Thought → Action → Observation
ReAct Agent Pattern
Bounded iteration loop with tool grounding — pick a scenario, watch the trace, read simulated metrics.
Run scenario
Metrics are simulated from the deterministic stub — not live LLM usage. Teach the pattern here; run open-ended work in VAP / OmniForge.
Steps
—
Tool calls
—
Stop reason
—
Latency (sim)
—
Run the system
This repo teaches one pattern. Compose it into a governed multi-agent OS or multimodal platform.