AGENT SKILL

prototype

Build disposable evidence for unresolved logic, interaction, or visual questions. Use when runnable evidence would resolve behavior or UI uncertainty before commitment.

License 许可
MIT
最近更新
2026-11-12
Stars
★ 4
Forks
1

01 前言 Front Matter

name
prototype
description
Build disposable evidence for unresolved logic, interaction, or visual questions. Use when runnable evidence would resolve behavior or UI uncertainty before commitment.

02 资源文件 2

03 SKILL.md 正文

For product/design forks, read Poteto.

Prove one named question, not an early production branch.

Choose cheapest faithful shape: logic/state executable model (LOGIC.md); UI several meaningful variants (UI.md); API/tool minimal sandbox/fixture call.

Use .context/prototypes/<question>/, or target-adjacent only when real runtime must load it; mark in-tree artifacts.

Retention

Keep runnable prototypes as primary sources; never merge prototype-only code to main.

  • When the requested endpoint authorizes commits, use isolated prototype/<name> plus an issue/decision context pointer.
  • Otherwise keep under .context/prototypes/<question>/, report path, and move/copy in-tree artifacts there before cleaning the ship diff. Do not delete.

Record the question, evidence, and verdict in an issue, ADR, implementation notes or commit. Main: only validated production decisions.

Constraints

Explicit interface workbenches: recommend variant (designs), state-machine (reachable states), or break (stress). Apply retention; user-only, never automatic dependencies.

  1. Standard library/existing deps; no unrelated scaffolding.
  2. One run command; scratch/in-memory persistence.
  3. Show relevant state/observations.
  4. Before trusting verdict, /dogfood once through decisive path and likely boundary, not every edit.
  5. Apply retention after the answer.

If evidence contradicts the plan, revisit that decision before production.

04 同仓库更多 Skills