Interactive · no signup · nothing stored
The Policy Playground
Paste an agent's action log — or pick a preset below — and run it through the real policy engine (allowlist → denylist → spend caps → risk tiers → kill switch, fail-closed). You get a live allow / deny / require_approval verdict with the reason, exactly as the production proxy would decide.
1 · Pick a scenario or write your own
2 · Set the policy
EVALUATED AS PRO
What just happened
Every action was decided by the production policy core, in order: kill switch → denylist → spend caps (fail-closed when a call would breach) → allowlist (unknown actions need a human on Pro, denied on Free) → risk tiers (dangerous verbs on high-risk agents always need a human). The engine is fail-closed: anything ambiguous is denied, never allowed. This is the same code path every POST /v1/proxy call hits in production.
FAQ
Is this the real policy engine or a mock?
The real one. Your log is evaluated server-side by the exact fail-closed function the production proxy uses. Nothing is stored — the log lives only in the request.
What do the verdicts mean?
ALLOW runs. DENY is blocked — the reason names the exact rule. REQUIRE_APPROVAL is held for a human in the approval queue (dashboard or signed email link, no login).
Why is this evaluated as Pro?
So you can see REQUIRE_APPROVAL verdicts. On Free, unknown and high-risk actions are denied outright instead of queued — the engine behaving honestly at each plan level.