| name | pdf-compression |
| description | Optimize and compress large image-heavy or scanned PDFs by downscaling images and re-encoding them using CCITT Group 4 (black and white) or JPEG (grayscale/color). Use when compressing scanned PDFs, splitting large court PDFs by bookmarks, applying N-up (2-up/4-up) layouts, or when the user says "compress this PDF" / "PDF muito grande". |
PDF Compression
Overview
This skill allows the agent to compress and optimize large, scanned, or image-heavy PDF files. It works by extracting images from the PDF, downscaling them to a reasonable resolution for reading, and re-encoding them with high-efficiency formats (specifically 1-bit CCITT Group 4 TIFF for black & white text, or compressed JPEG for grayscale and color images).
Quick Start
Run the helper script using uv run to compress a PDF. It dynamically installs the required libraries (pymupdf and pillow) so they don't have to be pre-installed globally.
uv run --no-project --with pymupdf,pillow <skill-dir>/scripts/compress.py --input <input.pdf> --output <output.pdf> --mode bw
<skill-dir> is resolved by skills.sh to the installed skill directory (e.g. ~/.claude/skills/pdf-compression), not the source repo checkout. Run skills.sh from the repo root to install with the placeholder resolved automatically.
Utility Scripts
compress.py
The CLI script provides several options to control the compression style and target size:
--input (required): Absolute path to the source PDF file.--output (required): Absolute path to save the compressed/optimized PDF file.--mode: The compression algorithm to use:
auto (default): Automatically detects scanned pages vs. native digital pages. Scanned pages are binarized using bw mode (CCITT Group 4) to achieve max compression. Native digital pages preserve color/grayscale images to prevent degradation of charts, logos, and diagrams.bw: Converts images to 1-bit Black & White and compresses using CCITT Group 4.gray: Converts images to 8-bit grayscale and compresses with JPEG.color: Preserves image colors and compresses with JPEG.
--max-dim (default: 1200): Downscale any image whose width or height exceeds this value, maintaining aspect ratio.--quality (default: 50): JPEG compression quality (1-100) for gray and color modes.--skip-small (default: 150): Do not compress images with both dimensions smaller than this threshold (useful to protect logos, icons, and small vector graphics from compression artifacts).
process_pdf.py
This script splits a large PDF based on its bookmarks (Table of Contents), applies a customizable N-up layout, compresses each split document (with binarization, downscaling, grayscale, and rasterization fallbacks), and re-merges the optimized parts back into a single PDF with rebuilt bookmarks.
Options:
--input (required): Absolute path to the source PDF file.--output-dir (required): Absolute path to the directory to save the split PDFs.--mode: Compression mode (same as compress.py). For documents whose bookmark contains "autos digitalizados" or "digitalizado", B&W mode is automatically forced.--threshold-kb (default: 150): The size limit in KB per page. If a split PDF exceeds this limit after standard compression, it is automatically rasterized to bypass vector/form bloating.--nup (default: 1): Combine N pages from the original PDF into a grid on each page of the output PDF (e.g. 2, 4, 8, etc.).
2up.py
This script combines consecutive pages of a PDF side-by-side (2-up layout) into a single landscape page in a new PDF, keeping text layers fully searchable.
Options:
--input (required): Absolute path to the source PDF file.--output (required): Absolute path to save the 2-up PDF file.
Example Commands:
1. Compress a scanned text document to minimum size (Black & White):
uv run --no-project --with pymupdf,pillow <skill-dir>/scripts/compress.py \
--input "/path/to/document.pdf" --output "/path/to/compressed.pdf" --mode bw
2. Compress a document while preserving colors:
uv run --no-project --with pymupdf,pillow <skill-dir>/scripts/compress.py \
--input "/path/to/document.pdf" --output "/path/to/compressed.pdf" \
--mode color --quality 55 --max-dim 1200
3. Split, N-up (2-up), compress, and re-merge a PDF (with dynamic rasterization fallback for heavy parts):
uv run --no-project --with pymupdf,pillow,opencv-python,numpy \
<skill-dir>/scripts/process_pdf.py \
--input "/path/to/document.pdf" --output-dir "/path/to/split_dir" --threshold-kb 150 --nup 2
4. Combine pages side-by-side (2-up layout) of a single PDF file directly:
uv run --no-project --with pymupdf <skill-dir>/scripts/2up.py \
--input "/path/to/document.pdf" --output "/path/to/2up_document.pdf"
Common Mistakes
- Running with standard Python instead of
uv run: Standard python invocation might fail if pymupdf or pillow are not installed in the global environment. Always run using uv run --no-project --with pymupdf,pillow.
- Skipping
process_pdf.py's optional dependencies: process_pdf.py can also use opencv-python and numpy for adaptive thresholding. If they're missing, it falls back automatically to plain Pillow thresholding (with a warning) rather than failing — add uv run --no-project --with pymupdf,pillow,opencv-python,numpy only if you want the OpenCV-based enhancement.
- Using
bw mode for photos/color-heavy figures: If the PDF has high-resolution colored graphs, photos, or diagrams where color is critical, bw mode will binarize them into high-contrast black and white, making them unreadable. Use color or gray mode for these files.
