How does this skill handle different file types?

It implements a specific decision tree: pdfplumber is used for text-based PDFs, AWS Textract is used for images and scanned documents, and email HTML is parsed directly without OCR.

What are 'golden fixtures' in this context?

Golden fixtures are pairs of saved OCR text and 'expected' JSON output that serve as ground-truth tests to ensure that any changes to the parsing logic do not cause regressions.

When should I use Claude Vision instead of a deterministic parser?

Deterministic parsers are preferred for high-volume vendors requiring reliable tax and line-item extraction. Claude Vision serves as a safety net for low-volume or highly variable receipts where you lack enough samples to stabilize patterns.

Does this skill require local OCR dependencies?

No. To maintain consistency and avoid environment-specific issues, the skill follows a core invariant of using AWS Textract for all OCR tasks rather than local binary wrappers.

Deterministic Receipt Parser Engineer

Name: Deterministic Receipt Parser Engineer
Author: ThomasMcCrossin

byThomasMcCrossin

0•

数据科学与机器学习

Develops and maintains high-accuracy receipt and invoice parsers using automated OCR routing and golden fixture testing workflows.

The Receipt Parser Engineer skill provides a structured framework for building robust accounting integrations within the curlys-books ecosystem. It streamlines the process of converting unstructured OCR text from receipts and invoices into standardized data structures by leveraging a deterministic parsing engine. The skill guides developers through sophisticated OCR routing—utilizing pdfplumber for text-based PDFs and AWS Textract for images—while ensuring reliability through a 'golden fixture' testing methodology. It balances precision with flexibility by maintaining specific vendor parsers and utilizing Claude Vision as an intelligent fallback for unknown or low-volume formats.

主要功能

01Automated OCR routing between pdfplumber and AWS Textract

02Standardized normalization of totals, taxes, dates, and line items

03Intelligent fallback to Claude Vision for unknown vendor layouts

04Golden fixture testing framework to prevent regressions in data extraction

050 GitHub stars

06Deterministic vendor-specific parsing logic for high-volume receipts

使用场景

01Debugging and fixing mis-detected fields in existing financial parsers

02Adding support for a new high-volume vendor with unique invoice layouts

03Establishing a regression-proof testing suite for accounting data extraction

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add thomasmccrossin/curlys-books receipt-parser-engineer

For use in Claude.ai and ChatGPT

Download Skill