feat(mist): Product Quiz

[sophiaerobinson]

What

Adds mist/product-quiz — a skill that builds an AI-driven product recommendation
quiz for the company it's run in: a Mist (Next.js) backend that generates the
questions and picks products from the live catalog, a native Liquid section that
renders it in the active theme, and the storefront Page that hosts it.

Why

Recommendation quizzes are a recurring ask and the naive build fails in
predictable ways — the model hallucinates products that aren't sellable, the
questions don't discriminate between anything in the catalog, and the quiz 500s
when the model is slow. This encodes the sequence that works: a deterministic
scorer generates a shortlist the model cannot escape, prices and URLs always come
from the catalog record, and every AI path falls back rather than failing.

Shape

Folder skill — the body is long enough to want its reference material factored out.

• mist/product-quiz/SKILL.md — the playbook
• references/backend-contract.md — endpoint payloads, catalog response shape, AI provider config, cache keys
• references/fluid-gotchas.md — Page/template and Mist-environment failure modes
• manifest.json entry with both references listed

Notes

• No hardcoded company values — catalog, theme tokens, and brand voice are all discovered at run time.
• Names the exact tool + endpoint for each step rather than deferring to the docs.
• scripts/validate_catalog.py passes.