2 projects
pyaigis
Open-source trust layer for adopting Claude Code and autonomous AI agents at work: deterministic guardrails on every tool call, tamper-evident audit logs, and a generated IT-approval pack (aigis trust-pack). Zero-dependency Python firewall — prompt injection, MCP tool poisoning, memory defense; 44 compliance templates across US/CN/JP/EU. Library, Docker sidecar, or CLI.
aig-guardian
AI agent security with provable guarantees: capability-based access control (CaMeL-inspired), atomic execution pipelines, and safety specification verification. 165+ patterns, 25 threat categories, OWASP LLM Top 10 + MITRE ATLAS. Zero-dependency core.
