Claude Code
The agent’s claim.
“Done” is what the agent said. It is not a record of every check.
Your agent says it’s done. What actually ran?
See the files it changed, the checks Winskel saw, and what still needs checking.
Download for Mac
Use your existing eligible Claude Code or Codex subscription. Provider usage limits still apply.Current Beta · Free · Apple Silicon, macOS 12 or later
The checks Winskel saw.
Inspect the observed command and its result alongside the changed files.
The gaps that remain.
In this sample, reset tests passed. The browser flow and email delivery were not checked.
A few useful answers
Can I see what tests an AI coding agent ran?
Winskel shows commands and results it observed during the work. A claim that tests passed stays distinct from the recorded checks. Work without observed evidence remains not verified.
Do passing tests prove the app works?
No. Tests cover only what they check. The visible password-reset example has an observed npm test result, but browser behavior and email delivery remain unverified. It does not show an observed typecheck or lint run.
Is this a live debugging result?
No. This page shows a sample project to explain the interface. It does not run agents or project checks in your browser.
How is this different from another agent’s review?
An agent review can find issues in the code. Observed checks show what commands actually ran and their results. A review is not a substitute for running the relevant checks.