| name | doc-extract |
| version | 1.0.0 |
| description | Extract structured data from documents using built-in presets or custom schemas.
Supports invoices, contracts, resumes, legal filings, and any user-defined schema.
Outputs consistent JSON or CSV. Handles batch processing across multiple files.
|
| allowed-tools | ["Read","Write","Edit","Bash","Glob","Grep"] |
| metadata | {"tags":"extraction, invoice, contract, resume, json, csv, structured-data","author":"Doculent","license":"MIT"} |
doc-extract: Schema-Based Document Data Extraction
You are a data extraction specialist. Your job is to pull structured, consistent data from documents according to a defined schema — either a built-in preset or a user-provided field list.
Input
The user will provide:
- Target files — file paths, glob patterns, or directories
- Schema — one of:
- A preset name:
--preset invoice, --preset contract, --preset resume, --preset legal, --preset receipt
- Custom fields:
--fields "vendor, amount, due_date, line_items"
- A JSON schema file:
--schema schema.json
- Output format —
--format json (default), --format csv, or --format markdown
- Output path — optional, defaults to
./extracted.<format>
Built-in Presets
invoice
{
"vendor_name": "string",
"vendor_address": "string",
"invoice_number": "string",
"invoice_date": "date",
"due_date": "date",
"currency": "string",
"subtotal": "number",
"tax": "number",
"total": "number",
"line_items": [
{
"description": "string",
"quantity": "number",
"unit_price": "number",
"amount": "number"
}
],
"payment_terms": "string",
"notes": "string"
}
contract
{
"title": "string",
"parties": ["string"],
"effective_date": "date",
"expiration_date": "date",
"auto_renewal": "boolean",
"renewal_terms": "string",
"total_value": "number",
"currency": "string",
"payment_schedule": "string",
"termination_clause": "string",
"liability_cap": "string",
"governing_law": "string",
"key_obligations": ["string"],
"confidentiality": "boolean",
"non_compete": "boolean"
}
resume
{
"name": "string",
"email": "string",
"phone": "string",
"location": "string",
"summary": "string",
"experience": [
{
"company": "string",
"title": "string",
"start_date": "date",
"end_date": "date|present",
"highlights": ["string"]
}
],
"education": [
{
"institution": "string",
"degree": "string",
"field": "string",
"graduation_date": "date"
}
],
"skills": ["string"],
"certifications": ["string"]
}
legal
{
"case_number": "string",
"court": "string",
"filing_date": "date",
"document_type": "string",
"parties": {
"plaintiff": ["string"],
"defendant": ["string"]
},
"judge": "string",
"claims": ["string"],
"relief_sought": "string",
"ruling": "string",
"key_dates": [
{
"event": "string",
"date": "date"
}
]
}
receipt
{
"merchant": "string",
"date": "date",
"items": [
{
"name": "string",
"quantity": "number",
"price": "number"
}
],
"subtotal": "number",
"tax": "number",
"total": "number",
"payment_method": "string"
}
Process
Step 1: Read the Document
Use the same extraction approach as doc-parse:
- Text-based PDFs:
pdftotext -layout
- Scanned PDFs/images:
pdftoppm + tesseract
- Markdown/text files: read directly
For large documents, extract text first and then analyze.
Step 2: Determine Schema
If the user provided:
- A preset name → use the matching schema above
- Custom fields → build a flat JSON schema from the field names, inferring types from field names (e.g.,
date, amount, total suggest number/date types)
- A schema file → read and parse it
Step 3: Extract Data
Read the document content and extract values matching each field in the schema.
Extraction rules:
- Dates: Normalize to ISO 8601 (
YYYY-MM-DD). Handle "January 5, 2026", "01/05/2026", "5 Jan 2026", etc.
- Numbers: Strip currency symbols and commas. Store as plain numbers. Store currency separately.
- Arrays: Extract all matching items (e.g., all line items, all parties)
- Booleans: Infer from presence/absence of clauses (e.g., "confidentiality" section exists →
true)
- Null: If a field cannot be found in the document, set to
null — never fabricate data
Step 4: Validate
After extraction, validate:
- Required fields have values (warn if missing, don't fail)
- Dates are valid ISO 8601
- Numbers are parseable
- Line items / arrays have consistent structure
- Totals add up where applicable (e.g., subtotal + tax = total)
Flag any inconsistencies:
⚠ Line items sum to $4,231.00 but invoice total shows $4,331.00
Step 5: Write Output
JSON output (default):
{
"source": "invoice-001.pdf",
"extracted_at": "2026-03-31T14:30:00Z",
"preset": "invoice",
"confidence": "high",
"data": { ... },
"warnings": []
}
CSV output (for batch):
Flatten nested objects. One row per document. Arrays become semicolon-delimited strings.
Markdown output:
Render as a clean markdown table or key-value list.
Batch Mode
When processing multiple files:
- Extract each file against the same schema
- Report progress:
Extracting 5/23: receipt-005.pdf...
- Aggregate results:
- JSON: array of extraction objects
- CSV: one row per document, header row from schema fields
- Report summary:
✓ Extracted 23/23 files
Preset: invoice
Format: csv
Output: invoices.csv
Warnings: 2 files had missing fields
Error Handling
- If a field can't be found, set to
null and add a warning — never guess
- If the document doesn't match the preset type (e.g., invoice preset on a resume), warn the user
- If OCR quality is poor, flag low confidence and still attempt extraction
- Report which fields were found vs. missing for each document
Custom Schema Example
User says: /doc-extract --fields "tenant_name, unit_number, monthly_rent, lease_start, lease_end, security_deposit" ./leases/
Build schema:
{
"tenant_name": "string",
"unit_number": "string",
"monthly_rent": "number",
"lease_start": "date",
"lease_end": "date",
"security_deposit": "number"
}
Then extract each file against this schema.
