Offline version-evidence checker for CVE-2026-56847: Node.js Permission Model trace-log filesystem bypass.
You supply the Node.js versions your runtimes report. The checker compares them with the upstream fixed releases and writes a CSV, JSON or HTML report. It is triage evidence, not an assessment of any deployed system.
- Purpose and scope
- Requirements
- Quick start
- Synthetic report preview
- Usage
- Rule snapshot
- Interpreting results
- Reports
- Limits and failure behavior
- Limitations
- Sources and provenance
- Project reference
- Contributing, security and license
This repository covers only CVE-2026-56847. It compares operator-reported runtime versions with the fixed releases in a dated rule snapshot.
The checker never contacts websites, executes Node.js or project code, or sends exploit payloads. No network mode is implemented. It does not scan the current machine, and a public URL is not an inventory. Running it authorizes nothing: assess only runtimes you are permitted to assess.
Every result carries exploitability: not_tested. Hidden versions, custom builds, conflicting evidence and uncovered release lines remain unknown.
- Python 3.10 or later, standard library only. There are no runtime dependencies, and
node_cve_2026_56847_check.pycan be copied and used alone. - CI runs the unit tests on Ubuntu and Windows with Python 3.10 and 3.14.
- An exact version string taken from the deployed process, which you must obtain yourself.
python node_cve_2026_56847_check.py --version 22.23.2 --format json
python node_cve_2026_56847_check.py --inventory examples/inventory.example.json --format html --output report.html
python node_cve_2026_56847_check.py --csv-list inventory.csv --fail-on-finding
python node_cve_2026_56847_check.py --help
python -m unittest -v
The default format is CSV, written to stdout. --version is repeatable, and each value describes a separate runtime. Existing report files are protected unless --force is explicit.
Open the synthetic example report. Example data only; this is not a live assessment.
At least one of --version, --inventory or --csv-list is required. They can be combined in one run, and all inputs are assessed with the same rules.
| Option | Purpose |
|---|---|
--version VERSION (alias --node-version) |
Assess one explicit Node.js version, with or without a leading v. Repeatable; each value is a separate runtime. |
--build-source SOURCE |
Provenance of the --version builds, at most 512 characters. Default unknown, which is compared as upstream. See Interpreting results for how vendor sources are treated. |
--permission-model-enabled, --trace-events, --filesystem-write-allowlist |
Operator observation (true or false) applied to every --version input. See Precondition observations. |
--inventory FILE |
JSON object or array of records. Repeatable. |
--csv-list FILE |
CSV with a node_version column. Repeatable. |
--format csv|json|html |
Report format. Default csv. |
--output FILE |
Write the report to a new file. Default is stdout. |
--force |
Replace an existing output file. Inputs can never be overwritten, even with --force. |
--brand NAME |
Organization name shown in the HTML report, at most 512 characters. Default CompleteTech LLC. |
--fail-on-finding |
Exit 1 when any result is potentially_affected or unsupported_runtime. |
--tool-version, --help |
Print the tool version, or the full option list and exit-status summary. |
The output directory must already exist. Without --force, the checker publishes by hard link and refuses to overwrite; on a filesystem without hard links it stops with an error unless you choose another location or pass --force.
Both formats feed the same assessment, and JSON, CSV and HTML reports are built from the same evidence. Without a label, a record is named <file name without extension>:<row number>. Use examples/inventory.example.json as a starting point.
JSON. One object or an array. Each record requires the integer schema_version: 1 and the string node_version. Optional fields are runtime (only "node" is accepted), label, build_source and the observation keys below, each a boolean or null. Unknown keys, duplicate keys, non-finite numbers and invalid types are rejected.
CSV. Needs a header row with node_version. Optional columns are label, build_source and the observation columns, whose values are exactly true, false or blank. Undocumented or duplicate columns, rows with the wrong number of fields, bare quotes inside unquoted cells, and NUL characters are rejected.
label,node_version,build_source,permission_model_enabled,trace_events,filesystem_write_allowlist
api-example,22.23.1,,true,true,
| Observation | Meaning | CLI flag |
|---|---|---|
permission_model_enabled |
permission model enabled | --permission-model-enabled |
trace_events |
trace events | --trace-events |
filesystem_write_allowlist |
filesystem write allowlist | --filesystem-write-allowlist |
Observations are recorded as reported_true, reported_false or not_reported. They are operator-supplied and unverified. They never change the version result and never establish reachability.
Reviewed 2026-10-24 against the official release advisory and the linked fixed releases.
| Node.js major | Fixed version |
|---|---|
| 22.x | 22.23.2 |
| 24.x | 24.18.1 |
| 26.x | 26.5.1 |
The vendor does not specify an introduced version in the release advisory, so earlier versions of each listed major are potentially_affected rather than confirmed affected. The advisory note carried into every result is:
For each listed major, versions below its fixed version require remediation. Unlisted, custom/backported, and prerelease builds need explicit evidence; do not label unknown builds safe. The vendor does not specify an introduced version in the release advisory.
| Status | Meaning |
|---|---|
potentially_affected |
A plain stable version below the fixed release of its listed major. It needs remediation review; exploitation was not tested. |
fixed_version_reported |
A plain stable version at or above the fixed release of its listed major. This is a reported version, not attestation that the deployed runtime is patched. |
unsupported_runtime |
A major past end-of-life in the Node.js release schedule, reported as upgrade guidance. It does not claim this flaw exists in every historical build. |
unknown |
Missing or unclassifiable evidence. It is never evidence of safety. |
version_status gives the specific reason:
version_status |
Status | Cause |
|---|---|---|
below_fixed_version |
potentially_affected |
Below the fixed release of a listed major. |
at_or_above_fixed_version |
fixed_version_reported |
At or above the fixed release on a listed major. |
unsupported_release_line |
unsupported_runtime |
End-of-life major: 0, 4 to 21, 23 or 25. |
release_line_not_in_rules |
unknown |
A major not covered by this snapshot, such as a newer or other active line. It is not assumed patched. |
version_not_disclosed |
unknown |
No version was reported. |
unrecognized_or_partial_version |
unknown |
Malformed or partial version, or non-text metadata. |
qualified_version_unverified |
unknown |
Prerelease, nightly or rc build, package suffix or other qualifier. |
vendor_build_unverified |
unknown |
A build source other than blank, unknown, upstream, official or official_upstream. The upstream version does not show which patches or backports the build contains. |
conflicting_versions |
unknown |
Several different versions for one runtime, or too many candidates. |
A vendor or custom build source is checked before the end-of-life rule, so such builds stay unknown. They need the vendor's own evidence. Every result also records evidence_type: operator_reported_version and an evidence_confidence of low, or insufficient when the status is unknown.
Exit codes describe the run, not the assessment: a completed run exits 0 even when it reports findings, unless you ask for --fail-on-finding. See Limits and failure behavior.
Each result has these fields: input_label, input_source (version, inventory or csv), node_version, build_source, status, version_status, fixed_version, exploitability, evidence_type, evidence_confidence, cve, rules_as_of, one precondition_* field per observation, recommendation, notes, sources and report_schema_version (currently 1.0).
- JSON is an object with
tool,tool_version,report_schema_version,cve,rules_as_of,sources,complete,generated_atandresults. - CSV has one row per result with
record_type=result. When no result exists, onerecord_type=summaryrow preservesreport_completeand the rule metadata, with no asset or verdict. Every field is quoted, and values that start with=,+,-or@after leading whitespace are neutralized against spreadsheet formula execution. Values containing NUL characters are rejected before a report is written. Existing column positions are unchanged. - HTML is a script-free, self-contained page with a restrictive content security policy and escaped values. It includes the fixed-release table and a reminder to check the binary in each serving process.
--brandchanges only the organization name text.
CSV column order
record_type, input_label, input_source, node_version, build_source, status, version_status, fixed_version, exploitability, evidence_type, evidence_confidence, cve, rules_as_of, precondition_permission_model_enabled, precondition_trace_events, precondition_filesystem_write_allowlist, recommendation, notes, sources, report_schema_version, report_complete, tool, tool_version, generated_at
Provenance is attached to every report. CSV appends tool, tool_version and generated_at to every result and to the empty summary row. One timezone-aware UTC generation timestamp is captured per run and reused when an interruption triggers partial-report recovery, across CSV, JSON and HTML. That timestamp is separate from the advisory rule date (rules_as_of). It is not an evidence-acquisition time or a deployed-patch attestation.
Keep real inventories, targets, reports and secrets outside version control. Synthetic examples are in examples/ and docs/example-report.html.
Inputs must be regular UTF-8 files. A run is limited to:
- 4 MiB per file
- 100 input files in total, across
--inventoryand--csv-list - 10,000 entries in total, including
--versionvalues - 512 characters per text field
Reports are rendered and encoded in memory, then published atomically to files. Use stable local copies of your inputs: file identity checks are not a sandbox against concurrent adversarial mutation.
| Exit code | Meaning |
|---|---|
| 0 | The run completed. |
| 1 | --fail-on-finding was given and a result was potentially_affected or unsupported_runtime. |
| 2 | Usage, input, output or cleanup failure (an operational error, not an assessment result). This does not establish whether report bytes were already written; inspect the output before retrying. |
| 130 | Interrupted. |
Interruption (exit 130) behaves as follows:
- Before the output path is validated, nothing is written.
- During assessment, completed results are written with
complete=falsein JSON,report_complete=falsein CSV, and an interrupted-run notice in HTML. Rerun the inputs that are missing. - After a complete report has been published, it is preserved and not retried.
- If publication to a file is indeterminate, the checker does not retry. Inspect the output before retrying; the diagnostic does not establish whether publication succeeded.
- During stdout output, the document may be truncated, and a second report is never appended.
- A second interrupt during recovery means the partial report may not have been written.
- A version string is operator-reported evidence, not verified runtime state. The checker cannot see what a process is actually running.
fixed_version_reportedis a reported version at or above the upstream fix. It is not a verified deployed patch. Confirm that the running process or container uses that release, not just a patched workstation or lockfile.- Version metadata is not a patch or build signature. Vendor, backported, custom and prerelease builds need vendor evidence and remain
unknown. - The vendor does not establish an introduced version, so
potentially_affectedis a prompt for remediation review, not confirmation of impact. - Exploitability is never tested. Configuration observations and version results do not show reachability.
- End-of-life status follows the Node.js release schedule, reviewed 2026-10-24. Major 0 is treated as end-of-life by inference, because the schedule lists only 0.8, 0.10 and 0.12 individually.
- The rule snapshot covers majors 22, 24 and 26 only. Other active or future majors stay
unknownuntil a reviewed rule exists.
Primary sources for the rule snapshot:
- https://nodejs.org/en/blog/vulnerability/july-2026-security-releases
- https://github.com/nodejs/security-wg/blob/main/vuln/core/185.json
- https://nodejs.org/en/blog/release/v22.23.2
- https://nodejs.org/en/blog/release/v24.18.1
- https://nodejs.org/en/blog/release/v26.5.1
End-of-life status is taken from the Node.js release schedule, reviewed 2026-10-24. The checker does not fetch it at runtime.
The embedded version engine is copied from node-cve-2026-105481-check commit de3161c4a8a7f7ca5c9e531df43b94e2805b52fd. No companion repository is required at runtime.
| Path | Contents |
|---|---|
node_cve_2026_56847_check.py |
The checker. |
examples/inventory.example.json |
Synthetic JSON inventory. |
docs/example-report.html |
Synthetic example report. Example data only. |
docs/REFERENCE.md |
CLI and evidence reference. |
docs/IMPLEMENTATION.md |
Implementation and verification status. |
advisory.json |
Source-preserving advisory metadata for the offline rule. |
test_*.py, .github/workflows/tests.yml |
Synthetic local-only tests and the CI workflow. |
Implementation status: the offline rule is implemented and has had local and independent review against the primary sources. No deployed-target assessment, runtime attestation or exploit verification is claimed. Release also depends on hosted CI for the exact published commit; see docs/IMPLEMENTATION.md.
Open a focused issue and pull request, use synthetic fixtures, and do not contact real targets in tests. Before submitting, run python -m unittest -v, python node_cve_2026_56847_check.py --help and git diff --check. Verify current primary sources before changing rules, keep unknown evidence uncertain, and keep real inventories, reports and secrets out of commits. Hosted CI must pass for the exact reviewed head. See CONTRIBUTING.md, which also sets out the contribution license grant.
To report a flaw in this project, follow SECURITY.md. Report vulnerabilities in Node.js itself through the Node.js project's own process.
This project is available under the GNU AGPL-3.0 (LICENSE) or a separate commercial license from CompleteTech LLC; see LICENSING.md. CompleteTech LLC names, logos and the artwork in docs/images/ are brand assets and are not covered by either code license; see BRAND_ASSETS.md. The report previews are screenshots of synthetic example data.
Copyright © 2026 CompleteTech LLC.
