Skills 2026-09-05 15:32 UTC
designer-skills
Committed by Owl-Listener. Update Category: New Features & Capability Expansion. Commit details: feat: add the quantitative half of design-research, with evals (#50). This update introduces new capabilities and expands supported use cases.
CommitRaw commit message
feat: add the quantitative half of design-research, with evals (#50) design-research was 12 skills and entirely qualitative. The adjacent skills each sit one step away from the problem: metrics-definition chooses what to measure, a-b-test-design creates the number, design-impact-reporting presents it, data-visualization encodes it. None covered being handed a dashboard you did not instrument and asked why people drop off. behavioural-analytics — reading funnels, retention curves and event data. Rules out the five things that make a surprising number wrong before forming any design hypothesis; reads shape rather than value, since a cliff means something blocks and a slope means the flow is merely long; and writes hypotheses that forbid something, ranked by how cheaply each can be killed. qual-quant-triangulation — what to do when the dashboard and the interviews disagree. A taxonomy of what each kind of disagreement means, which source can answer which question (a why-question asked of a dashboard is a category error, not a conflict), and how to design a resolving study narrower than either original. Both ship with EVALS.md: three realistic inputs each, what a good output must contain, what a failing output looks like, and why the case exists. Cases are chosen so a plausible but shallow answer fails at least one check — a funnel step users are not forced through, a slope mistaken for a cliff, a bimodal distribution read as a mean, a proxy metric that moved while the experience got worse, stated preference mistaken for revealed. This follows /start-here passing all seven CI checks while being wrong on three of five realistic inputs. Structural validation is not evidence of quality, so these are the first skills to carry their own. No new eval skills: ai-design-skills already ships a seven-skill evaluation plugin, and duplicating it here would recreate the overlapping-claims defect fixed in #47. A third candidate, reading-significance, was dropped because a-b-test-design already carries sample sizing. Boundaries: each description names its near-neighbours across three plugins, and INDEX.md gains one Start-here entry and five Frequently-confused rows. The four places design-research is described are updated so the marketplace no longer advertises a purely qualitative plugin. Closes #49