How to test prompts and projects on this site

What this site gives you (and what it does not)

Prompts and project pages are briefs: structured instructions you copy into an AI builder or editor. They are not magic buttons that always produce perfect apps by themselves.

Some items include a playable reference on this site (for example Sky Dash). That proves the idea can work. Your AI run may still differ — you must play-test what you get.

Pick a lane for the test

Choose one:

• Fun / low stakes — games (Sky Dash, Snake, memory). Best for learning the loop.
• Practical thin slice — landing page or habit tracker. Best for product practice.
• Serious path — anything with login, money, or other people’s data. Use the production checklist before calling it done.

Step-by-step test method

Use the same method every time so you can compare tools fairly.

• 1) Read the project or prompt once. Write your own “done when” in one sentence.
• 2) Open your tool (Lovable, Bolt, Grok Build, Cursor, etc.). Prefer a fresh project.
• 3) Paste the full prompt. Do not strip the constraints section — that is the safety rail.
• 4) Run once. Do not re-roll immediately if it is 70% right.
• 5) Play-test with a checklist (below). Note failures in plain language.
• 6) Send a short fix prompt: only the failures. No “make it amazing.”
• 7) Stop after 2–3 fix rounds or when credits get silly. Capture what worked.

Play-test checklist (games)

For Sky Dash–style games:

• Starts from a title screen without errors
• Input works: keyboard and tap/click
• Score increases; game over is reachable
• Restart works without a frozen state
• Best score survives a refresh (if promised)
• Usable on a phone-width window
• No console errors you cannot explain (open browser developer tools if you can)

Play-test checklist (apps)

For tools and SaaS-shaped projects:

• Happy path works end-to-end once
• Empty states exist (no blank confusion)
• Forms show errors for empty/invalid input
• Refresh does not destroy critical state without warning (or state is saved on purpose)
• No API keys visible in the browser page source
• If accounts exist: two users cannot see each other’s data

How to use the Sky Dash demo

Play the on-site demo first so you know what “good enough” feels like: /examples/sky-dash/

Then run the project prompt in your AI tool of choice. Compare: does your version flap, score, and restart? If not, fix those before polish.

The demo is a reference, not a rule that every AI output must match pixel-for-pixel.

Scoring a session honestly

Pass: you would show a friend without apologising every click.

Partial: core loop works; polish or edge cases missing — fine for learning.

Fail: blank screen, constant errors, or you do not understand a single part — revert and narrow the prompt.

Never call Partial “production.” That is how the 80% wall wins.

When the model fails

Try a smaller scope (“only the title screen and bird flap”). Switch tools if one environment keeps breaking. Read When to stop accepting diffs if the tool is thrashing your files.

Log what failed (tool name, date, one sentence). That notes file becomes your personal operator’s manual — which is the spirit of this whole site.