PDF Intelligence Studio

Insights, Evidence & Action from Complex PDFs

PDF Intelligence Studio transforms one or more PDFs into a reusable, evidence-backed knowledge pack for humans and AI systems.

It is designed to go far beyond a generic summary. The Skill maps the full readable document, adapts to the document type and user goal, separates evidence from interpretation, preserves page traceability, identifies conflicts and limitations, and can produce Markdown, structured JSON, an evidence ledger, and a modern standalone HTML dashboard.

Core promise

Understand the document, verify the evidence, and turn the knowledge into useful decisions and action.

Why this is a premium document-analysis Skill

Generic PDF summarizers often:

• focus on the first pages;
• flatten every document into the same template;
• lose tables and page context;
• mix facts with recommendations;
• invent missing metadata or page references;
• provide summaries that are difficult to reuse.

PDF Intelligence Studio adds a structured document-intelligence workflow:

1. PDF preflight and extraction-health review
2. Page-aware document map
3. Complete section coverage
4. Adaptive document-type analysis
5. Evidence status and traceability
6. Conflict and limitation detection
7. Context-aware practical implications
8. Action planning when requested
9. Human-readable Markdown
10. AI-ready JSON and evidence ledger
11. Standalone knowledge dashboard

Analysis modes

Executive Brief

Produces a concise decision-oriented brief with the central conclusion, material evidence, risks, opportunities, and next steps.

Deep Structured Analysis

Reconstructs the document's logical structure and extracts themes, insights, evidence, frameworks, limitations, applications, and final synthesis.

Strategy & Action Plan

Converts document-grounded ideas into prioritized next steps while clearly separating source recommendations from analyst interpretation.

Research Review

Analyzes thesis, methodology, evidence, findings, limitations, validity, implications, and further research.

Technical / Process Blueprint

Extracts requirements, workflows, architecture, dependencies, roles, failure modes, implementation steps, and validation checks.

Learning Guide

Creates learning objectives, concept explanations, frameworks, examples, glossary, reflection questions, and application exercises.

Document Comparison

Compares two or more PDFs across purpose, assumptions, evidence, agreements, contradictions, methods, risks, recommendations, and unique contributions.

Focused Extraction

Answers specific questions such as:

• What are the requirements?
• What are the main risks?
• Which statistics matter?
• What decisions are required?
• What does the contract say about a specific clause?
• Which process steps and dependencies are described?

Complete Knowledge Pack

Generates the most comprehensive deliverable set:

• Markdown report
• Structured JSON
• Standalone HTML dashboard
• CSV evidence ledger
• Comparison pack when multiple PDFs are attached

Adaptive document intelligence

The Skill changes emphasis according to the document type.

Evidence-first architecture

Every material conclusion can use one of five statuses:

• Verified — directly supported by readable document evidence
• Derived — responsibly synthesized or calculated from verified evidence
• Interpretive — analyst-added implication requested by the user
• Unclear — ambiguous wording, visual, OCR, or source claim
• Unavailable — content could not be read or verified

The evidence ledger can include:

• claim or insight;
• status;
• document;
• page or section;
• evidence description;
• confidence note;
• analyst note.

Document Intelligence Dashboard

The standalone HTML dashboard uses a professional enterprise knowledge-tech visual system.

It can include:

• document identity and metadata;
• extraction and coverage health;
• executive summary;
• core thesis;
• themes;
• key insights;
• section map;
• frameworks and processes;
• evidence ledger;
• risks and opportunities;
• recommendations;
• action plan;
• conflicts and limitations;
• glossary;
• reflection questions;
• final synthesis.

The dashboard is responsive, standalone, evidence-first, and free from fabricated visual metrics.

Open the sample dashboard

Inputs

The user can provide:

LANGUAGE: English
USER_CONTEXT: Marketing Director
DEPTH: Deep
DOCUMENT_TYPE: Business report
OUTPUT_MODE: Strategy and action plan
FOCUS_QUESTIONS: What should we prioritize in the next 90 days?
PRIVACY_MODE: Minimize sensitive data
DELIVERABLES: Markdown, JSON, HTML dashboard, CSV evidence ledger

Missing inputs are inferred conservatively so the user does not need to complete a form before analysis.

Example prompts

Complete analysis

Analyze the attached PDF as a complete knowledge pack.

Language: English
User context: Operations Director
Depth: Deep
Document type: Business report
Output mode: Strategy and action plan

Generate a Markdown report, structured JSON, evidence ledger, and standalone HTML dashboard. Preserve page traceability and clearly label interpretation.

Executive brief

Turn the attached PDF into a concise executive brief. Focus on the central conclusion, material evidence, business risks, opportunities, and decisions required.

Research paper review

Review the attached research paper. Analyze the research question, methodology, sample, findings, limitations, validity, and practical implications. Do not overstate causal claims.

Technical blueprint

Convert this technical PDF into an implementation blueprint with requirements, architecture, dependencies, workflow, failure modes, security considerations, and validation checklist.

Compare documents

Compare the attached PDFs. Preserve each document's perspective and show agreements, contradictions, methodological differences, evidence quality, unique insights, and decision implications.

Focused extraction

From the attached PDF, extract only the obligations, deadlines, exceptions, risks, and unresolved questions. Include page references and do not provide legal advice.

Reliable PDF workflow

The package includes local helper scripts.

Extract a PDF

python scripts/extract_pdf.py document.pdf \
  --out outputs/extracted.json \
  --render-low-text-dir outputs/rendered-pages

The extractor records:

• metadata;
• page count;
• page-aware text;
• word and character counts;
• image count;
• likely scanned pages;
• low-text pages;
• heading candidates;
• extraction coverage.

Validate the structured analysis

python scripts/validate_analysis.py analysis.json --extracted outputs/extracted.json

Validation checks include:

• required output fields;
• duplicate evidence IDs;
• invalid evidence statuses;
• impossible page references;
• missing limitations for partial extraction.

Render the dashboard

python scripts/render_dashboard.py analysis.json --out outputs/dashboard.html

Export the evidence ledger

python scripts/export_evidence_ledger.py analysis.json --out outputs/evidence-ledger.csv

Accuracy rules

PDF Intelligence Studio must not:

• invent metadata;
• invent page numbers;
• invent statistics;
• invent quotations;
• convert unreadable content into facts;
• imply full coverage when extraction is partial;
• silently reconcile contradictions;
• reproduce long passages from the PDF;
• present interpretation as source fact.

Privacy mode

When PRIVACY_MODE: minimize-sensitive-data is selected, the output should avoid unnecessary repetition of names, emails, telephone numbers, addresses, account numbers, IDs, and signatures.

The Skill does not claim legal confidentiality, privilege, or compliance certification.

High-stakes documents

The Skill can summarize and structure legal, medical, tax, financial, insurance, or regulatory documents. It must remain informational and should identify questions for qualified review rather than provide personalized professional advice, diagnosis, or definitive compliance determinations.

Important limitations

• Some scanned PDFs require OCR or page-image inspection.
• Complex tables and charts may need visual review.
• Password-protected or corrupted PDFs require an accessible copy.
• Extraction quality depends on the source PDF and runtime capabilities.
• The analysis does not replace the original document.
• High-stakes conclusions should be independently reviewed.

Package structure

pdf-intelligence-studio/
├── SKILL.md
├── README.md
├── agents/
│   └── openai.yaml
├── capafy/
├── config/
├── schemas/
├── scripts/
├── templates/
├── references/
├── workflows/
├── examples/
└── tests/

Version

1.0.0