Patching treats each finding as a task. Avoidance removes the source: a base image with fewer packages and a faster rebuild cadence, and versions picked because they are clean, not because they are newest. What is left to patch is short enough to finish.
Move the base, not the findings
Compare each container image with hardened equivalents, then follow a written migration plan built from its SBOM, its Dockerfile and where it runs.
Upgrade to the closest clean version
One version policy across every manifest picks releases with no known vulnerability, exploit or malware report, and opens the pull request.
Never install what you would have to patch
The Package Firewall blocks vulnerable, malicious and untested versions at install time, and vendor aware advice covers what cannot move yet.
Move the base, not the findings
Most of what a container scan reports came with the base image. Safe Harbour compares each image's base with the equivalents hardened registries publish, scores every candidate itself, and shows what a move would change before you make it.
Where the candidates come from
- Zero CVE. Images rebuilt as fixes land, with the vendor aiming for no known vulnerability at release, such as Chainguard, Docker Hardened Images and Red Hat Hardened Images.
- Hardened. Images built to a published hardening standard, such as Iron Bank, Bitnami Secure and the SUSE Application Collection.
- Minimal. Small bases with less to attack, such as distroless, Wolfi, Canonical chiselled images and UBI.
- Standard. The general purpose images most Dockerfiles start from, such as Docker Official Images, so a move away is measured from where you are.
Scored by Vulnesis, not taken on trust
Each image gets a score out of 100 from safety, freshness, provenance, maintenance, attack surface. Strong 85 and over, Good 70 to 84, Fair 50 to 69, Poor under 50. A candidate must run the same runtime, libc and architecture and score at least as well as the base you run.
- Signature. The vendor signature is verified against Sigstore and the signer the vendor publishes.
- Vendor claim against Vulnesis. Vulnesis assesses each image on its own and lists any vulnerability the vendor claim leaves out.
- SLSA level. Build provenance is read from the attestations the image carries.
- What you will have to change. A non root user, a release change, an account to pull, no shell or no package manager, each named before you move.
How close the match is
- Same release. The same distribution release from another registry: the smallest move.
- Release change. A newer release of the same distribution, with each base package that changes version listed.
- Distribution change. A different distribution with the same runtime and libc, offered when it removes enough to be worth it.
A plan written from what you build and what you run
A recommendation is only useful once someone makes the change. The plan reads three things you already have and writes the change for you, so a better base is a review and a merge rather than a project.
- The SBOM. vulnetix image records what the image holds, so the plan knows every package your own layers add and which of them the new base ships.
- Where it is built. vulnetix containers in CI uploads the final FROM line and the install lines of each Dockerfile, so an image is planned from the file that builds it, even before an SBOM exists.
- Where it runs. A connected AWS account lists where EKS, ECS, Lambda and ECR use the image, so a move is weighed against the workloads it changes.
- Review what changes. Each change for the people running the image, and each base package that moves version.
- Pin your packages. Every package your layers install, pinned under the policy you choose to a version the new release ships. A package with no pick says so; the plan never guesses.
- Change the Dockerfile. FROM pinned by digest, a build stage when the new base has no shell, and install lines in the package manager's own pin syntax.
- Registry access and CI. The sign in your machine and your CI need to pull the new base, written as files to commit.
- Check the result. Commands for cosign, Grype, Trivy and the Vulnesis CLI, so you check the result with tools you already run.
What is written for you
- A Dockerfile. FROM pinned by digest, a dev stage when the target ships no shell, and install lines in the package manager's own pin syntax under the policy you chose.
- Your manifests. package.json, requirements.txt, go.mod, Cargo.toml, pom.xml, composer.json and Gemfile edited as text, with transitive fixes written in the override section each ecosystem uses.
- Generated lockfiles. Lockfiles are generated from the edited manifests with install scripts switched off, shown read only, and never edited by hand.
- Local setup. The registry login, a .env, compose and devcontainer overrides, a CI login step and a Renovate rule, so the new base keeps working after the first build.
- A pull request. Opened or updated through the GitHub App, with the checks that read your manifests listed first; or the files to download when the app is not installed.
The evidence behind each move
- Advisories. Every vulnerability id links to its record in the Vulnesis database, with the exploit and known-exploited evidence that decides its urgency.
- The registry. Each alternative links to the registry page it came from and the digest its tag resolved to when it was read.
- Signed SBOMs and claims. The vendor signature, the signed SBOM and the vendor vulnerability claim are published through the Transparency Exchange API, with the trust anchors to verify them.
- Independent checks. Copyable commands for cosign, Grype, Trivy and the Vulnesis CLI, so the result is checked with tools you already run rather than taken on trust.
- The gap. When a vendor claim omits a vulnerability Vulnesis found in the same image, the page says so and links each omitted id.
Safe Harbour: upgrade to the closest clean version
When you do need to move, the question is not what is newest. It is what is safe, and how little you have to change to get there. Apply one version policy across every manifest, fix problems deep in the resolved tree and open a pull request with the dependency curator.
Safest
The highest stable release that Safe Harbour clears. Chosen by version order. A pre-release is never picked.
Stable
The stable release closest to the one you run. Closest by major, then minor, then patch, with a tie going to the higher version. It can be lower than the version you run.
Latest
The highest stable release by the ecosystem's own version rules. A pre-release is never picked. Where versions follow SemVer, the highest version wins whatever its date. Where versions cannot be ordered by number, the most recent release by date wins. A point release on an old branch is newer by date but is not the latest.
Rolling
What shipped most recently, closest to the one you run. Takes every version published on the release day of the highest version, pre-releases included and malware removed, and picks the one closest to your version.
Pix investigates, Safe Harbour decides what to move to
Pix does the investigation and the edits around a Safe Harbour move, either as a plan for you to approve or as an agent session that ends in a pull request. Each one starts from a button on the Resolve Platform page where the question comes up.
- Find out why the checks failed (Safe Harbour pull request). Pix reads the failing checks and their logs, says whether the version changes caused the failure or the base branch fails too, and writes a plan to fix it.
- Open the pull request for a migration (Migration plan). Pix applies the plan's Dockerfile in the repository that builds the image, builds it and opens a draft pull request that the Safe Harbour workspace tracks.
- Upgrade a package in this repository (Dependency page). Pix takes the Safe Harbour picks the server worked out, upgrades the package in that repository with its own package manager, and says what broke and the smallest fix when the tests fail.
- Move this repository off an end of life release (End of life). Pix finds where the release comes from, whether a dependency, a base image, a runtime or a CI version, and moves it to the recommended release.
- Look it up while it works (Any session). Through the Vulnesis MCP server Pix looks up the vulnerabilities, exploits and packages behind a move before it makes it.
What else Pix does · Safe Harbour work in the Software Factory
Never install what you would have to patch
Once the base and the versions are clean, keep them that way. The Package Firewall sits in front of the registries your developers and pipelines already pull from and enforces one organisation policy on every request, so risky, malicious or untested versions are turned away before they reach a build.
Vendor-aware patch advice
Real vendors backport fixes, ship errata for a specific release and set end of life and KEV due dates. Vulnesis sources patch guidance from the VDB, so what cannot move yet is patched the way your vendor actually ships the fix. Browse the vendor patching archives
Avoid first, patch what is left
A hardened base removes the findings that came with the image. Safe Harbour pins every package to a version with no known vulnerability, exploit or malware report. The firewall keeps risky versions out, and vendor aware advice covers what cannot move yet.
How the Vulnesis Package Firewall compares
Evaluating Aikido SafeChain, Socket, JFrog Artifactory or DevGuard? Each is a capable supply-chain tool; here is where the Vulnesis Package Firewall is different, and where it wins.
One de-duplicated corpus, every major feed
The firewall checks each install against a malware corpus that aggregates OSSF Malicious Packages, OSV.dev malware advisories, GitHub Advisory and Vulnesis first-party research into one de-duplicated set: one of the largest in the industry, and broader than any single-feed scanner.
12 policy controls, including EOL & exploit intelligence
Beyond malware, gate installs on CVSS, EPSS, Coalition ESS, CISA KEV, weaponized / active / PoC exploit maturity, bad-actor association, cooldown and version lag, plus end-of-life blocking that competitors do not offer. Tune every threshold per ecosystem.
Your own allow-list and deny-list, with a review queue
Signals decide most packages; you decide the rest. Keep per-ecosystem allow and deny lists scoped to an exact version or a whole package, each with an append-only audit trail, a last-seen timestamp recorded when the entry actually stops or permits traffic, and a staleness queue that surfaces waivers nobody has reviewed or used. An allow entry waives policy, never exact-version malware, a known-bad hash or a malscan verdict.
Observe first, then enforce or default-deny
Warn mode downgrades every block to a logged advisory carrying the reason and the status that would have been returned, so you can measure blast radius against a live build fleet before enforcing. Allow-list-only mode inverts the default so nothing installs without approval, and falls back to normal enforcement when the list is empty rather than halting every build. Set either org-wide or per ecosystem; the deny-list is enforced in all of them.
Block, then fix
When a version is blocked, Safe Harbour resolves the safest, latest, stable or rolling version that clears the finding and autofixes to it. This remediation step is unique to Vulnesis, found nowhere else among package firewalls.
Vulnesis Package Firewall vs Aikido SafeChain, Socket, JFrog & DevGuard
Aikido SafeChain
A free CLI that wraps npm / pnpm / yarn to block known-malicious packages at install time.
They focus on Malware interception for JavaScript package managers.
Vulnetix edge Vulnesis firewalls 25 registries (beyond JavaScript) and layers 12 configurable policies (CVSS, EPSS, Coalition ESS, CISA KEV, end-of-life, exploit maturity) on top of malware blocking, then offers Safe Harbour autofix to a known-clean version.
Socket
Socket.dev and Socket Firewall (sfw): behavioural / AI analysis of package signals such as install scripts, obfuscation and network access.
They focus on Detecting and alerting on suspicious package behaviour, mostly across npm and PyPI.
Vulnetix edge Vulnesis adds policy-grade gating on CVSS, EPSS, Coalition ESS, CISA KEV, end-of-life and exploit maturity, draws on a de-duplicated malware corpus aggregated from OSSF Malicious Packages, OSV.dev and GitHub Advisory, and turns a block into a fix with Safe Harbour autofix, not just an alert.
JFrog Artifactory, Curation & Xray
Artifactory is a universal repository manager whose remote repositories cache upstream artefacts and can be restricted with include/exclude patterns; Curation gates packages on the way in; Xray scans stored artefacts for CVEs and licence issues.
They focus on Owning the binary as a stored, cached artefact (allow-list patterns, offline resilience and retention), with policy gating and SCA layered on for teams standardised on the JFrog Platform.
Vulnetix edge Vulnesis needs no Artifactory because it proxies any package manager directly, and it adds end-of-life and exploit-intelligence policies, threat-actor maintainer attribution and Safe Harbour autofix, with the pkg.go.dev API and AUR free on every plan. Artifactory keeps the edge on artefact storage: it caches what it proxies and Vulnesis does not.
DevGuard
An open-source DevSecOps platform covering SCA, SBOM, VEX and in-toto supply-chain attestations.
They focus on Self-hosted, open-source software-composition and supply-chain attestation.
Vulnetix edge Vulnesis pairs a managed VDB and de-duplicated malware corpus with exploit intelligence, a 25 registry install-time firewall, and Safe Harbour autofix, with no infrastructure to run.
See all package firewall alternatives compared →
Safe Harbour: questions people ask
What is vulnerability avoidance?
Choosing sources that do not carry known vulnerabilities in the first place: a hardened base image and package versions with no known vulnerability, exploit or malware report. It removes findings before a scan reports them, so patching is left for the few that remain.
Which hardened images does Safe Harbour compare against?
Images from zero CVE, hardened, minimal and standard registries, including Chainguard, Docker Hardened Images, Red Hat, Iron Bank, distroless, Wolfi and Docker Official Images. A candidate must run the same runtime, libc and architecture and score at least as well as your current base.
Do I have to trust the vendor's vulnerability claim?
No. Vulnesis assesses every image itself, publishes the signed SBOM and both assessments through the Transparency Exchange API, and lists any vulnerability the vendor claim leaves out. The verify commands let you check the result with cosign, Grype and Trivy.
What does a migration plan need from me?
Any one of an SBOM from vulnetix image, a Dockerfile uploaded by vulnetix containers in CI, or a connected AWS account. The more of the three it has, the more of the plan it can write: the SBOM gives the package pins, the Dockerfile gives the lines to change and the deployments say which workloads move.
What is Safe Harbour autofix for packages?
Safe Harbour resolves the Safest, Stable, Latest or Rolling version that clears a finding and writes it into your manifest, with lockfiles generated and a pull request opened, fixing the problem rather than only reporting it.
Is this an alternative to Socket, Aikido SafeChain, JFrog Curation or DevGuard?
Those tools block or alert at install time. Vulnesis does that with the Package Firewall too, and adds the avoidance side: a hardened base with a written migration, and versions Safe Harbour clears written into your manifests with a pull request.
How does Pix help with Safe Harbour?
Pix finds out why checks failed on a Safe Harbour pull request, upgrades a package to its Safe Harbour pick in a repository, and moves a repository off an end of life release, including when the release comes from a base image.