v0.4.0: PDF Reference Extraction + AI Prompt Injection Detection

What's New

PDF Reference Extraction (bibguard pdf)

• Extract bibliography directly from compiled PDF files — no .bib required
• PyMuPDF-based reference section detection (supports [1], 1., and blank-line formats)
• Two-strategy API resolution: Crossref bibliographic query → Semantic Scholar fallback
• Parallel resolution with thread pool
• Outputs standard .bib files ready for full verification pipeline

bibguard pdf paper.pdf --bib-out refs.bib          # extract only
bibguard pdf paper.pdf --bib-out refs.bib --verify  # extract + verify

AI Prompt Injection Scanner (bibguard scan)

Detects adversarial text hidden in PDFs that targets LLM-based review systems. 5-layer detection:

Layer	What it catches
Visible text	Instruction hijacking, role injection, score manipulation, weakness suppression, jailbreak patterns
Hidden text	White-on-white text, micro font (<1pt) rendering
Off-page text	Content positioned outside visible page area
Metadata	Prompts embedded in PDF title/subject/keywords
Invisible Unicode	Zero-width character clusters carrying hidden payloads

bibguard scan paper.pdf              # standalone scan
bibguard scan paper.pdf --json       # machine-readable output

Auto-reject: In bibguard pdf mode, injection scan runs first — poisoned PDFs are blocked before reference extraction even begins.

Also included

• Feedback logging system (bibguard feedback)
• AI coding agent integration templates (Claude Code, Codex)

Install / Upgrade

pip install --upgrade bibguard[pdf]   # with PDF support
pip install --upgrade bibguard        # core only

Full Changelog: v0.3.1...v0.4.0