Kane CLI Online Hackathon · TestMu AI

Your agent and Kane, talking until the app is verified green.

kane-loop watches your code. On every save it runs your Kane CLI flows in a real browser, and when one fails it hands the failure to your agent, which patches the code. The app heals itself until Kane is green, with no human in the middle.

Open the live checkout demo View the source

Why this exists

The loop that never closed

AI agents write code fast. Features ship from a prompt. But one part of the loop stayed manual: when the agent ships something, a human still opens a browser and clicks through to see if it actually works.

That is the bottleneck now. Not writing the code, trusting it. kane-loop closes it.

How it works

Save. Verify. Heal. Green.

1
You save
A file changes and kane-loop notices.
2
Kane runs it in a real browser
Your plain-English flow runs headless. The total is wrong, so Kane fails the run.
3
The agent reads the failure
The reason and the value Kane observed go to your coding agent, which edits the code.
4
Green
The save re-fires Kane. It passes. No human in the middle.

Built on the whole Kane suite

Not one command. All of it.

run --agent

The live verification, its NDJSON parsed to a pass or fail.

testmd run

Committable _test.md flows, cached and replayed for free.

context + design

A PRD becomes acceptance criteria and scenarios.

testrun

Many tests, one sealed evidence pack, gating CI.

evidence

Screenshots and logs the agent fixes from.

balance · export · skill

Live credits, a Playwright export, an agent skill.

Provenance

Everything here is real

red to greenreal self-heal, on camera
2 / 2committable tests, sealed
11 ACsdesigned from the PRD
MITopen source

Live checkout: kane-loop-demo.pages.dev · Source: github.com/zkasuran/kane-loop

Built for the Kane CLI Online Hackathon (TestMu AI). Built with an AI coding agent; a human designed, reviewed and verified it. The self-heal shown is real and nothing is mocked.