CoursesFree Quizzes

Loop Engineering Quiz: Can You Run Claude Code on Repeat?

Twelve scenario questions about running Claude Code on repeat. Each one describes a loop run that worked, stalled, or did something you did not ask for, and asks which choice works. Every answer explains why the other three choices fail, with facts checked against the Claude Code docs.

About the Loop Engineering skill quiz

This Loop Engineering quiz is a Practical fundamentals check. It covers agent loops, task files, checkable tasks, CLAUDE.md, /goal, /loop, evaluator evidence, stuck loops, handoffs, headless claude -p, permissions. The quiz has 12 questions, takes about 8 minutes, and a score around 8 out of 12 means you have a workable baseline.

The 12 questions on this page are built for developers who have used Claude Code for a week and want it to keep working after they walk away. They focus on practical decisions you would make while building, debugging, querying, or operating real projects.

Related Loop Engineering Practice Questions

What Is An Agent Loop Question Explained
Loop Basics
Run Claude Code Overnight Question Explained
Loop Basics
Task File For A Loop Question Explained
Loop Basics
Write A Checkable Task Question Explained
Checkable Tasks
Repeated Instructions Question Explained
Checkable Tasks
Goal Evaluator Evidence Question Explained
Checkable Tasks

Frequently asked questions about the Loop Engineering skill quiz

What is loop engineering?+

Loop engineering is the practice of running an AI coding agent on repeat with a trigger, an attempt, a fresh check of the result, a rule that decides whether to continue, stop, or ask a person, and limits that end the run. The agent makes each attempt. The check and the stop rule decide whether another attempt runs.

What is the difference between /goal and /loop in Claude Code?+

/goal sets a completion condition. Claude keeps working toward it, and a separate evaluator model checks the condition after each turn using only what appears in the conversation. /loop runs a prompt on a schedule while the session stays open and idle. Use /goal to finish a defined job now and /loop to check back for new work.

Can I run Claude Code unattended overnight with /loop?+

Only while Claude Code keeps running on an awake machine. Scheduled tasks fire only while the session is running and idle, so closing the terminal or letting the laptop sleep stops them. Missed fires do not catch up, and recurring tasks expire seven days after creation. For runs that do not depend on an open session, the docs point to cloud scheduled tasks, Desktop scheduled tasks, or GitHub Actions. A system cron job that calls claude -p also works, but then you supply the checks and limits yourself.

Why did /goal say my goal was met when the tests fail?+

The evaluator judges the condition against what Claude surfaced in the conversation. It does not run commands or read files itself. If Claude wrote "tests should pass" without showing test output, that sentence was the only evidence. Write goals so the check output itself must appear in the conversation.

Does --allowedTools keep a loop from touching other files?+

No. --allowedTools lists tools that run without a permission prompt. It removes questions; it does not restrict access. To limit what a loop can reach, run it in a dedicated folder, use deny rules or --tools, and use OS-level sandboxing or a disposable environment.

Is this quiz the same as the Loop Engineering course?+

No. The quiz checks whether you understand the ideas. The Loop Engineering course on learn.cloudyeti.io is ten lessons where you build the loop yourself: a task file, CLAUDE.md, a checks file, /goal and /loop runs, a stuck-loop prompt, a handoff note, a permission review, and a claude -p report. The quiz is free; the course is paid.

Sample Loop Engineering questions

Three examples from the 12-question quiz. Answers stay hidden until you reveal them.

Sample Question 1Limits and Safety

Your overnight loop tried the same task fourteen times. Each run edited the same line, the check failed the same way, and the run ended. What should you change before restarting?

AShorten the interval so the fix lands sooner
BRemove the check so the task can be marked done and the queue moves on to the next one
CAdd a stop rule for repeated failure, then inspect the task before restarting
DSwitch to a larger model so the fourteenth attempt succeeds
Show answer
Answer: C — Add a stop rule for repeated failure, then inspect the task before restarting

Fourteen identical failures mean the task or the environment is wrong, and another try will fail the same way. Give the loop a no-progress rule, such as an attempt cap or a repeated-failure rule, that ends the loop and hands the problem to a person. Inside a /loop prompt, that rule is an instruction to Claude, not an enforced limit. For claude -p runs, Claude Code enforces the --max-turns and --max-budget-usd flags, though spend can pass the budget cap. Then read what failed and fix the task before you start again.

Sample Question 2Limits and Safety

It is 6 pm. A loop has finished three of seven tasks and the session context is getting long. Tomorrow you want a new session to continue without redoing work. What do you do tonight?

AMark all seven tasks done so tomorrow starts with a clean file
BCancel the loop, write a short handoff note, and leave unfinished tasks unchecked
CCopy the whole conversation into a text file for the new session to read tomorrow morning
DLeave the session running and the laptop open so nothing is lost
Show answer
Answer: B — Cancel the loop, write a short handoff note, and leave unfinished tasks unchecked

The next session only knows what is in files. Cancel the loop, then have Claude write a short note with what changed, what was checked, and the next task. Leave the four unfinished tasks unchecked so the file itself says where to start. Tomorrow a new session reads CLAUDE.md, tasks.md, and the note, and you confirm its summary before restarting the loop.

Sample Question 3Limits and Safety

You plan to run a loop unattended on a laptop that also holds client repositories and personal documents. Which setup actually limits what the loop can touch?

APass every tool you use to --allowedTools so Claude never has to ask
BRun it in a dedicated folder with deny rules or an OS-level sandbox
CTell Claude in the prompt to edit only index.html and tasks.md
DUse bypass permissions mode so a blocked permission prompt cannot leave the loop waiting
Show answer
Answer: B — Run it in a dedicated folder with deny rules or an OS-level sandbox

By default Claude has access to the directory where you launched it, so a dedicated practice folder is the first boundary. The CLI reference says --allowedTools lists tools that run without prompting. To restrict tools you use --tools and deny rules. The permissions docs describe OS-level sandboxing for shell commands. A file list in the prompt states your intent, but it does not block access to other files.

Try another tech quiz

Compare your skills across Git, AWS, Linux, SQL, Python, Docker, harness engineering, and loop engineering.

See all free quizzes