
Early in my career I pushed a container image with a development password baked into one of its layers. I had written the Dockerfile myself, built it, tested it, and looked it over twice. A colleague glanced at it for thirty seconds and said, “Isn’t that your real token on line twelve?” It was. I have thought about that moment ever since, because it taught me how many comfortable beliefs developers hold about reviewing their own work. Here are six of them, and why I no longer trust any of them.
Myth 1: “If I built it and it runs, it is fine”
Running is the lowest bar there is. A container can start perfectly and still carry a leaked secret, an oversized base image, a root user or a forgotten debug port. Those problems do not show up when you run it; they show up when somebody else reads the file. The author sees a working image. A reviewer sees every choice that was made on the way there, including the ones the author has forgotten making.
When people ask me what the first habit is for anyone learning containers, I say this: let another person read your Dockerfile before it leaves your machine. It is free, and it works.
Myth 2: “Reviews are for people who do not know what they are doing”
The opposite is true. The more experienced you are, the more you work from muscle memory, and muscle memory skips steps. I have watched very senior engineers paste an old snippet into a new project and forget to change a registry name. Beginners at least double-check because they are nervous. Experts often do not, and that is exactly when a reviewer earns their keep.
That is also why I like treating review as mutual. I read my friends’ files, they read mine, and nobody keeps score. It stops feeling like supervision and starts feeling like what it is: a habit between people who want each other’s work to be good. If you are brushing up on typical slips before asking someone to check yours, the list of frequent Docker errors that newcomers make makes a good pre-flight read.
Myth 3: “A linter or scanner does the same job”
Tools are brilliant at what they know, and I use plenty of them. A scanner will flag a known-vulnerable package and a linter will complain about an unpinned tag. Neither will notice that the image copies your whole home directory in, or that the entrypoint script does something surprising on the first run. A person reading with a bit of curiosity asks questions that nobody wrote a rule for.
My own routine pairs both: tools first, to clear the obvious, then a person, to find the unexpected. The practices described in the write-up on good Docker habits for development and live environments will give your tools and your reviewer a shared baseline.
Myth 4: “It takes too long to find someone”
Everyone says this, and nearly everyone has a person within reach. A teammate, a classmate, someone in an online community, a friend who works in a different company. Keep the request small. “Could you look at these forty lines for five minutes?” is a request people say yes to. “Could you review my whole project?” is not.
I also keep a note with every request: what the image is for, what I am unsure about, and where the secrets are supposed to be kept. It speeds everything up, and half the time writing it makes me spot the problem before I send it.
Myth 5: “Once it is in the registry, the review window has closed”
Some of my most useful reviews happened after publishing. Someone pulled the image, tried to run it on their own machine and found it assumed a path that only existed on mine. Image housekeeping benefits from a second pair of eyes too: tags, old layers, naming. The piece on looking after container images across an engineering team shows how much there is to keep tidy, and a colleague’s occasional glance keeps the clutter from building up.
If you only review before release, you miss the drift that happens after. A short monthly look at what you have published, with a friend, is a cheap way to keep a registry healthy.
Myth 6: “Asking for review makes me look junior”
This one matters for careers and for personal brand, so I will be direct. The engineers I most respect ask for review constantly. It signals that they care about quality more than appearance, and people remember that. When I mentor learners, I tell them to say out loud, “Can you check this before I ship it?” That sentence has never hurt anyone’s reputation. A habit of review, shown in public, builds trust faster than a hundred polished posts.
What a ten-minute review actually sounds like
People imagine review as a formal ritual. Mine usually sounds like this. I share my screen, scroll to the top of the file and say what each block is for. My friend interrupts: “Why is that package installed?” I explain, then realise it was only needed for a test I deleted. Further down: “What is this port for?” I pause, because I do not remember. We remove it. Near the end: “Where does this variable come from?” I show the build command, and he points out that it would print into the build log. Three catches, ten minutes, no jargon, and nobody feeling clever or foolish. That is all the process there is. The value lies in having a second mind in the room that does not share my assumptions.
Afterwards I note each catch in a small text file. After a few months that file became my own checklist: secrets, ports, packages, users, tags. I now run through it before I ask anyone, which means my friends see cleaner files and spend their time on the harder questions.
What does a good pair look like?
- Two people with slightly different backgrounds, for example one comfortable with application code and one with servers.
- A shared willingness to say “I do not understand this line.”
- A quick turnaround, ideally the same day.
- Mutual reading: both of you review, both of you are reviewed.
A word on non-technical forms
I use the same trick for everyday paperwork. When a relative was planning care overseas, we browsed directories, among them KneeHospitals, a patient-facing directory that sorts its hospitals and doctors under countries and offers forms to ask for a price or book a visit. I knew nothing about the medical side and did not pretend to. What I could do was sit alongside and read each field back before it was submitted: spelling of the name, the dates, the email address. A buddy check on a form is the same instinct as a buddy check on a Dockerfile.
Try it this week
Pick the next thing you are about to ship, maybe a Dockerfile, a script or a short post. Send it to one person with three lines of context. Ask them to look only for what surprises them. Then offer to do the same for them.
I still check my own work. I just no longer believe it is enough, and my image history, with its one embarrassing token, is why.







