One engine.
Six surfaces.

Intent analysis is a horizontal capability. MCP, A2A and Skills are our beachhead.
The same reasoning engine extends to every permission-based attack surface.

Intent Analysis

LLM-graded assessment of tool declarations. Manipulation, injection, abuse patterns flagged with reasoning trace.

Composition Analysis

Cross-server capability chains. Identifies emergent 1+1=3 risks. The attack paths each tool hides on its own.

Red Team Sandbox

Adversarial AI agents in ephemeral K8s jobs. Honeypod evidence capture. Proves exploitability with a signed attack log.

25 Security Policies

Plain-English rules with two-phase evaluation. Auto-maps to OWASP Agentic AI, LLM Top 10, MCP Cheat Sheet.

Continuous Monitoring

Background drift detection. Ed25519-signed snapshots. SmellDex cross-references 1,500+ known AI tools every poll.

CI/CD Gate

mcp-gate with configurable risk thresholds. Blocks risky tool changes before they reach production.