Smaller, faster builds.
A slow Docker build taxes every single push. You wait on it in CI, you wait on it locally, and the bloated image it produces is slower to pull and pricier to store. Most Dockerfiles get written once, in a hurry, and never revisited, so they reinstall the world on every code change. This blueprint takes the one you already have and rebuilds it to be fast and lean, without you memorizing the dark corners of Docker's cache behavior.
A rewritten Dockerfile that does two things your current one probably doesn't. It splits the build into multiple stages, so heavyweight build tools and dev dependencies stay out of the final image and only the artifact you run gets shipped. And it orders the layers so the parts that rarely change, like dependency installs, sit above the parts that change every commit, letting the cache do its job instead of rebuilding from scratch.
You watch it happen in the task workspace: the agent reads your Dockerfile, restructures the stages and layer order, and builds the image to confirm it still works and comes out smaller. The updated file lands on a branch in your repository with a pull request ready.
The first pass covers stages and caching and is yours to steer with plain follow-up messages:
Each follow-up wakes the same repository and the agent continues on its own code.
Do I need to know how to code? No. You hand over your Dockerfile and say what matters; the agent rewrites it, builds it to prove it works, and ships the change to your repository.
Will my image still run the same? Yes. The goal is a smaller, faster build that produces the same running app, and the agent builds it to confirm before opening the pull request.
How much smaller will it get? It depends on your starting point, but moving build tooling out of the final stage often cuts image size sharply. The pull request shows the numbers.