What changed
Once GitHub is connected, Write my checklist reads the app's repositories (or folders on your computer, read in the browser) and writes checks for each feature area: both sides of every feature, the unhappy paths and concrete expected results. It can ask up to six questions about what the app is meant to do, adds a check for each fixed bug so it can't quietly come back, and extends the checks you already have instead of writing them twice. Nothing reaches the board until you've reviewed it.
Every app has a test inbox: fresh addresses for sign-ups, with the code and link picked out of each email, kept for 7 days. Test configuration also keeps test accounts, the logins testers use when a check needs an existing account, each with a note on when to use it. Passwords are sealed and never shown back.
Your own AI assistant (Claude Code, Cursor, Codex) can test the app over MCP: it opens the app on a connected computer, sees each screen and decides what to do, and the computer taps and types. No AI credit is used. Saved passwords are typed on the device, so they never pass through the assistant.
Why it matters
No AI is assumed any more. Running with AI and writing a checklist first ask who does the work: qarunbook's AI, your own key, or your assistant. On qarunbook's AI, work is paid from AI credit, topped up by card, at what the AI actually cost plus a set markup, never a flat guess. A check it couldn't settle stays untested and isn't charged.
Your own keys live in Profile → AI keys: a testing key and a Claude key for checklists and imports. Each app chooses which to use.
Who it's for
Teams starting from a codebase with no test plan, teams whose app signs people up by email, and developers who'd rather their own assistant did the testing.
Apps with an API platform are now tested with requests, written in each check's steps, and checks can be made from an OpenAPI document, uploaded as a file. Only endpoints the document actually describes become checks; the rest are named, never guessed.
How to try it
Connect GitHub to an app and choose Write my checklist. The test inbox and test accounts are in App configuration → Test configuration.
What shipped
Write your test checklist from your code
Pick the app's repositories or folders on your computer, and the AI writes checks for each feature area, adds what the existing checks don't cover, and puts each check on the platforms the code builds. You review everything before it lands.
- GitHub
A test inbox and test accounts for every app
Fresh email addresses for sign-up and reset checks, with codes and links picked out of each email, and saved logins with a note on when to use each. Passwords are sealed and never shown back.
Test with your own AI assistant
Claude Code, Cursor or Codex opens the app on a connected computer over MCP, sees each screen and decides what to do. No AI credit is used, and a computer can hold up to four assistant sessions at once.
Every change
New14
- Write an app's test checklist from its code: pick repositories or folders on your computer, answer a few questions about the app, and review the checks before they're imported.
- A test inbox for every app: fresh addresses, the code and link picked out of each email, kept 7 days. Assistants read it over MCP.
- Test accounts in Test configuration: the logins testers use when a check needs an existing account, with when to use each. Passwords are sealed and never shown back.
- Testing with your own AI assistant over MCP (start_device, device_action, end_device), on a browser, Android emulator or phone, or iPhone simulator on a connected computer.
- Sign up and sign in with GitHub. A new workspace is connected to GitHub in the same step.
- API testing for apps with an API platform: requests and expectations written in each check's steps, and checks from an OpenAPI document, which can be uploaded as a file.
- Retest fixes automatically: when it's on, a fixed issue's check is retested on each set-up platform, and the people on the issue hear how it went.
- Delete an issue, with its files, if you raised it or are an admin of the app. Also over MCP.
- Reset asks what to reset: everything, or only the platforms and sections you tick.
- AI credit: top up by card, and pay for AI work at what it actually cost plus a set markup. The balance is always in the sidebar.
- Your own AI keys in Profile → AI keys, chosen per app.
- A folder on your computer as the source for Android and iPhone simulator apps: it's built before every run, so each run tests the latest code.
- Bugs that arrive from Sentry, PostHog or email are placed on the check they plainly concern, which tells that section's owners.
- A regression testing checklist guide with a 15-check template, and a Testmo alternatives page.
Improved10
- Running with AI and writing a checklist ask first who does the work: qarunbook's AI, your own key, or your assistant.
- A workspace connects GitHub, GitLab, Linear, Jira, PostHog or Sentry once, and each new app offers that connection. GitHub settings say which account an app reads, with a way to switch.
- Test configuration is a settings screen with its own navigation, and the test inbox works like a mail client, with search.
- The run screen has a tab for each run going and a History of every run, with what it covered and what it cost. Waiting runs show amber and running ones green.
- Forms such as Add an app, Add a check and Raise an issue open as dialogs, and fields are easier to see in light and dark.
- Descriptions across the app are one short line, with the details one click away behind More.
- A refresh keeps you on the app, page and tab you were on.
- Testing apps carry a green checklist mark and live apps an amber bug, so the two read apart at a glance.
- Connected computers keep their helper up to date by themselves, and runs go to your own computer first when one is free.
- Emails reach the Outlook inbox instead of Junk.
Fixed5
- Connecting GitHub or Slack, email verification links and invite links no longer land on a localhost address.
- A run cut off by a restart is picked up again from the next unfinished check.
- Checks running side by side read only their own mail, so they never use each other's codes.
- Imported checks join an existing section of the same name instead of making a second one.
- Billing and pricing no longer say a checklist from code costs a flat $10.