DARINORCold Lab

Assessments / Agent Boundary Assessment

Agent Boundary Assessment

16 questions across trust boundaries, shared-state isolation, per-node capability scoping, and blast-radius containment. Answer one at a time — step back whenever you want to change an earlier answer. Nothing is uploaded.

Trust Boundaries1 / 16

Each agent in the graph has its own identity and credentials — no shared service account or shared API key across agents.

// About this tool

Agent Boundary Assessment

This assessment scores a single agent graph — not an org program — across four technical areas: trust boundaries between agents, shared-state isolation, per-node capability scoping, and blast-radius containment. Each area maps to a maturity tier: Ad hoc, Developing, Managed, or Optimized.

It complements the org-level Agentic AI Maturity Assessment. That one asks whether your program has identity, supply-chain, memory, and governance controls. This one asks whether a specific multi-agent deployment would survive one agent being compromised — the per-graph technical view.

Everything runs locally in your browser. Nothing is uploaded, and your answers reset the moment you leave the page. Any section that isn't yet Optimized gets a written recommendation on the results screen, worst-scoring first.

// When to use it

  • Score a deployment before it ships

    Walk through the questionnaire before wiring a multi-agent system to production to see which boundaries are missing before an incident finds them for you.

  • Prioritize hardening work

    Use the per-section breakdown to decide whether trust boundaries, shared-state isolation, capability scoping, or blast-radius containment is the highest-leverage next step.

  • Pair with the org-level assessment

    Run the Agentic AI Maturity Assessment for the program view, then this one for the graph view — together they cover policy and architecture.

// Questions

Is this a substitute for a real security audit?

No. It's a fast, self-reported baseline to find the biggest gaps in a multi-agent deployment — not a replacement for a penetration test or formal architecture review.

How is this different from the Agentic AI Maturity Assessment?

The maturity assessment scores an org program (identity, supply chain, memory, governance). This one scores a specific agent graph's technical boundaries — whether one compromised agent can reach the rest of the system.

Is my data uploaded?

No. All scoring runs locally in your browser — nothing is sent to a server, and nothing persists after you close the tab.

What do the maturity tiers mean?

Ad hoc (0-24) means boundaries are largely absent. Developing (25-49) means some controls exist but are inconsistent. Managed (50-74) means controls are in place and mostly consistent. Optimized (75-100) means controls are consistently applied and verified.

This site uses minimal cookies and local storage to keep features like the chat widget and games working. We do not use third-party tracking cookies. Privacy Policy