Semantic intent evaluation for agentic systems. Detects instruction-layer risks including authority laundering, consent bypass, concealed execution, and semantic attack patterns in SKILL.md and agent configuration files.
mcp cybersecurity semantic-analysis ai-security ai-alignment prompt-injection llm-security prompt-security agentic-ai semantic-intent semantic-execution skillmd invariant-evaluation
-
Updated
May 7, 2026 - Python