rf-doc-extract

name

rf-doc-extract

description

Extract specific content from a document — key concepts and definitions, data tables, figures descriptions, methodology details, named entities (authors, institutions, tools), or a specific section. Targeted extraction for when you need particular content rather than a full summary. Trigger phrases: "extract the data tables", "pull out the methodology", "find all definitions", "get the key concepts", "extract figures", "/rf-doc-extract".

Targeted Document Extraction

Extract specific types of content from a document with precision. This is a surgical tool — the user tells you what they want pulled out, and you find it cleanly.

Before You Begin

Load the ingest report for this document from _rf-output/doc-analysis/. If none exists, ask the user to run /rf-doc-ingest first.

Step 1 — Extraction Target

Ask: "What do you want to extract from this document?"

Present the extraction types:

1. Key concepts and definitions — technical terms defined in the document, with their definitions
2. Data and statistics — numerical results, percentages, measurements, sample sizes
3. Tables — reproduce all or specific tables in Markdown format
4. Figures — describe figures, charts, and diagrams with captions
5. Methodology — research design, tools, instruments, procedures
6. Named entities — authors, institutions, tools, datasets, geographic references
7. Arguments and claims — logical claims with their supporting evidence
8. Specific section — the user names a section or page range to extract verbatim or summarized

Allow multiple selections.

Step 2 — Scope

Ask: "Should I extract from the entire document, or specific sections/chunks?"

If the document is long (from ingest report), extracting from priority chunks first is faster. Ask if they want to start with priority sections.

Step 3 — Execute Extraction

For each selected extraction type:

Key Concepts and Definitions

• Find all instances where a term is explicitly defined (look for "is defined as", "refers to", "we define", "means", "i.e.")
• Also extract technical jargon used without definition (flag as "implied term")
• Format: **Term:** [definition] (p.[N])

Data and Statistics

• Extract all numerical claims with context (what was measured, in what conditions)
• Note sample size, confidence intervals, p-values where present
• Flag data that seems inconsistent or surprising
• Format as a table: | Metric | Value | Context | Page |

Tables

• Reproduce each table in Markdown
• Add a brief note on what each table shows and its relevance
• Flag tables with incomplete or ambiguous data

Figures

• Describe each figure: type (bar chart, scatter plot, diagram, photo), what it shows, key takeaway
• Note if the figure supports or contradicts the text claims

Methodology

• Research design label (RCT, case study, meta-analysis, simulation, etc.)
• Sample: size, selection method, demographics
• Instruments: surveys, sensors, software, APIs
• Procedure: step-by-step as described in the document
• Analysis method: statistical tests, qualitative coding, ML models, etc.

Named Entities

• Authors and their affiliations
• Institutions and organizations mentioned
• Software tools and versions
• Datasets and their sources
• Geographic locations

Arguments and Claims

• Each claim as a numbered statement
• Supporting evidence (direct quote or paraphrase + page)
• Type: empirical / theoretical / anecdotal / assumed

Specific Section

• Reproduce the target section with page reference
• Option to summarize instead of verbatim reproduction for long sections

Step 4 — Output

Save _rf-output/doc-analysis/{date}_{run}_{doc-name}_extract.md:

# Extraction — [Document Title]
**Extraction types:** [list]
**Scope:** [full document / sections: list]

## [Extraction Type 1]
[Content]

## [Extraction Type 2]
[Content]

## Flagged Items
[Anything unusual, ambiguous, or worth reviewing]

After saving, suggest: "Use /rf-cite-generate to create a citation for this document, or /rf-lit-synthesize to combine it with other papers."