- Add Identity Truth and Honesty principles to SOUL mapping - Expand input sanitizer with audit logging and 7 new injection types: * Social engineering, researcher impersonation, context flooding * Token smuggling, multilanguage bypass, Unicode spoofing, hypothetical framing - Integrate input sanitization into run_agent.py message processing - Add pytest markers for conscience/soul/security tests Security hardening against prompt injection attacks (Issue #87)
4.2 KiB
4.2 KiB