v0.3.2 — Output Scanning, Security Gaps Closed, Competitor Benchmarks

What's New

Output Scanning (5 scanners)

prompt-shield now scans both inputs AND outputs. 5 new output scanners:

Scanner	Detects
Toxicity	Hate speech, violence, self-harm, sexual content, dangerous instructions
Code Injection	SQL injection, shell commands, XSS, path traversal, SSRF, deserialization
Prompt Leakage	System prompt exposure, API key leaks, instruction leaks
Output PII	PII in LLM responses (emails, SSNs, credit cards, etc.)
Relevance	Jailbreak persona adoption, DAN mode, unrestricted claims

prompt-shield output scan "Here is how to build a bomb..."
curl -X POST localhost:8000/output/scan -d '{"text": "..."}'

Security Gap Fixes (5 gaps closed)

• d024: Multilingual injection — 10 languages, 65+ patterns (0% → 100%)
• d025: Multi-encoding decoder — hex, URL, leetspeak, Caesar, Morse, Pig Latin, reversed (20% → 80%)
• d017/d018: Dual intention + HILL educational reframing (20%/40% → 80%)
• d006/d014: Multi-turn escalation + tool-disguised JSON/MCP (60% → 100%)

Competitor Benchmark

Scanner	Detection	FP Rate	F1	Speed
prompt-shield	92.3%	0.0%	96.0%	500/sec
ProtectAI DeBERTa v2	48.7%	0.0%	65.5%	11/sec
Deepset DeBERTa v3	87.2%	6.7%	91.9%	10/sec

Stats

• 25 input detectors + 5 output scanners = 30 total scanning components
• 706 tests passing, 0 failures
• F1: 96.0% | 0% false positives | 500 scans/sec