kordoc

The engine handles word breaking (line wrapping) differently for English (Latin) and Korean (Non-Latin) characters to match Hancom (HWP) behavior. Note that the property names for Non-Latin characters are counter-intuitive:

Recommended Production Setting: To match standard official documents, use:

• breakNonLatinWord="BREAK_WORD" (Keeps Korean words/eojeol intact)
• breakLatinWord="KEEP_WORD" (Keeps English words intact)

Note: Only single tokens that are longer than the line width (like long URLs) will be forcibly broken.

kordoc supports automatic form filling for official document templates. It can recognize and process several patterns:

• Label-Value cells: Identifying pairs in tables.
• Checkboxes: Converting □ to ☑.
• Parentheses: Filling empty spaces like 일반( )통 $\rightarrow$ 일반(3)통.
• Annotations: Converting patterns like (한자：) to (한자：金).

Additionally, AI agents can interact with these forms using the fill_form MCP tool.

Every BIFF record consists of a 4-byte header followed by variable-length data. The header contains an opcode (16-bit) and a length (16-bit). If the data length exceeds 8224 bytes, the record is split into multiple segments using CONTINUE (0x003C) records.

Record Header Layout:

• opcode: 16-bit Little-Endian (LE) record type.
• length: 16-bit LE data length.
• data: The actual payload of length bytes.

Implementation Logic: Iterate through the stream by reading the opcode and length, then advancing the offset by 4 + length until the end of the stream is reached.

while offset < stream.length:
  opcode = readU16LE(offset)
  length = readU16LE(offset + 2)
  data = stream[offset+4 : offset+4+length]
  yield { opcode, length, data }
  offset += 4 + length

Format Profiles allow you to extract and apply specific document styles (borders, shading, column widths, and cell fonts) without sharing the original document content.

1. Extract a profile: Use hwpxToProfile(hwpx) to get a JSON representation of the reference style.
2. Apply a profile: Use markdownToHwpx(md, { profile }) to apply that style to a new document.

The schema for these profiles is defined in docs/format-profile-spec.md.

When parsing PDFs with parsePdf, kordoc provides quality signals to help determine if OCR (Optical Character Recognition) is required. kordoc does not perform OCR itself but provides the metadata needed to route documents to an OCR pipeline.

Key Quality Fields:

• pageQuality: A per-page quality signal.
• qualitySummary: A summary of the document's text quality.
• needsOcr: Boolean indicating if the text layer is likely broken or unreadable.
• ocrReason: A string explaining why OCR is recommended (e.g., broken ToUnicode/CMap mappings or control characters).

While specific laws do not dictate table formatting, administrative and academic conventions (KAMJE) are widely used.

Ending a Table

• If the table ends at the last cell: Leave two spaces (one character width) from the left boundary and write 끝.
• If the table ends mid-way (empty cells remain): Do not write 끝.. Instead, write 이하 빈칸 on the next line/cell.

Cell Alignment (Administrative Default)

• Text (문자): Left-aligned
• Numbers (숫자): Right-aligned (align by decimal point if applicable)
• Headers (제목행): Centered

Table Styling

• Units: Place (단위: 천원) at the top right of the table or in the column header.
• Empty/N/A Cells: Use - or 해당없음.
• Font: Typically slightly smaller than body text (13-14pt) using sans-serif fonts like 한컴돋움.
• Line Spacing: Default is 123% in HWP (Hangul).

kordoc provides parsers for a wide range of document formats, converting them into structured Markdown or IR (Intermediate Representation) blocks:

• HWP/HWPX: Supports HWP 3.0 (legacy binary), HWP 5.x (binary), HWPML 2.x (XML), and HWPX (modern XML). Includes support for DRM-protected (read-only) files via COM API fallback on Windows.
• PDF: Supports standard text extraction, 2-column layout detection (for papers/reports), and table detection (both line-based and cluster-based for borderless tables).
• XLSX: Excel spreadsheet parsing with support for shared strings, merged cells, and multiple sheets.
• DOCX: Microsoft Word parsing with support for styles, lists, footnotes, hyperlinks, and images.
• OCR: For scanned or image-based PDFs, you can enable OCR using the ocr: true option (supports PP-OCRv5 Korean) or use external providers like Tesseract or Claude Vision.

Format profiles allow you to reproduce a table's borders, shading, measured column widths, and cell fonts without needing the original source document. This is useful for sharing an organization's formatting (e.g., a specific government report style) without leaking sensitive content.

To use this feature:

1. Extract a style-only JSON profile from a reference HWPX file using hwpxToProfile(hwpx).
2. Apply that profile to a new document using markdownToHwpx(md, { profile }).

The schema for these profiles is defined in docs/format-profile-spec.md.