BuildKit

Password Deleted from Git but Still Present: Major CI/CD Security Mistakes

Developers often leave secrets in configuration files, .env examples, or test scripts that end up in Git history. Removing a file in a new commit does not erase previous versions, and .gitignore offers no protection for already tracked content. Pre-commit hooks with tools like Gitleaks can scan staged changes before they reach the repository. Docker images can retain secrets in layers even after removal commands, requiring BuildKit secret mounts instead. Terraform plans and state files may still expose sensitive values marked only as sensitive. Kubernetes workloads need both non-root containers with dropped capabilities and narrowly scoped RBAC roles. Multiple scanning stages from pre-commit through post-deployment help catch issues before they reach production.

Where Secrets Really End Up in Docker Images: Testing 8 Common Methods

A detailed analysis of eight different approaches for handling secrets during Docker image builds reveals that most methods leave sensitive data exposed in layers, configuration files, or build history. Experiments conducted on Ubuntu 24.04.1 LTS with Docker Engine 29.1.3 and BuildKit v0.26.2 showed that simply using RUN rm after COPY or overwriting files fails to remove secrets from immutable layers. Environment variables via ENV and build arguments via ARG both persist in image metadata and history. Only BuildKit Secrets with --mount=type=secret and multi-stage builds that avoid copying secrets into the final stage successfully prevent leakage. The study demonstrates that secrets can be extracted from saved images without running any containers. Developers are advised to adopt these two secure patterns to protect credentials in container supply chains.