Auto-remediation
In the repository’s Settings → Dependencies, enable automatic PR creation for the dependency types you want to cover. Each type has a separate minimum vulnerability score.- Auto-Create PRs for Direct Dependencies proposes an upgrade to a dependency listed in the manifest. Set the Direct Dependency PR Threshold.
- Auto-Create PRs for Transitive Dependencies looks for a direct parent it can upgrade to remove the affected transitive version. Set the Transitive Dependency PR Threshold.
- For scheduled dependency scans, set Dependency scan branches to the branches you want to monitor.
Ivy findings use Maven package identities, but the Maven
dependency-upgrade patcher does not update
ivy.xml. Ivy advisories open
standard patch review instead of a chosen-release upgrade. Review the Ivy
revision or its source property, validate the project and re-scan.Reading patch and issue status
Patch PR accepted records acceptance of the patch pull request. The finding’s Current workflow is shown separately: an accepted patch PR can coexist with a reopened linked issue. A resolved linked issue can also appear under Patch PR → Not accepted when it has no accepted patch PR. Use both facts when reviewing remediation; patch acceptance alone does not establish the current issue status.Choosing a safe upgrade
ZeroPath selects an upgrade target rather than always taking the latest version. It first looks for the nearest version with zero known CVEs. If none exists, it falls back to the nearest version that fixes the specific vulnerability without introducing any new CVEs you aren’t already exposed to. It avoids impractical major-version jumps. Selection is against current advisory data, so it cannot account for vulnerabilities disclosed after the upgrade is proposed.When a clean upgrade doesn’t exist
- Remediation guidance — when a vulnerability can’t be auto-patched (no safe version, or the fix needs architectural changes), ZeroPath provides step-by-step instructions so a developer knows exactly what to do manually.
- Force-generate — for a finding marked unpatchable, you can override the verdict and ask the patch agent for the smallest-blast-radius compensating control it can produce — an authorization check, a feature flag, or a defensive guard around your use of the package. This reduces exposure but does not remove the vulnerable dependency; the advisory remains in your inventory until a real upgrade exists.
- Reliable delivery — if a patch run completes without writing any edits, ZeroPath automatically retries it once, so an empty patch doesn’t reach your review queue.
- Repository-aware patchability decisions — the patch agent explores the affected file and its surrounding code before concluding that a vulnerability cannot be auto-patched. A finding is only declared unpatchable for substantive reasons — high regression risk, a fix that requires secret rotation, or a cross-cutting architectural change — not because the agent lacked context it could have fetched itself.
Adjusting and comparing patches
- Adjust Patch — after a patch is generated, you can reject it and submit free-form feedback describing what you want changed. The agent reads your feedback and produces a revised patch that honors it.
- Choose a Patch — you can request multiple alternative patches for the same vulnerability. Each variant uses a meaningfully different fix technique, location, or structure, so you can compare the options and select the one that fits your codebase best.
Linking to your tracker
If findings are linked to Jira issues, you can use{jira_id}, {jira_url}, and
{jira_title} placeholders in your custom PR title, description, branch name, and
commit message templates. ZeroPath fills them in automatically, so each
remediation PR links back to the corresponding tracker issue.
Verify a dependency fix
After upgrading or removing a vulnerable package, commit the updated manifest and lockfile, then run a fresh full or SCA scan. ZeroPath can resolve a confirmed finding when the scan verifies that the package was removed or that all installed versions are outside the advisory’s affected ranges. Verification uses the scanned commit, the resolved inventory, and dependency evidence ZeroPath can verify. For pnpm workspaces, commit the workspace’s updatedpnpm-lock.yaml along with any changed member manifests. For uv, commit the
updated uv.lock and pyproject.toml as applicable. Branches and monorepo
partitions are assessed in their own scope.
A later scan of the same PR can also clear a previous PR dependency finding when
the new commit verifies the fix. Run a scan of the target branch after merging
to update its findings and dependency inventory.
If a finding stays open, check that the scan used the intended commit and branch,
then review coverage warnings. An incomplete
graph, unsupported lockfile configuration, or unavailable advisory can leave
the fix unverified. Check for other installed copies of the affected package.
CVE alerting
CVE alerting proactively notifies you when a new vulnerability is discovered in a package you already depend on, without waiting for the next scan to run. In the repository’s Settings → Dependencies, turn on Enable CVE Alerting. Then open Issues → Dependencies to review affected packages and findings across repositories. For one repository and branch, open Inventory → Risk → By CVE to group its findings by advisory.Reachability and exploitability assessments are AI-assisted and probabilistic. For
critical supply-chain vulnerabilities, confirm the assessment before relying on it
to deprioritize a finding or auto-merge a fix.
Accuracy safeguards
- Cross-package advisory filtering — when an advisory covers multiple packages, ZeroPath verifies its affected entry matches your package before raising an alert, removing the false positives that broad multi-package advisories would otherwise produce.
- Noise control — the pipeline skips metadata-only advisory updates (a CVSS tweak or a new reference URL) and only re-processes a CVE when its affected packages or version ranges actually change. Transient processing failures are retried automatically and do not stall alerting for other advisories.