prototype
When a design question needs runnable evidence — choose a useful experimental surface, protect real data, and let observed behavior inform the decision.
SKILL.mdskills/workflow/prototype/SKILL.md
GitHubPrototype
Build the cheapest experiment that can answer the question for its recipient, not a compulsory TUI, web page or production scaffold.
- State the question, useful observations and allowed scope before coding. A logic/state question usually needs LOGIC.md; a visual/interaction comparison needs UI.md. Choose a script, terminal, offline artifact or native preview according to the question and recipient. Ask when that choice is consequential, rather than guessing from backend/frontend file location.
- Use existing language, tools and preview seams. Keep experimental code clearly disposable and isolated from production routing and mutations. Start with in-memory or approved fixture data; persistence experiments need an explicitly isolated data boundary and recovery path.
- Keep relevant state visible, scenarios resettable and results reproducible. Leave a small decision check for nontrivial logic. Throwaway status does not waive input validation, failure handling, accessibility or data protection.
- Present observed evidence separately from assumptions and unrun checks. A human preference remains the human's decision; if the recipient has not tried the artifact, do not invent their verdict.
Deliver: question, artifact/run instructions, scenarios, evidence and decision or remaining uncertainty in the task's existing working notes. Absorb only validated logic or the selected direction under production standards and authorization. Remove owned throwaway work only within cleanup authority; preserve user edits and records. No new decision diary.