← Back to Module 0 - Watch this first

The walkthrough

The exact process, working end to end.

1 min read

This is the whole loop. You write a task. A context agent scores it and asks you questions until it clears 85. An execution agent then builds it and opens a pull request, and you review it. Nothing ships without you merging it. Expect at most one PR an hour, and 30 to 90 minutes from writing a task to its PR, because the two agents fire half an hour apart.

How the loop worksA task is read by the context agent and given a readiness score. Below 85 it writes questions back to the task and waits for answers, then the task is scored again. At 85 or above the execution agent writes the diff and opens a pull request. A human reviews it and either merges it or sends it back to the queue.NoYesTask in the queueSmall enough for one PRContext agentReads the task and codeReadiness scoreRubric, 0 to 100Gate: score 85+?Checked each hourly fireQuestionsWaits for answersExecution agentWrites diff, runs checksPull requestWhat changed and whyHuman reviewMerge or send backContext halfExecution halfWaiting on you

You don't have to read the rest of this course page by page. The install runs from inside Claude, and once it's in, your Claude can read the course for you. Ask it anything about a lesson and talk it through there.

Once you own the course, go straight to Module 1: it's the install. You paste one prompt into Claude Code and Claude installs the loop with you.

Buy The Loop