← ClaudeAtlas

show-dont-telllisted

Turn data-shaped answers into designed, self-contained inline UI instead of walls of text. Renders comparisons as comparison matrices, plans as timelines, numbers as stat tiles, and processes as step-flows — as theme-aware, responsive HTML/SVG that any capable chat renders. Governed by a strict anti-slop design doctrine. Triggers when the user says compare, dashboard, plan, breakdown, visualize, rank, "which should I pick", or when an answer is data-shaped enough that a visual reads better than prose.
turtk5799-afk/show-dont-tell · ★ 0 · AI & Automation · score 70
Install: claude install-skill turtk5799-afk/show-dont-tell
# Show, Don't Tell MCP servers give chat more tools. This gives it taste. Most answers should stay text. But some answers are **data-shaped** — a comparison, a plan, a set of numbers, a ranking, a process — and for those, a designed visual reads better than a paragraph. This skill decides *when* that's true and *how* to render it well, in whatever chat you're using. Two things make this work and not annoy: a strict trigger (**don't render everything**) and a strict aesthetic (**don't render slop**). Both are below. Treat them as hard rules, not suggestions. --- ## Rule 0 — When NOT to render (read this first) Rendering everything gets a skill uninstalled in a day. **Default to text.** Only produce a visual when the answer passes the **table test** — it contains at least one of: - **≥3 comparable items** (products, tools, plans, options) - **≥3 meaningful numbers** (metrics, stats, prices, scores) - **a sequence** — steps, a timeline, a schedule, a process - **a decision with trade-offs** — "which should I pick?" - an **explicit ask**: compare, dashboard, plan, breakdown, visualize, rank, chart, schedule, "show me". **Stay plain text** for: prose explanations, code help, short factual answers, ongoing back-and-forth, and **anything sensitive** — health, grief, conflict, money troubles, anything where a reader could be personally affected. Never decorate those. Three hard limits, always: 1. **One visual per answer.** Never stack two. 2. **Always pair the visual with