How can we stop a newly published typosquatted package before it runs during a build?

Put a gate in front of package installation rather than relying only on image scanning. Resolve dependencies against a lockfile, require exact versions and integrity hashes, and route installs through an internal repository that only serves approved artifacts. That also prevents an unreviewed change from silently arriving through a floating version.

Disable lifecycle or post-install scripts by default and maintain a small allowlist for packages that genuinely need build steps. Package managers can often prevent arbitrary install hooks from executing, which closes the biggest gap between pulling a dependency and scanning the resulting image.

CVE scanners are useful, but they will miss something that was published yesterday. Add pre-install reputation and behavior checks: package age, maintainer history, download patterns, ownership changes, install hooks, and whether the package tries to access the network. Run that against the lockfile before the package manager starts installing anything.

Use a repository proxy or mirror for both packages and base images. Scan and approve artifacts as they enter the mirror, then make production builds pull only from that controlled source. Avoid floating tags such as “latest,” and pin image digests as well as dependency versions.

Also enforce default-deny network egress during builds. Even if a malicious install script gets through, it should not be able to download another payload or send credentials out. This is defense in depth, not a replacement for dependency approval and script restrictions.