Claude Code

How to Make Claude Test Its Own Work: The Verification Loop

Anthropic's Claude Code team has Claude check its own work before it says it's done. Here is how /verify works, and how to set it up once.

5min read
1prompts to copy
7sections

You ask Claude for a feature, it says it's finished, you try it and it's broken. Anthropic's Claude Code team has a name for the fix: a verification loop. Claude checks its own work, fixes what fails and checks again, before it tells you it's done. Here is how it works, and how to set it up once so it runs the same way every time.

Read Anthropic's post, Building verification loops in Claude Code with skills: https://claude.com/blog/building-verification-loops-in-claude-code-with-skills

Who is behind it

The post is by Delba de Oliveira, a member of the Claude Code team, published on July 22, 2026. Boris Cherny, who created Claude Code, works on the same team. They describe the pattern they use every day: /code-review hunts for bugs, /simplify cleans up the diff, /verify confirms the change actually works, and a custom /design skill checks the UI against their guidelines.

What a verification loop is

It is a repeating cycle:Claude makes a change, runs a check (tests, a linter, or a custom check of its own), and if something fails it fixes it and runs the check again. The difference from a normal session is who does the checking. Instead of you trying the app and reporting back, Claude tries it itself, and only reports done once the check passes.

Start with the built-in /verify

Claude Code ships three skills for this. They work with no setup: they work out how to launch your project from its type (a CLI, a server, a TUI, a web app) and from your README, package.json or Makefile.

SkillWhat it doesUse it when
/runLaunches and drives your app: it can click, interact and take screenshotsYou want to see a change working
/verifyBuilds and runs your app to confirm a change does what it should, against the running app, not only your testsYou want proof before you trust "done"
/run-skill-generatorGets your app running from a clean environment and records the recipe as a project skillLaunching needs a database, an env file, a graphical session or a multi-step build

The example from the video: ask for a new button on your website, and Claude opens the app, clicks the button itself and takes a screenshot as proof. If the click fails, it fixes the code, runs it again, and keeps going until it works.

Free, for your email

Get the full setup steps + the copy-paste setup prompt

The rest of this guide (4 more sections, 1 copy-paste prompt) opens right here. The setup prompt, one paste into Claude that sets it all up, comes by email only. Use your real address.

Unlocks: Set it up once · Turn your own check into a skill · Where each check should run · Before you rely on it

50,000+ follow where these guides come from. No spam, reply “unsubscribe” any time.

Read next