AGENT SKILL

prototype

Build a small disposable artifact to answer a specific UI, logic, data, or feasibility question.

License 许可
MIT
最近更新
2026-09-16
Stars
★ 0
Forks
0

01 前言 Front Matter

name
prototype
description
Build a small disposable artifact to answer a specific UI, logic, data, or feasibility question.

02 资源文件 4

03 SKILL.md 正文

Prototype a question

State the question, what observation would answer it, and the available budget. Choose the cheapest format that exposes the uncertainty. Read only the matching reference; the formats below are options, not a mandatory sequence.

  • UI alternatives: UI.md when comparing visual arrangements.
  • Interactive state transitions: LOGIC.md when someone needs to drive a state model by hand. HTML is useful here, not required for all logic.
  • Data, numerical logic, model feasibility: EXPERIMENT.md for a script, notebook, benchmark, or small data sample. Call the Skill tool with "ml-experiment" when the question is a model-quality comparison requiring a valid protocol.

Keep the artifact isolated and clearly provisional. Use the project's existing runtime and conventions. Restrict persistence and external mutations to what the question actually requires and the user authorized. Verify that the artifact runs and that its output can answer the question; numerical checks are warranted when an incorrect calculation would invalidate the result.

Stop when the question is answered, the budget is exhausted, or a named blocker prevents the observation. Report the answer, its evidence and limits, and the artifact path or reproduction command. Retain it according to the caller's request and repository conventions; do not create a branch or commit by default. Promotion into production is a separate implementation task with appropriate TDD and review, not an automatic consequence of a successful prototype.

04 同仓库更多 Skills