evonx
Log in
All pages

Automated testing & action nodes

Changes are split into verified steps; each step is tested before the next begins.

Evonx doesn't build a whole change and hope it works. Execution splits your request into action nodes — small, self-contained steps — and each one is verified before the next begins. This page explains what that buys you and how to read a run's progress.

How a run is structured

When execution starts, the plan becomes a sequence of action nodes: "add the photo field to the data model", "extend the API", "build the gallery on the job-detail screen". For each node the AI:

  1. Makes the change in the thread's isolated workspace.

  2. Verifies it — builds the project and runs the relevant checks, extending or adding automated tests where the change warrants them.

  3. Only then moves to the next node. A failure stops the line: the AI fixes the node before continuing, instead of stacking new work on a broken base.

You can watch nodes complete live on the thread's run progress — useful for spotting early whether the work is heading where you expected.

Why it matters

  • Failures are localized. When something breaks, it broke in the last small step — not somewhere in a thousand-line change.
  • Regressions get caught in the run, not by your staff next Monday. Tests added along the way keep guarding the behavior in later threads.
  • Reviews shrink. The pull request you receive arrives already building and passing its checks; your review is about intent, not "does it run".

Your part

Verification covers what's checkable mechanically. Whether the change is right for your business is still confirmed the same way as always: click through it in a preview before merging. Edge cases you state in the request ("same-day bookings close after 16:00") become checks the run can verify — one more reason precise prompts pay off.

Last updated: September 4, 2026