sentinel-seed/sentinel: v2.25.0 - Multi-turn Conversation Support

Memory Shield v2.0 Pre-signing content validation for memory entries, addressing Princeton CrAIBench attack vectors where malicious content is injected before memory protection is applied. Highlights MemoryContentValidator: Detects injection attacks BEFORE content is signed 23+ detection patterns across 9 injection categories < 1ms latency per validation < 5% false positive rate with benign context detection > 90% detection rate for known attack patterns Injection Categories | Category | Severity | |----------|----------| | Authority Claim | High | | Instruction Override | Critical | | Address Redirection | Critical | | Airdrop Scam | High | | Urgency Manipulation | Medium | | Trust Exploitation | Medium | | Role Manipulation | High | | Context Poisoning | High | | Crypto Attack | Critical | Integration Support Letta: create_memory_guard_tool(validate_content=True) Virtuals: SentinelConfig(memory_content_validation=True) Solana Agent Kit: SentinelValidator(memory_content_validation=True) Installation pip install sentinelseed==2.26.0 Links PyPI Documentation Memory Shield v2.0 Spec References Princeton CrAIBench: https://arxiv.org/abs/2503.16248 OWASP ASI06: Memory and Context Poisoning