Dependencies first, source second
The manifest and lockfile are copied and installed before the source arrives, so editing a component does not reinstall the tree. The multi-stage split carries the compiler only as far as the build.
Pikkola Docker, tool No. 63
Pick the stack, the build and the extras, and copy a Dockerfile with the multi-stage split, a non-root user, a health check and cache mounts already in the right order. The same bench writes the .dockerignore. A compose file puts a database, a cache and a proxy around it, and the audit names every line of an old Dockerfile that will cost you later.
The studio
Build, compose, audit. Each bench writes a file you can copy or download, and neither of them sends it anywhere.
Node.js runs dist/server.js, Next.js reads the standalone output, Python runs uvicorn, Go builds ./cmd/server, Rust builds the Cargo target and Nginx serves a built dist.
No request, no account, no storage. The Content-Security-Policy sets connect-src to none, so the page could not upload a file even if it tried.
A Dockerfile is a cache key with a shell attached. The order of the lines decides what has to run again when one file changes.
The manifest and lockfile are copied and installed before the source arrives, so editing a component does not reinstall the tree. The multi-stage split carries the compiler only as far as the build.
A pinned base tag, a non-root user, a health check that fails when the process is up but the app is not, and cache mounts that survive a rebuild. Turn one off and the file shows exactly what you gave up.
Instructions are joined across continuation lines and read one by one, so a finding knows which instruction it sits in. That is how COPY . . before the install is caught, and why a RUN split in two is caught as well.
Eight rules about the file, each with the line it starts on. Two of them are errors, the rest are warnings and notes, and a clean file earns nothing at all.
It reads the file, not the build. Whether the image runs, which port the network reaches and what your .dockerignore leaves out are outside what a text file can say.
No. The generators are functions in a JavaScript file on this domain, and the page's Content-Security-Policy sets connect-src to none, so a request has nowhere to go. Load the page once and it keeps working with the network off.
It is a starting file, not a verified build. Every path it names, such as dist/server.js, app.main:app, ./cmd/server or the Cargo target, is the convention of that stack. Rename yours and change that one line. Your application was not run here.
node:22-bookworm-slim and its siblings are chosen because the tag names a version rather than latest. A base that moves under you cannot be reproduced, which is the first thing the audit reports. Bump the tag when you mean to.
It is the wiring: build, ports, restart policy, depends_on with a health condition, named volumes, and a health check for the app itself. The app still needs its environment in .env, and a proxy needs the nginx.conf that the second tab writes beside it.
It reads text, so it cannot tell whether the runtime really needs curl, whether the port is reachable, or what your .dockerignore leaves out. It checks the file, not the build.
The Dockerfile reference as it stands in 2026: FROM, WORKDIR, COPY, ADD, RUN, ENV, ARG, USER, EXPOSE, HEALTHCHECK, ENTRYPOINT and CMD. Cache mounts come in through the syntax line, and the audit reads the same set of instructions.