About this skill

Problem

Bank statements often arrive as MT940, domestic bank Excel files, or PDFs with wrapped headers, multi-page tables, and inconsistent column names. Manual normalization can mismatch transaction dates, debit/credit direction, counterparty names, and balances. imp-trm-accstmt converts these files into standard transaction import templates.

How It Works

Parsing: Supports MT940, bank Excel, and PDF. PDF parsing can use PyMuPDF text positions to reconstruct columns, handle wrapped headers, and merge multi-page tables.
Mapping: Field mappings are managed via JSON configuration, with cosine similarity used to match close column names.
Output: Supports target templates such as BIPV5, EAS_YXH, FINGARD, NSTC, and YYNCC.
Batch merge: Multiple statements in a directory can be merged into merged_BIPV5.xlsx.

Boundaries

It is best for statements with a text layer or relatively stable layouts. Encrypted PDFs must be decrypted first, and scanned PDFs require an optional OCR backend. Some mixed English/Chinese layouts, such as the Pufa Bank sample, have extraction limits and may benefit from prior conversion to structured text.

Screenshots

Use Cases

Finance converts China Merchants and CITIC bank Excel statements into a BIPV5 import template each month.
Treasury staff merge multiple bank PDF statements in a directory into merged_BIPV5.xlsx.
Data engineers parse MT940 files and map fields to Yonyou NCC or Kingdee EAS templates.
Analysts reconstruct wrapped PDF column headers from text positions before exporting transaction details.

Best For

Finance: monthly conversion of multiple bank Excel statements into BIPV5 import templates.
Treasury: merging a directory of bank PDF statements into one standard file.
Data engineers: mapping MT940 fields to Kingdee EAS or Yonyou NCC templates.
Analysts: extracting transaction details from wrapped-header text PDFs.