helio-creating-test
Installation
SKILL.md
You are helping the user design a Helio test that produces a signal worth using — one a designer, a PM, or a leader can carry into a Design Review and act on. Not just "ran a study."
Core idea
Every good Helio test runs the same arc — Hunch → Shape → Test → Signal — and each step constrains the next. Skip the hunch and you can't write the right questions. Pick the wrong shape and you collect the wrong metrics. Don't tie the signal back to a hunch and the result has nowhere to land.
The working parts:
- Four house shapes — marketing page eval (5 Qs), multi-screen flow eval (6–8 Qs), content prioritization (5 Qs with MaxDiff), concept comparison — with a decision rule for picking one. If two feel possible, pick the simpler.
- The hunch template — We believe that [change] for [audience] will [impact] because [reason]. Every question must tie to it; drop questions that don't.
- One metric per question, behavioral + attitudinal paired across the test.
- The followup pattern — required "why" after every Likert and multi-select; optional after click tests. The followups are where the design move lives.
- Validate before you spend — CLI dry-run, web preview, pre-launch checklist (≤ 8 questions, every question metric-tagged, hunch still answerable).
- Three ready-to-edit templates (A: marketing eval, B: multi-screen flow, C: content prioritization) lifted from real Zurb tests.
Surface rule: person → web app, script → CLI, assistant → MCP. Test design is the same either way.