Imported from Yoojin-nam/hwp-pipeline (SKILL.md). Install upstream with npx skills add Yoojin-nam/hwp-pipeline. Copyright stays with the author (MIT).

hwp-pipeline

Read, convert, and edit Korean Hangul documents (.hwp / .hwpx) without Hancom Office.

What this skill does

Three capabilities in one skill:

1. Read HWP: Convert .hwp to JSON, Markdown, or HTML via @ohah/hwpjs (Node.js)
2. Convert HWP to HWPX: Transform .hwp to editable .hwpx via hwp2hwpx (Java)
3. Edit HWPX: Create, modify, and template-fill .hwpx documents via python-hwpx

When to use

• "Convert this HWP file to Markdown"
• "Extract images from a Hangul document"
• "Fill in this government form template (.hwp)"
• "Replace placeholders in this HWPX document"
• "Create a new HWPX document with tables"

When not to use

• The file is .docx or .pdf (use the appropriate tool)
• OCR or scanned PDF recovery is needed

Prerequisites

Run setup.sh to install all dependencies automatically.

Quick decision tree

1. Extract text from HWP → Use @ohah/hwpjs (Section A)
2. Convert HWP to Markdown/HTML/JSON → Use @ohah/hwpjs (Section A)
3. Edit or fill an HWP form → Convert to HWPX first (Section B), then edit (Section C)
4. Extract text from HWPX → Use scripts/text_extract.py (Section C)
5. Replace placeholders in HWPX → Use scripts/zip_replace_all.py (Section C)
6. Create a new HWPX document → Use python-hwpx directly (Section C)

Section A: Read HWP

Uses @ohah/hwpjs (Node.js CLI). Works on macOS, Linux, and Windows.

Install

npm install -g @ohah/hwpjs
export NODE_PATH="$(npm root -g)"

Convert to JSON

hwpjs to-json document.hwp -o output.json --pretty

Convert to Markdown

hwpjs to-markdown document.hwp -o output.md --include-images
# or save images separately:
hwpjs to-markdown document.hwp -o output.md --images-dir ./images

Convert to HTML

hwpjs to-html document.hwp -o output.html

Extract images only

hwpjs extract-images document.hwp -o ./images

Batch processing

hwpjs batch ./documents -o ./output --format json --recursive

Section B: Convert HWP to HWPX

Uses hwp2hwpx (Java, Apache 2.0 by neolord0). Converts binary .hwp to ZIP-based .hwpx so it can be edited with python-hwpx.

First-time setup

bash setup.sh

This clones hwp2hwpx, patches the compiler version, builds a fat JAR, and installs python-hwpx.

Convert

SKILL_DIR="$(cd "$(dirname "$0")" && pwd)"  # or set to skill install path
java -jar "$SKILL_DIR/java/hwp2hwpx-fat.jar" input.hwp output.hwpx

If the fat JAR is not available, use the classpath method:

cd "$SKILL_DIR/java"
CP=".:target/hwp2hwpx-1.0.0.jar:$(find ~/.m2/repository/kr/dogfoot -name '*.jar' | tr '\n' ':')"
java -cp "$CP" Convert input.hwp output.hwpx

Limitations

• hwpparser (Python) is NOT supported: it loses table structure during conversion.
• Polaris Office cannot produce HWPX output.

Section C: Edit HWPX

Uses python-hwpx (MIT by airmang). Full API reference: references/api.md.

Install

pip install -U python-hwpx lxml

Extract text from HWPX

python3 scripts/text_extract.py input.hwpx
python3 scripts/text_extract.py input.hwpx --include-nested --format json

Create a new HWPX document

from hwpx import HwpxDocument

doc = HwpxDocument.new()
doc.add_paragraph("Document title")

table = doc.add_table(2, 2)
table.set_cell_text(0, 0, "Name")
table.set_cell_text(0, 1, "Value")

doc.save_to_path("output.hwpx")

See examples/01_create_and_save.py.

Replace placeholders (including table cells)

python3 scripts/zip_replace_all.py template.hwpx output.hwpx \
  --replace "{name}=John" "{date}=2026-01-01" --auto-fix-ns

See examples/03_template_replace.py for the full pipeline.

Style-based search and replace

with HwpxDocument.open("input.hwpx") as doc:
    doc.replace_text_in_runs("TODO", "DONE", text_color="#FF0000")
    doc.save_to_path("output.hwpx")

Fix namespaces after ZIP-level edits

python3 scripts/fix_namespaces.py input.hwpx --inplace --backup

Workflow patterns

Pattern 1: Fill a government/institutional HWP form

input.hwp → [hwp2hwpx] → template.hwpx → [zip_replace_all.py] → filled.hwpx

1. Convert HWP to HWPX: java -jar java/hwp2hwpx-fat.jar form.hwp form.hwpx
2. Identify placeholders: python3 scripts/text_extract.py form.hwpx --include-nested
3. Replace placeholders: python3 scripts/zip_replace_all.py form.hwpx filled.hwpx --replace "{field}=value" --auto-fix-ns

Pattern 2: Extract and analyze HWP content

input.hwp → [hwpjs] → output.json or output.md

Pattern 3: Create a new structured HWPX

doc = HwpxDocument.new()
# Add paragraphs, tables, memos
doc.save_to_path("output.hwpx")

Unstable areas

• set_header_text() and set_footer_text() may produce inconsistent layouts depending on the document version. If headers/footers cause problems, fix them in the template and only automate body, tables, and memos.
• Always verify output files after automated processing.

Bundled resources

• references/api.md — python-hwpx API signatures and usage notes
• scripts/text_extract.py — One-command text extraction CLI
• scripts/zip_replace_all.py — Global placeholder replacement including table cells
• scripts/fix_namespaces.py — XML namespace normalization after ZIP-level edits
• examples/01_create_and_save.py — Create a new document with paragraphs and tables
• examples/02_extract_and_inspect.py — Extract text and inspect document structure
• examples/03_template_replace.py — Full pipeline: placeholder replacement + namespace fix

Done when

• Requested conversion output exists and is valid
• Template placeholders are replaced and verified
• Batch processing count matches input count
• Output file opens correctly (verify with text extraction)