Every AI coding tool can produce a diff. The question that keeps the diff from shipping is the same one you would ask a junior engineer: "did you run it?" With most tools the answer is "you run it" — the burden of verification lands back on you.
CloudAxis Coding flips that. Before you ever see the pull request, the agent has already started the app, opened it in a real browser, screenshotted it, and run the tests. The evidence rides along in the PR.
What is in the proof bundle
When a coding session finishes a change on a web app, the pull request body includes a proof section assembled from the isolated machine while it was still running:
- A live preview URL — the app actually running, so you can click the change yourself.
- Multi-viewport screenshots — the changed page rendered on desktop and mobile widths.
- Console and page errors — captured while the page loaded, so a change that renders but throws in the console is not a silent pass.
- Test results — the outcome of your own test command, run after the change.
- Optionally, a short recorded walkthrough of the flow, when enabled.
Why "the event exists" is not proof
A detail we care about: a proof bundle only counts if it actually captured something. It is easy to build a system that always claims "proof attached" even when the screenshots failed. Ours does not. If a preview was possible for the repo but the screenshots did not succeed, the change is flagged rather than dressed up as verified.
And when a repo has no dev server to preview — a backend service, a library — the honest proof is the test suite, not a picture. We show you that instead of faking a screenshot. Evidence, matched to what the repo can actually produce.
Tests you can trust, not keyword matches
The "Tests: passing" badge only appears when your actual test command ran. We are deliberately strict about what counts as a test run — installing a test library or making a folder called tests is not running tests, and it will not paint a green badge. If the agent finishes without running your tests, that is recorded honestly instead of implied to be green.
What this changes about reviewing AI code
Reviewing a raw AI diff means holding the whole program in your head and guessing at runtime behavior. Reviewing a proof-of-work PR means looking at a picture of the result and a test outcome, then reading the diff with that context. It is the difference between "does this look plausible?" and "I can see it works — does the code look clean?"
That is the same review you would do for a human teammate who attached a screenshot to their PR. CloudAxis just makes the AI do the polite thing.
Frequently asked questions
What if the screenshots fail?
The change is flagged as unverified rather than presented as proven. For web repos where a preview was possible, real screenshot proof is required before the merge gate treats the change as verified.
What about repos with no UI?
Backend and library repos are proven by their test suite instead of screenshots. The proof is matched to what the repo can actually produce.
How CloudAxis Coding works · Approve-to-merge from your phone · We audited our own coding agent