Give Claude a Check It Can Run

Claude says it's done. Is it?

Without a way to find out, Claude can only judge by how the work looks. Then you become the one who checks every answer.

What is a check?

A check is anything that tells Claude pass or fail.

The guide to Claude Code from Anthropic, the company that makes Claude, puts it this way:

"Give Claude a check it can run: tests, a build, a screenshot to compare. It's the difference between a session you watch and one you walk away from."

Tests are small checks the computer runs for you.

A build is the step that turns your files into the finished thing.

A session is one conversation with Claude.

In plain words: Claude runs the check, reads the result, and keeps fixing until it passes.

What should I do first?

Save a copy of your project first.

Keep Claude's permission checks on. They are Claude's safety step before it changes or runs things. Don't switch them off to go faster.

Why does it help with a learning game?

A quiz or scenario has many small things that can go wrong. One answer scores wrong. One screen shows the wrong feedback. A button does nothing.

You could click through all of it yourself, every time Claude changes something. Or you can ask Claude to do the clicking and show you what it found.

What can I copy?

Here is a prompt of mine you can copy. Say it in your first message about an onboarding quiz:

"Before you tell me it's done, write a check that plays through every answer and shows me the score each time."

Then read the scores it shows you. Do they match what you expect for each answer?

No scores? Ask Claude what stopped it.

Then play the result yourself once. A check Claude wrote can be wrong too.

What's next?

Pick one thing in your game that would embarrass you if it broke. Ask Claude for a check on that first.

Sources

• Anthropic, "Best practices for Claude Code," section "Give Claude a way to verify its work": code.claude.com/docs/en/best-practices. Checked October 3, 2026.