dsh-pdf: A PDF Reading Plugin for DeepSeek Harness
dsh-pdf is a DeepSeek Harness (DSH) plugin designed to address pain points when AI assistants read PDFs, including file size limits (overcoming the 64KB cap) and the inability to read scanned or image-only pages. Maintained by henryxiao709, it extracts the text layer using pdfjs-dist and provides an automatic OCR fallback mechanism. It supports page-by-page reading, multilingual documents, and multiple OCR engines (such as Windows WinRT and tesseract.js), allowing users to control the reading range and mode via the `read_pdf` tool. Installation requires a Node.js environment, and the plugin can be loaded automatically through command injection or a configuration file.
Read Morezhtx2024/dsh-pdf: DeepSeek Harness PDF Parsing Toolkit
zhtx2024/dsh-pdf is a PDF parsing plugin for DeepSeek Harness (DSH), designed to resolve garbled characters and missing text when standard rendering engines process non-embedded Chinese fonts, such as in schematics exported from JLCPCB or EasyEDA. The plugin provides three tools: retrieving document metadata, extracting text, and rendering images as PNG. It employs a hybrid engine architecture: by default, it uses the PDF.js engine, and automatically switches to a built-in decoder when non-embedded CJK fonts are encountered, mapping CID codes to Unicode and rendering the text using system fonts. A session restart is required after installation. Note that the plugin cannot extract text from scanned documents, and large files may consume significant memory.
Read Moredsh-pdf: Reading PDF Text, Metadata, and Page Ranges Locally in DeepSeek Harness
`sunshine-lang/dsh-pdf` is a DeepSeek Harness (DSH) plugin licensed under MIT, designed to locally parse PDF files. It leverages PDF.js to extract text, metadata (such as title and total page count), and content from specified page ranges in an offline environment, without requiring an API key or network connection. Its core functionality includes the `pdf_read` tool, which supports text extraction page by page with page number annotations and allows flexible selection of page ranges via formats like `"1-3,5"` or `"all"`. To ensure security and efficiency, the plugin incorporates built-in limits for file byte size (default: 20 MiB), maximum pages per parse (default: 500 pages), and maximum characters per invocation (default: 12,000 characters), triggering an error when these limits are exceeded.
Read More