Prompt Shield - Dify Plugin
A Dify tool plugin that provides local prompt injection detection, output scanning, and PII redaction powered by prompt-shield (v0.4.0+).
27 input detectors including novel cross-domain techniques (Smith-Waterman sequence alignment, stylometric discontinuity detection), 6 output scanners, 10-language support. Research paper: arXiv:2604.18248.
Tools
Prompt Injection Scan (scan)
Scans text for prompt injection attacks using 27 detection strategies including pattern matching, encoding bypass detection (Base64, ROT13, homoglyphs, zero-width), role hijacking, semantic analysis, Smith-Waterman sequence alignment, and stylometric discontinuity detection. Returns a risk score (0.0-1.0), a recommended action (pass, log, flag, or block), and details about detected threats.
PII Scan (pii_scan)
Detects personally identifiable information in text, including email addresses, phone numbers, SSNs, credit card numbers, IP addresses, and more. Returns entity types and counts without modifying the text.
PII Redact (pii_redact)
Redacts PII from text by replacing sensitive data with entity-type-aware placeholders (e.g., [EMAIL_REDACTED], [PHONE_REDACTED]). Returns the redacted text along with counts of what was redacted.
Output Scan (output_scan)
Scans LLM output for harmful content including toxicity, code injection, system prompt leakage, PII exposure, and jailbreak persona adoption. Returns whether the output was flagged, along with details about each flag raised.
Installation
-
Ensure prompt-shield-ai is installed in your Dify plugin runtime environment:
pip install prompt-shield-ai>=0.4.0
-
Package and install the plugin into your Dify instance following the Dify plugin development guide.
Configuration
No API keys or credentials are required. All processing runs locally using the prompt-shield library.
Privacy
All text analysis is performed locally. No data is sent to external services. See PRIVACY.md for details.