Hold on
You are not looping?
EVERYONE ELSE IS LOOPING
Your agents are sitting there doing one thing at a time. One. Thing. While you watch. Reading every result yourself like some kind of person.
Paste this at your agent.
It writes the loop. You did nothing. Beautiful.
I am about to do the same thing over and over.
Do not start yet. Turn it into a loop first.
1. THE LIST
Write the command that lists every item. A real
command, not a guess. Show me that it works.
2. ONE AT A TIME
Do one item per run, fresh each time.
3. A CHECK THAT CAN FAIL
Write something that says pass or fail on its own,
without asking me and without asking you. If you
cannot think of one, say so instead of pretending.
4. WHEN TO STOP
Three tries per item. Then put it in a quarantine
file with the reason and move on. Never save an
item that failed.
Show me the list and the check before you touch anything.
Here is the job:
Then say what the boring thing is. That is the whole move.
A one minute silent screen recording. Someone asks their AI to add meta descriptions to all 80 pages of a site. It replies that it is done. It is not: one page has no description, one says "Welcome to our website", and one describes a different page. The same request is then made with one extra sentence asking it to run a checker on each page and fix what the checker reports. This time failures are caught with the reason, retried, and the one page that fails three times is set aside in a quarantine file. It finishes with 79 fixed, 1 flagged, and the words "Not done".
A loop needs four things
Questions
Is this free?
Yes. MIT licensed. No account, no email, no plan, no "book a demo".
Does it work outside Claude Code?
Any AI coding tool that reads Agent Skills. Check your tool’s docs for the Agent Skills format.
Does it only do code?
No. Writing, spreadsheets, files, research, tagging. Anything you are about to do a hundred times.
Ten loops worth stealing are on the loops page. Longer thinking is on the blog.