# Agentic Demo

Use the sibling `build-async-agentic-web-demo` skill for implementation. This reference defines how it participates in the web deliverable family.

An agentic demo is appropriate only when agents, asynchronous work, handoffs, tool use, or human approval are part of the concept being demonstrated. Do not turn an ordinary research explanation into a fictional multi-agent workflow.

The task contract must name:

- the user role and consequential decision;
- agents and their bounded responsibilities;
- case states and transitions;
- evidence or tool receipts shown at each handoff;
- the human review gate;
- requested observable effects;
- which behavior is simulated.

Evaluate state-machine integrity, human authority, provenance, replay stability, and truthful simulation labeling. Use research-quality scoring only for research claims contained inside the demo, not as a replacement for workflow evaluation.