LLM agent · feed a brief, watch every tool call gated

Brief

Sample briefs · each exercises a different gating path
BriefPOST /v1/agents/run
Agent identity
Briefnatural language
Max steps
How it works: the agent gets the brief + its tool catalog, calls Gemini (or the stub planner if no key), and every tool it picks is gated by Sentinel before the mock result is fed back. Each gated call writes an audit receipt + cost event.
Trace
no run yet — submit a brief on the left