Video: Trust It: Verifying Unsupervised Runs

1. Video: Trust It: Verifying Unsupervised Runs

You handed Claude a task and let it run without watching every step. Now it says it's done, but before you ship that, you need a way to check work that you didn't even supervise. A check is what makes hands-off Claude Code safe to rely on. Verify in proportion to the rope that you gave the run. A short session where you watched the messages scroll by needs a glance. An unattended run or a job that fired in continuous integration with nobody in the loop actually needs a real check, because no one saw it happen. And when a run does go unattended at work, keep it in auto mode rather than bypass permissions. The classifier still reviews each action for danger. It never judges whether the code is right, though, so the verification bar stays where it was. Set that bar from how unsupervised this run was. So start with the diff itself, rather than Claude's summary of it. Run /code-review to walk the changes and flag issues. Then put your eyes on git diff. The trap is a tidy summary that reads fine while the diff touched a file that you honestly didn't even expect it to. Read what changed and read the files that were in the plan first. The real gate on an unsupervised run is whether the test passed and whether Claude ran them or only claimed that it did. Don't leave that to trust. Wire it as a hook so Claude can't skip it. A stop hook that runs your tests and refuses to end the turn on a failure, or a PostToolUse hook that lints and type checks after every edit. A hook that exists with code 2 feeds the failure straight back to Claude, which reads it and fixes it without you asking. The check fires on every run, whether or not you remember to ask. The subagent code review you'd run before a PR works here, too. Point it at an unsupervised run. Open a fresh session or subagent and have it review the change cold, with no memory of how the code was built. It has no stake in the approach and catches what the original run talked itself past. Make the check as serious as the run was unsupervised. Read the diff, turn the test into a hook that gates the turn, verify headless runs by their JSON result and exit code, and get a cold second opinion on anything that matters. Do that and Claude did it while I wasn't looking. No longer takes faith.

2. Let's practice!

Create Your Free Account

By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.