Things will break. The skill isn't avoiding it — it's responding without making it worse.
"It doesn't work" makes the agent guess. Paste the actual material: the full error message, the stack trace, the failing test output, the console log. The difference in outcome is dramatic — the error text contains the file, the line, and usually the cause.
Via the terminal / VS Code: copy the output directly. Via the Desktop app: paste the log into the chat, or point the agent at the log file if it's in the project.
The agent asks to scan folders outside the project — a harmless command, but exactly why you read every one before approving. Answered "No" and gave it the real error instead.
One pasted stack trace: from guessing to the exact line in six seconds. The error text contained the file, the cause, and the fix.
A healthy debugging session converges: each attempt narrows the problem. A looping session cycles: the agent tries A, then B, then A again with confidence. Signs you're in a loop:
Three strikes rule: if three attempts haven't converged, stop prompting "try again." Change something structural instead: roll back, start a fresh session (clean context — Week 1), or narrow the task.
This is why Week 2 insisted on branches and small commits: rolling back is cheap. When the working state is gone and the agent is digging deeper, revert to the last good commit and re-approach with a smaller step. Ask the agent itself: "revert everything after the last commit" — and read the commands it proposes before approving.
A fresh session on a clean state, with a better-phrased task (see the prompting page), routinely fixes in ten minutes what a looping session couldn't in an hour. That's not the agent getting smarter — that's you removing the noise it was drowning in.
The HackYourFuture curriculum is licensed under CC BY-NC-SA 4.0 *https://hackyourfuture.net/*
Built with ❤️ by the HackYourFuture community · Thank you, contributors
Found a mistake or have a suggestion? Let us know in the feedback form.