Test the mandate before the agent ships it.
Drop in any AP2 mandate this showcase emits — AML,
BaaS,
DORA gap,
instant-pay fraud,
stablecoin readiness — queue a
candidate agent action, and watch the runtime traverse the mandate's rules. The trace shows
which rule fired, why, what state followed, and the final allow / step-up / refer / refuse
decision. The artifact you keep is the trace, exported as AP2 against
@ainumbers.co/sandbox-trace-v1.
EU AI Act Article 14 (human oversight) and Article 15 (accuracy, robustness, cybersecurity) apply to high-risk AI systems in financial services. DORA Articles 25–26 require resilience testing of ICT systems including threat-led penetration testing for in-scope entities. The Financial Stability Board (FSB) AI principles (Nov 2024) explicitly call out pre-deployment testing of agentic FS AI. The mandate is the contract; the sandbox is the test rig that proves the contract holds before production.
Load a mandate
Pick a bundled template (each is a representative mandate this showcase emits) or paste your own AP2 JSON. The sandbox parses the schema, runs the action against it, and emits a trace.
…or paste your own AP2 mandate JSON
Queue a candidate action
Actions are scoped to the loaded mandate's schema. The default parameters trigger different evaluation branches; tweak and re-run to explore the policy lattice.
What the runtime did
Run the sandbox to see the per-step trace.
— run sandbox to populate —
Compose with
This demo closes the Compose → Test → Apply arc on the Agentic Runtime hub. Compose with #7 — AP2 Visual Guardrail (visualise the same mandate as agent-traversal guardrails) and #23 / #24 (compose the mandates this sandbox tests). Position in the 259-tool atlas: Tool Chain Composer, RBE-06 + T102 in the Agentic Runtime cluster.