All machine-readable outputs produced by vulncheck use deterministic key ordering, sorted finding lists, and stable finding fingerprint hashes. Serialization is byte-reproducible for a fixed timestamp, while per-run metadata (generated_at, duration_ms, and SBOM/VEX document serials/timestamps) reflects the scan execution unless a fixed timestamp is supplied.
1. Native JSON Schema (--format json)
{
"duration_ms": 142,
"findings": [
{
"advisory_id": "EEF-CVE-2026-101028",
"aliases": [
"CVE-2026-101028"
],
"context_routines": [],
"cvss_score": 7.5,
"cwes": [
"CWE-863"
],
"fingerprint": "159f0146487b7b5140153f9b6a77bb94432c2ec3383d3cf47abcd1b8f6d04010",
"fixed_versions": [
"3.34.6"
],
"installed_version": "3.34.2",
"package": "ash",
"package_scope": "prod",
"reason": "call_chain_proven",
"reason_detail": "",
"recommended_fix": "3.34.6",
"references": [
"https://cna.erlef.org/cves/CVE-2026-101028.html"
],
"severity": "high",
"summary": "Ash.count, Ash.exists and Ash.aggregate skip related resources' read policies",
"suppressed": false,
"suppression_expired": false,
"suppression_expires": null,
"suppression_reason": null,
"target_mfa": "Ash.Actions.Aggregate.run/4",
"trace": [
{
"file": "lib/my_fintech_web/controllers/report_controller.ex",
"kind": "entrypoint",
"line": 41,
"mfa": "MyFintechWeb.ReportController.index/2"
},
{
"file": "lib/my_fintech/ledger.ex",
"kind": "static_call",
"line": 112,
"mfa": "MyFintech.Ledger.count_filtered_entries/2"
},
{
"file": "deps/ash/lib/ash.ex",
"kind": "static_call",
"line": 1184,
"mfa": "Ash.count/2"
},
{
"file": "deps/ash/lib/ash/actions/aggregate.ex",
"kind": "static_call",
"line": 29,
"mfa": "Ash.Actions.Aggregate.run/4"
}
],
"verdict": "reachable"
},
{
"advisory_id": "EEF-CVE-2026-103502",
"aliases": [
"CVE-2026-103502"
],
"context_routines": [],
"cvss_score": 5.9,
"cwes": [],
"fingerprint": "32cebf59e14a9377e324c49c74ac8afa420d4be22093b3ae7279d4965a75c265",
"fixed_versions": [
"0.10.2"
],
"installed_version": "0.10.0",
"package": "hex_core",
"package_scope": "prod",
"reason": "otp_version_unaffected",
"reason_detail": "requires OTP <= 25; running OTP 27.2",
"recommended_fix": "0.10.2",
"references": [],
"severity": "medium",
"summary": "hex_core verifies registry signatures with an outdated default on old OTP releases",
"suppressed": false,
"suppression_expired": false,
"suppression_expires": null,
"suppression_reason": null,
"target_mfa": ":hex_core.default_config/0",
"trace": [],
"verdict": "unreachable"
}
],
"generated_at": "2026-11-18T09:00:00Z",
"policy_violations": [],
"schema_version": "1.0.0",
"summary": {
"expired_suppressions": 0,
"noise_reduction_percent": 50,
"potentially_reachable": 0,
"reachable": 1,
"retired": 0,
"suppressed": 0,
"total": 2,
"unreachable": 1
},
"target": {
"elixir_version": "1.18.2",
"env": "prod",
"lockfile": "mix.lock",
"non_prod_packages_count": 19,
"otp_release": "27.2",
"prod_packages_count": 74,
"project_name": "my_fintech_api",
"project_type": "mix"
},
"tool": {
"name": "vulncheck",
"version": "1.0.1"
},
"warnings": []
}cvss_score and severity come from each source of the advisory:
- an Erlang Ecosystem Foundation CNA record gives its published score (
metrics[].cvssV4_0, elsecvssV3_1, elsecvssV3_0); - an OSV record gives the score computed from its CVSS v4.0 vector, else from its CVSS v3.1 or v3.0 vector.
Within an OSV record, valid ratings for the installed package and version take precedence over the record-wide rating. Within each scope, valid v4.0 vectors take precedence over v3.x vectors. When several sources describe one vulnerability, the highest applicable score wins. Optional JSON fields cvss_vector, cvss_method and cvss_source describe the selected score as one rating. A computed v4.0 score includes the Threat and Environmental metrics when the vector carries them. Scores range from 0.0 to 10.0 and map to none, low, medium, high and critical. none counts as low for --min-severity. A qualitative label is the fallback when no valid score is available. unknown counts as critical for the gate.
trace[].kind is how each hop was reached: entrypoint for the root, then static_call, macro_expanded, behaviour_callback, supervisor_child, dynamic_apply or imports. A chain whose root the project does not call itself (reason: "daemon_call_chain", or a module named in the configuration) starts with a step 0 hop of kind daemon_activation or config_activation. That hop points to the project code, configuration line or .app file that makes the root live, and has a note such as "starts Bandit". Its mfa is null when it points to a file rather than a function. A chain into a protocol client the project uses (mint, gun, finch) contains a hop of kind incoming_data: the client function that processes data received from the server, with the note "processes data received by the mint client". Only activation and incoming_data hops carry note.
package_scope is prod, build_only (a runtime: false dependency or a package only such dependencies need) or non_prod (only: [:dev, :test]). git_match_status ("confirmed" or "uncertain") appears on findings for Git dependencies, and feed.sources summarizes dependency source counts (hex, git, path). target.elixir_version records the Elixir version of the scanner VM (or null when unavailable), while target.otp_release records --otp-version if provided or the scanner VM's OTP release.
context_routines lists functions that the advisory names on the vulnerable path but that belong to Erlang/OTP, Elixir or another dependency (for example List.delete/2 in an HTTP/2 advisory). They explain the advisory and are never used as reachability targets.
2. SARIF 2.1.0 with codeFlows (--format sarif)
Designed for GitHub Code Scanning (github/codeql-action/upload-sarif):
- Each unique advisory maps to a
reportingDescriptorintool.driver.ruleswithsecurity-severitypopulated from CVSS so GitHub displays accurateCritical/High/Medium/Lowbadges. - The primary location is the first trace hop in the project's own code, including the activation site when present. Paths are normalized relative to the project root. An entirely external trace points to the package's entry in
mix.lock,rebar.lockormanifest.toml. GitHub workflow annotations and GitLab Code Quality use the same rule. SARIF includes the remaining project hops inrelatedLocations. codeFlows[].threadFlows[].locationsencodes the complete multi-hop call trace from entrypoint to vulnerable dependency routine. An activation hop hasexecutionOrder: 0andkinds: ["daemon_activation"]or["config_activation"]; the entrypoint hasexecutionOrder: 1andkinds: ["enter"].properties.contextRoutineslists routines named by the advisory that live outside the vulnerable package (seecontext_routinesbelow); they are never targets.UNREACHABLEadvisories are omitted by default (or emitted withlevel: "none"when--show-unreachableis passed).partialFingerprints.primaryLocationLineHashcarries the finding fingerprint (stable across runs), so Code Scanning tracks alerts across commits instead of reopening them.
3. OpenVEX v0.2.0 (--vex vex.openvex.json)
Compliance frameworks (SOC 2, PCI DSS v4.0, ISO 27001, EU CRA) increasingly require machine-verifiable Vulnerability Exploitability eXchange (VEX) documents explaining why unpatched transitive CVEs in an SBOM do not impact production.
vulncheck maps its formal verdicts directly to official OpenVEX v0.2.0 statuses and justifications:
| VulnCheck Verdict & Reason | OpenVEX status | OpenVEX justification |
|---|---|---|
REACHABLE | "affected" | — (action_statement lists fixed version) |
POTENTIALLY_REACHABLE | "under_investigation" | — (status_notes explains the reason) |
UNREACHABLE (function_not_called, module_not_loaded, adapter_not_configured, otp_app_not_started) | "not_affected" | "vulnerable_code_not_in_execute_path" |
UNREACHABLE (non_prod_dependency of a build-time only package, runtime: false) | "not_affected" | "vulnerable_code_not_in_execute_path" |
UNREACHABLE (non_prod_dependency of a :dev/:test package) | "not_affected" | "component_not_present" |
UNREACHABLE (otp_version_unaffected) | "not_affected" | OTP constraint in impact_statement |
Policy suppressions preserve the original vulnerability status in OpenVEX and CycloneDX. Both formats include the complete scan snapshot before diff and baseline filtering.
Retired Hex releases produce no VEX statements. When a scan has zero statements, the CLI reports that OpenVEX was omitted on stderr and leaves stdout or a new destination empty. An existing OpenVEX destination produces exit code 2 and preserves the file so automation can distinguish a stale artifact. The API returns nil from OpenVEX.document/3 and an empty string from OpenVEX.encode/3. The document @id is derived from the content.
4. CycloneDX 1.6 SBOM (--sbom sbom.cdx.json)
Generates a standards-compliant CycloneDX 1.6 Bill of Materials for the BEAM release:
- Component
purlidentifiers in canonical Hex format (pkg:hex/phoenix@1.7.18), plus git commit refs for git dependencies. scope: "required"for:prodruntime packages andscope: "excluded"for:dev/:test-only dependencies.- A
dependenciesgraph built from the lockfile's dependency edges, with the project's direct dependencies (frommix.exs,rebar.configorgleam.toml) as the root component'sdependsOn. vulnerabilities[].analysisrecordsexploitablefor a reachable match,in_triagefor a potential match, andnot_affectedwithcode_not_reachable,code_not_presentorrequires_environmentas supported by the evidence. Retired releases are omitted from the vulnerability list.- SHA-256 component hashes from the lockfile checksums and a deterministic
serialNumber, so identical scans produce identical SBOMs. - The project component's reference includes the declared project version when available and a
scan_scopedigest. It identifies the dependency inventory, runtime and any analyzed compiled evidence. The VEX product uses the same identifier. CycloneDX metadata gives the build profile, evidence status and snapshot scope. The identifier covers the scanned workspace and is scoped to the supplied OTP environment. Aliases for the same package URL have distinct component references in both outputs.