DocumentIndex Skills

Comprehensive skills for processing, indexing, searching, and extracting information from structured documents like SEC filings, earnings calls, and research reports.

Quick Decision Guide

Choose the right skill based on your needs:

•Need to process a new document? → document-indexing.md
•Looking for specific content in a document? → node-searching.md
•Have a specific question to answer? → agentic-qa.md
•Need ALL evidence on a topic for compliance/audit? → provenance-extraction.md

Available Patterns

Core Skills

•Document Indexing - Transform unstructured documents into hierarchical tree structures with summaries, metadata, and cross-references
•Node Searching - Find relevant document sections using LLM reasoning instead of vector similarity
•Agentic Question Answering - Answer questions through iterative reasoning with confidence scoring and full reasoning traces
•Provenance Extraction - Exhaustively scan entire documents to find ALL evidence for topics with progress tracking

When to Use This Skill

Use DocumentIndex patterns when you need to:

•Process and index financial documents (10-K, 10-Q, 8-K, earnings calls)
•Search for specific information within structured documents
•Answer questions about document content with citations
•Extract comprehensive evidence for compliance, audits, or research
•Navigate complex document hierarchies (PART → ITEM → Section → Note)
•Follow cross-references between document sections

Complexity Levels

•Beginner: Skill 1 (Document Indexing) - Start here to understand the foundation
•Intermediate: Skills 2-3 (Node Searching, Agentic QA) - Build on indexed documents
•Advanced: Skill 4 (Provenance Extraction) - Comprehensive evidence gathering

Learning Path

•Start with Document Indexing - Learn how to transform raw documents into queryable structures
•Explore Node Searching - Understand how to find relevant sections using LLM reasoning
•Try Agentic QA - Answer specific questions with confidence scoring and citations
•Master Provenance Extraction - Extract exhaustive evidence for compliance and research

Recommended progression:

•First-time users: Read skills in order (1 → 2 → 3 → 4)
•Experienced users: Jump to the skill matching your use case via the Quick Decision Guide

Typical Workflow

code

1. Index Document (Skill 1)
   ↓
2. Choose your approach:
   ├─ Quick question? → Agentic QA (Skill 3)
   ├─ Find sections? → Node Searching (Skill 2)
   └─ Need ALL evidence? → Provenance Extraction (Skill 4)

Configuration Trade-offs

Speed vs. Quality

•Fast/Cheap: Lower thresholds, fewer iterations, no summaries
•Balanced: Default configurations (recommended for most use cases)
•Thorough: Higher thresholds, more iterations, with summaries

Coverage vs. Precision

•Maximum coverage: Low relevance thresholds (0.5-0.6)
•Balanced: Medium thresholds (0.6-0.7)
•High precision: High thresholds (0.7-0.8+)

Working Examples

See examples/ directory for complete, runnable code examples:

•indexer_deep_dive.py - Document indexing patterns
•searcher_showcase.py - Node searching strategies
•agentic_qa_tutorial.py - Question answering workflows
•provenance_patterns.py - Evidence extraction patterns
•basic_usage.py - Getting started examples
•caching_example.py - Performance optimization
•streaming_example.py - Streaming responses
•multi_provider_example.py - Multi-LLM support

See examples/README.md for setup instructions and detailed descriptions.

Integration Patterns

Sequential Processing

code

DocumentIndexer → NodeSearcher → AgenticQA
(Index once)   → (Find sections) → (Answer questions)

Parallel Analysis

code

DocumentIndexer → ProvenanceExtractor (multiple topics in parallel)
(Index once)   → (Extract evidence for: climate, cyber, regulatory, etc.)

Hybrid Approach

code

DocumentIndexer → AgenticQA (quick questions)
              ↓
              → ProvenanceExtractor (comprehensive evidence when needed)

Performance Considerations

Component	Time	Cost	Use Case
Document Indexing	30-60s	$0.10-0.50	One-time per document
Node Searching	2-5s	$0.01-0.05	Per search query
Agentic QA	5-15s	$0.02-0.10	Per question
Provenance Extraction	30-90s	$0.05-0.20	Per topic extraction

Additional Resources

•Source Code: /documentindex package
•Documentation: Project README
•Main Skills README: ../README.md

Decision Framework Summary

code

┌─────────────────────────────────────────┐
│ What do you need to do?                 │
└─────────────────────────────────────────┘
                 │
    ┌────────────┼────────────┐
    │            │            │
    ▼            ▼            ▼
New Doc?    Find Content?  Answer Q?
    │            │            │
    ▼            ▼            ▼
Indexing    Searching      QA
(Skill 1)   (Skill 2)   (Skill 3)
                             │
                    Need ALL evidence?
                             │
                             ▼
                        Provenance
                        (Skill 4)

Last Updated: 2026-02-01 Skills Version: 1.0 Compatible with: DocumentIndex v0.1.0+