — New regulations may require product modifications
by Q3 2025, estimated compliance cost of $40–60M (page 35)
2. Supply Chain Concentration — 73% of component sourcing from two suppliers in
Vietnam and Thailand; single supplier disruption could reduce production by 30% (page 37)
3. Currency Risk — 41% of revenue denominated in non-USD currencies; 10% USD
strengthening would reduce reported revenue by ~$198M (page 39)
Extraction confidence notes:
Input: A 45-page SaaS Master Service Agreement (MSA). Identify key obligations, risk clauses, and any non-standard terms.
Output:
Contract overview:
Document: Master Service Agreement
Parties: TechVendor Inc. (Vendor) and ClientCorp Ltd. (Customer)
Effective Date: January 1, 2025
Term: 3 years with automatic renewal
Governing Law: State of Delaware, USA
Key obligations by party:
| Obligation | Party | Section | Notes |
|---|---|---|---|
| Maintain 99.9% uptime SLA | Vendor | §5.2 | Credits apply for downtime > 0.1% |
| Pay invoices within 30 days | Customer | §8.1 | 1.5%/month late fee |
| Provide security incident notice within 48 hours | Vendor | §9.3 | Non-standard: industry norm is 72 hours |
| Maintain SOC 2 Type II certification | Vendor | §9.1 | Annual report required |
| Provide 6-month termination notice | Customer | §12.1 | ⚠️ Non-standard: typical is 30–90 days |
Risk flags — clauses requiring legal review:
⚠️ HIGH: §11.3 — Liability cap set at 3 months of fees. Standard is 12 months.
Recommendation: Negotiate to minimum 6 months or last 12 months of fees paid.
⚠️ MEDIUM: §7.2 — Vendor retains right to use Customer data for "product improvement."
Recommendation: Add exclusion for confidential or PII data; require opt-out option.
⚠️ MEDIUM: §12.4 — Auto-renewal with 6-month notice requirement (Customer).
Standard notice period is 30–90 days. Risk of unintended renewal.
ℹ️ NOTE: §13.1 — Vendor can modify pricing with 60-day notice during term.
Standard is fixed pricing for the term. Flag for pricing stability discussion.
Missing standard clauses:
pdfplumber (Python) for accurate table extraction from PDFs with complex layoutsdiff-pdf or Word's compare feature works wellPyMuPDF (fitz) is fastest for extracting raw text; pdfplumber is best for tables; camelot is best for financial tables