The loop · How it works
Break it. Prove it. Heal it. Prove it again.
Every incident runs the same loop. Kane CLI judges both ends of it, and only a green Kane run closes the case.
- 01
Break
The Breaker
The Breaker breaks the store on purpose: a wrong total, a removed safety check — the way a careless edit would.
- 02
Verify
Kane CLI
Kane CLI opens the store in a real browser and uses it like a customer. Only Kane's verdict counts: if Kane says it works, the break was not real.
- 03
Heal
The AI Healer
An AI agent reads Kane's failure report plus the broken code and rewrites it. No answer key, no patch table.
- 04
Re-verify
Kane CLI
Kane checks again. Green closes the case. If the Healer fails within three attempts, the incident is recorded as failed — honestly.
Steps 03 and 04 repeat, up to three attempts.
Nothing is marked healed on the Healer's word alone.
- Green: case closed
- Still red after three: recorded as failed
The evidence
The failures stay in the log.
The public Immune Log shows every incident step by step — attack, Kane verdict, patch, re-check — with raw Kane output and patch files stored in the repo.
Credibility beats a fake 100%.
loopmune.onrender.com/log.html (opens in a new tab)Passed
Kane said the store still worked, so the break was not real.
3 of 15 incidents
Healed
The Healer rewrote the code and a green Kane run closed the case.
6 of 15 incidents
Failed
The Healer could not fix it within three attempts.
3 of 15 incidents
Error
The run itself errored, and it stays in the log too.
3 of 15 incidents
Inside every case file
- Attack
- Kane verdict
- Patch
- Re-check
Raw Kane output and patch files are stored in the repo (opens in a new tab).

Watch it heal itself.
Open the live store, follow every case in the Patient Monitor, or read the code on GitHub.