v0.3.0 — PII Detection & Redaction

What's New

PII Detection & Redaction

New d023_pii_detection detector and standalone PIIRedactor for detecting and redacting personally identifiable information before prompts reach the LLM. Mapped to OWASP LLM02 (Sensitive Information Disclosure).

• 6 entity types: email, phone, SSN, credit card, API key, IP address (16 regex patterns)
• Entity-type-aware redaction: [EMAIL_REDACTED], [SSN_REDACTED], [CREDIT_CARD_REDACTED], etc.
• CLI commands: prompt-shield pii scan and prompt-shield pii redact with --json-output support
• Per-entity configuration: enable/disable individual entity types and add custom patterns via YAML config
• AgentGuard integration: automatic PII redaction in the sanitize flow
• Standalone Python API: PIIRedactor usable directly without AgentGuard

New Files

File	Purpose
src/prompt_shield/pii/entity_types.py	EntityType enum, 16 default patterns, replacement strings
src/prompt_shield/pii/redactor.py	PIIRedactor class with overlap dedup and entity-aware placeholders
src/prompt_shield/detectors/d023_pii_detection.py	Regex-based PII detector, configurable via YAML
src/prompt_shield/models.py	Added RedactionResult model

Modified

• CLI: Added pii command group with scan and redact subcommands
• AgentGuard: Entity-type-aware redaction when d023 detections are present
• OWASP mapping: d023 → LLM02 (Sensitive Information Disclosure)
• Config: d023_pii_detection block with per-entity enable/disable

Tests

• 46 new tests (24 detector + 14 redactor + 8 CLI)
• Full suite: 482 tests passing