Help centre · Understanding

What does Diavi do, step by step, during a run?

The pipeline's seven steps, what each produces, and which ones you can turn off.

A run chains seven steps, from cloning your repository to commenting on the merge request. Each produces something the next consumes, and you watch them go by live in Runs while the run is in flight.

What are the seven steps?

1. Clone — Diavi takes a fresh working copy of your repository, private to this run.
2. Reading the issue — an agent reads the issue, its comments, and identifies the files likely involved. If a decisive piece of information is missing, it stops here and asks.
3. Development — an agent writes the code, file by file.
4. Checking — a second agent reads back what was just written and marks it. This is the step that produces the score.
5. Merge request — the branch is pushed and the merge request opened against your target branch.
6. E2E tests — if you enabled them, an agent writes and runs end-to-end tests against your staging environment, and attaches the screenshots.
7. Review — a final agent reads the whole thing and comments on the merge request.

Why do two agents read the code?

They look for different things. The checker verifies conformance to the issue and the conventions before anything leaves — it is a guardrail. The reviewer comments on the merge request the way a colleague would, afterwards, and its opinion is for you.

Can I turn any of them off?

E2E tests and review are optional, at launch or by default in your settings. The other five are the pipeline itself.

What does "awaiting input" mean?

Step 2 stopped on a question. See Diavi asked me a question.