The Prompt FilesCase File № 04 · research
The Agent Cited as a Math Co-Author
Receipts on file
| Agent | OpenClaw |
|---|---|
| Category | research |
| Outcome | New mathematics record set; the agent formally cited as a contributor |
| Filed | September 9, 2026 · documented in a public report on Reddit |
The Job
A vibe coder set an OpenClaw agent loose on a mathematics problem and ended up with a new record — with the agent formally cited as a contributor on the result. The writeup credits a zero-tolerance verification loop plus a human maintainer checking the output as the reason the result held up under scrutiny.
The Prompt — try it yourself
Help me attack [open problem or computation]. Set up a verification loop where every claimed result must pass an independent check before we accept it, keep a log of dead ends, and flag anything you cannot verify for my review.
Field note: swap the bracketed details for your own, keep the approval steps — and make the agent show its work before it acts. That's the whole trick.— V.
← All case files← № 03: The $1,577 Vibe-Coded Client Win№ 05: Tetris Inside Minecraft, From One Prompt →
From the files · The product
Your agent is capable of real work. It just never got set up.
The Installation File is the setup program your own AI agent runs — interview, memory, email + calendar, three working routines — inside your account, with your data never leaving it.
$199regularly $249
· the Set-Up Guarantee