Detect prompt-injection content in PDFs using an eight-layer analysis pipeline.
npx skills add https://github.com/Tuguberk/GhostPrompt --skill ghostprompt
Please help me install this Agent Skill. Skill: ghostprompt Source: https://github.com/Tuguberk/GhostPrompt/tree/main Command: npx skills add https://github.com/Tuguberk/GhostPrompt --skill ghostprompt
GhostPrompt detects prompt-injection attempts and other malicious content in PDFs, enabling safe AI workflows by preventing data leakage and instruction manipulation.
Upload a PDF to GhostPrompt and request a safety assessment prior to ingestion.
High-intent search queries and answers about installing and using this skill.
To detect prompt injection in PDF documents, an eight-layer analysis pipeline scans metadata, invisible text, encoding, streams, zero-width characters, annotations, XMP updates, and semantic camouflage to output a SAFE, SUSPICIOUS, or DANGEROUS verdict.
Malicious content hidden in PDFs for AI workflows includes invisible text, obfuscated encodings, zero-width characters, annotation streams, and semantic camouflage designed to manipulate language model instructions or leak data.
Scanning PDFs for invisible text and zero-width characters is handled by specific layers in an eight-step analysis pipeline that flags obfuscated content and hidden prompt injections before outputting a final safety verdict.
Automated PDF safety assessment is suitable for enterprise document intake, content moderation, and QA pipelines, providing SAFE, SUSPICIOUS, or DANGEROUS verdicts to quarantine risky files before language model processing.
Pdfminer.six extracts text for prompt injection detection by feeding document content into an eight-layer pipeline that analyzes streams, metadata, and semantic camouflage to identify malicious instructions.
The best way to quarantine risky PDFs in content moderation pipelines is applying an eight-layer analysis that outputs a DANGEROUS or SUSPICIOUS verdict, allowing systems to isolate files before AI ingestion.
tomlupoExtract text, images, tables, and metadata from PDF documents.
RedHatProductSecurityDetect malicious components in registry models using signature scanning and metadata analysis.
robotlearning123Create, edit, and analyze Word documents using Python libraries.
jbovetDetect obfuscated payloads in base64, hex, and encoded tokens.
