What problem does it solve? Testing a LiveKit voice or chat agent before shipping is hard: ad-hoc manual calls miss edge cases, refusals, and regressions. This Skill reads the agent's actual code locally, builds a risk checklist from its constraints, and generates a diverse suite of simulation scenarios that can be run with lk agent simulate and reused as a regression suite. ## Core Features & Use Cases - Code-grounded scenario generation: Reads the deployed agent's code locally (nothing is uploaded) to produce a test-oriented description and a risk checklist covering constraints, guardrails, and adversarial cases. - Coverage-enforced assembly: The build_scenarios.py script validates authored scenarios and fails the build in strict mode if any risk-checklist item lacks a dedicated scenario. - User-steered testing: Captures what the user wants stress-tested (flows, refusals, edge cases) and biases scenario authoring toward that focus without dropping coverage. - Use Case: Before shipping a drive-thru ordering agent, generate ~10 scenarios covering unavailable items, withheld required inputs, and prompt-extraction attempts, then run them with lk agent simulate and re-run the saved suite after every prompt or model change. ## Quick Start Ask the assistant to generate simulation scenarios for your LiveKit agent and stress-test the flows you are most worried about.
