focusgrouplisted

# focusgroup A skill for **persona-driven dynamic testing** of conversational AI products. The mental model: treat the system under test (SUT) the way a real focus group treats a prototype. Recruit diverse personas, give each one a real intention, watch them try to accomplish it, then debrief them in character and aggregate findings across runs. Static evals and unit tests catch logic bugs. focusgroup catches the things that only emerge when a stubborn novice, an impatient expert, and a probing adversary actually use the thing. ## When to use this skill **Use it when:** - The user has a conversational agent / chatbot / LLM-backed module to test - They want acceptance tests tied to product requirements, not just regression tests - They want to surface IX failures (confusion, frustration, abandonment) that judges-of-correctness miss - They want regression coverage that survives prompt and model changes **Do not use it for:** - Pure unit-testable logic (use pytest) - UI / browser automation (use Playwright) - Non-conversational systems (batch jobs, classifiers without dialogue, etc.) If the user's target isn't conversational, say so and stop. ## The four commands This skill orchestrates four operations. When the user invokes focusgroup, identify which command applies and follow the matching playbook below. | Command | When | Side effects | |---|---|---| | `init` | Once per project, before anything else | Creates `requirements.md` (or confirms existing), scaffolds `adap