The Package Firewall is a policy proxy in front of 25 package registries. From the Resolve Platform you set the thresholds, order your upstream mirrors, and review a full decision log of every package that passed, was blocked, or errored.
Security policies you control
Block known malicious packages
blockMalware: true rejects any dependency flagged as a known malicious package before the module metadata downloads. Intelligence from OSSF Malicious Packages, OSV.dev malware advisories and researcher-submitted samples is de-duplicated into a single corpus.
Block everything a compromised maintainer publishes
When a maintainer account is taken over, the damage is rarely one package. blockBadActors: true matches the maintainer behind a package (their registry account, or the GPG or SSH key their releases are signed with) against the Vulnesis threat-actor graph and blocks their whole catalogue. No CVE and no malware verdict are needed.
Block by vulnerability severity
Every request is checked against the Vulnesis VDB for known CVEs. If the highest CVSS v3 score on any CVE for that exact version meets or exceeds your threshold, the download is refused before it reaches your machine.
Block by exploitation probability
CVSS measures how severe a vulnerability is. EPSS, the Exploit Prediction Scoring System, measures how likely it is to be exploited in the wild in the next 30 days. A CVE with CVSS 7.5 and EPSS 0.97 is more dangerous than one with CVSS 10.0 and EPSS 0.01.
Block by exploit availability probability
Coalition ESS reports Exploit Availability Probability: the likelihood that working exploit code will be published for attackers to reuse. It is not another severity score. It answers whether exploit code is likely to be easy to find and put to use.
Block known exploited vulnerabilities
A KEV listing is not a prediction, it is confirmation that attackers are using the vulnerability now. US federal agencies must patch within 2–3 weeks of a CISA listing. blockKev: true applies the same standard to your dependency pipeline.
Block weaponized exploits
A weaponized exploit is mature, reliable attack code that needs no expertise to use. Honeypot intelligence tracks when proof-of-concept code graduates into an operational weapon. These are the CVEs behind ransomware campaigns and mass exploitation waves.
Block actively exploited vulnerabilities
blockActive: true blocks dependencies with confirmed active exploitation in the wild. CrowdSec tracks real attack traffic across sensors worldwide, and when a dependency's CVE shows up in that traffic the proxy blocks it.
Block when exploit code exists
A published proof of concept lowers the barrier to exploitation: attackers without deep expertise can follow the recipe. blockPoc: true closes that window for your dependency pipeline, before exploitation is confirmed in the wild.
Block end-of-life dependencies
End-of-life packages accumulate unpatched CVEs. Their maintainers have stopped shipping security fixes, so any new vulnerability becomes a permanent exposure. blockEol: true stops teams adopting one before it becomes a liability.
Delay newly published versions, carefully
Fast-follow attacks publish a malicious version seconds after a legitimate one, timed to catch automated upgrade scripts. cooldownDays creates a time buffer, but it is a freshness guard, not proof of safety: it also delays urgent fixes.
Avoid being first on a release
versionLag: N blocks a version until at least N newer versions exist. It counts releases instead of days, and like cooldown it only guesses that later releases bring more scrutiny. It does not prove the older version is safe.
Every block explains itself
Blocks return a semantic HTTP status (423 malware, 426 vulnerable, 428 bad actor, 425 cooldown, 422 EOL, 402 plan) with a full set of X-Vulnetix-Firewall-* headers carrying the reason, package, version and a docs link, so the reason survives even when a package manager such as Homebrew discards the response body.
Your call overrides ours, in both directions
Signals decide the overwhelming majority of packages; for the rest your team keeps a per-ecosystem deny-list and allow-list, scoped to an exact version or a whole package. The deny-list is enforced in every mode, and wins when a package appears on both. An allow entry waives every policy-derived block but never exact-version malware, a known-bad artefact hash or a malscan verdict. Every change writes an append-only audit row, each entry records when it last stopped or permitted real traffic, and a review queue surfaces waivers nobody has re-approved or used.
Roll it out without stopping the line
Warn mode downgrades every block to a logged advisory that still names the reason and the status that would have been returned, so you can measure the blast radius against a live build fleet before committing. Enforce applies the policy, and is the default. Allow-list-only inverts the default so a package with no allow entry is not installable, and falls back to normal enforcement when the list is empty rather than halting every build. Set any of them organisation-wide or per ecosystem.
Across every ecosystem
The pkg.go.dev API and AUR are free on every plan; every other registry is the Package Firewall add-on. 25 registries are supported: