Your agent is waiting on a check that already failed

You hand a task to a coding agent, it opens a pull request, and it waits for the checks. One goes red. The agent keeps waiting, and so do you, until you happen to look.

Why it happens

An agent only acts when something prompts it. If it is waiting on something outside the session, such as a check running on GitHub, nothing tells it the wait is over. It will not notice a failure on its own, and it will not tell you, because from its side nothing has happened.

The same thing happens closer to home. A session stops on a failed push hook or a failed test run, writes a polite summary, and sits there. If you are in another window, that summary might as well not exist.

What you actually need

Something that watches from outside the agent and is impossible to miss:

• Failed checks on your open pull requests, noticed within minutes.
• A session that stopped on a failure and has been idle since.
• Anything your own scripts know has failed, such as a deploy or a nightly job.

Doing it yourself

If you use the GitHub command line tool, this lists the checks on a pull request:

gh pr checks 123 --watch

That works for one pull request in one terminal you are looking at. It does not help when you have several open, or when you have walked away.

What Cost Doctor does

It checks your open pull requests every few minutes using the GitHub command line tool you already have, and watches your sessions for ones that stopped on a failure. When something fails, the menu bar badge turns red, a red frame appears around every screen, and a notification is sent. Each alert can be dismissed, and the frame does not block clicks.

Any script can raise the same alert with one line:

echo "deploy failed" > ~/.costdoctor/alerts/deploy.txt