Prototype

Adapted from Matt Pocock's prototype skill at commit 3cca18b368ae95cdbdebbff572ccafa662551015 (MIT). Observable-uncertainty routing is adapted from Cursor's poteto-mode prototype playbook at commit 93b00b89ef425a9c1bac0d0b317dfc49c930ac99 (MIT, Copyright (c) 2026 Cursor).

A prototype is disposable code that answers one written question. When shape-design delegated a Spike, return the evidence and verdict to that workflow; shape-design owns the resulting production decision.

Separate observable uncertainty from user-owned decisions. Behavior, timing, layout, compatibility, output, and performance questions are observable uncertainty: answer them by building and running the probe, not by asking the user. Product intent, preferences, trade-offs the user owns, and irreversible decisions remain questions for the user.

Choose the artifact

• Logic/state: read references/logic.md, then make the smallest runnable harness that exercises hard cases and displays full state after each action.
• UI: read references/ui.md, then make several meaningfully different variants that can be compared from one route or artifact. Build multiple variants only when the alternatives are genuinely viable; a routine question needs one probe.
• Feasibility: build only the risky seam needed to answer whether the approach works.

If the question is ambiguous, ask. If the user is unavailable, infer from context and state the assumption prominently.

Constraints