vuln_scan


What it does	Host vulnerability scanning — CVE matching and configuration compliance checks
Version	1.0.0
Kind	Collector · read-only · gathered (security.vuln_scan.scan, security.vuln_scan.cve_scan, security.vuln_scan.config_scan, security.vuln_scan.summary, security.vuln_scan.inventory)
Platforms	Windows ✅ · macOS ✅ · Linux ✅
Actions	`config_scan` (definition `security.vuln_scan.config_scan`) · `cve_scan` (definition `security.vuln_scan.cve_scan`) · `inventory` (definition `security.vuln_scan.inventory`) · `scan` (definition `security.vuln_scan.scan`) · `summary` (definition `security.vuln_scan.summary`)
Security	`scan`: securable `Security` · operation Read · risk Low · dispatch ReadOnly · approval gate None; `cve_scan`: securable `Security` · operation Read · risk Low · dispatch ReadOnly · approval gate None; `config_scan`: securable `Security` · operation Read · risk Low · dispatch ReadOnly · approval gate None; `summary`: securable `Security` · operation Read · risk Low · dispatch ReadOnly · approval gate None; `inventory`: securable `Inventory` · operation Read · risk Low · dispatch ReadOnly · approval gate None
Roles	execute: endpoint-admin, endpoint-operator, security-admin · author: content-author

How it works

Every action is a read; nothing on the host is changed. scan first reads the installed-software list and matches it against a fixed table of 41 built-in CVE rules, then runs the fixed list of configuration checks for the operating system, and returns the CVE rows followed by the configuration rows (progress is reported at 0, 50, 90 and 100). cve_scan and config_scan are those two halves on their own. summary runs the same combined scan, throws the individual rows away and returns only a count per severity. inventory runs neither scan: it returns the raw installed-software name|version pairs and nothing else.

The CVE half is deliberately small and local. It is not a vulnerability feed: the rule table ships inside the plugin, each rule names a product by a short text (for example openssl or git), and a rule fires when an installed product’s name contains that text, ignoring case, and its version compares lower than the rule’s fixed version. The configuration half reads a handful of well-known hardening settings (registry values on Windows, command output and a few files on macOS and Linux) and reports every check it ran, passing ones as INFO. inventory exists so a caller can hand the raw software list to the server for a fuller match instead of relying on the built-in table.

flowchart LR
  OP["Operator or workflow"] --> SRV["Server: authz Security.Read, or Inventory.Read for inventory"]
  SRV -- "gRPC mTLS" --> HOST["Agent plugin host"] --> EX["vuln_scan.execute"]
  EX --> WIN["Windows leg: registry reads"]
  EX --> MAC["macOS leg: system_profiler, brew, system commands"]
  EX --> LIN["Linux leg: package manager, /proc, sshd_config, firewall tools"]
  WIN --> ROWS["Pipe-delimited rows, result status UNDECLARED"]
  MAC --> ROWS
  LIN --> ROWS
  ROWS --> RS[("ResponseStore")] --> API["REST /api/responses, MCP"]

What you can decide with it

Question	Action	What you get
Which known-bad versions of a short list of common products are installed?	`cve_scan`	Windows: from the registry Uninstall lists (machine-wide, 32-bit, and the agent account’s own). macOS: apps from `system_profiler` plus the packages `brew list --versions` reports (formulae, and casks if present). Linux: packages from the first package manager found (`dpkg-query`, else `rpm`, else `pacman`, else `apk`).
Is this host’s basic hardening in place?	`config_scan`	A fixed per-OS list of checks, listed below. Every check that ran is reported, passing ones as `INFO`.
Both of the above in one call	`scan`	CVE rows, then configuration rows.
How bad is it at a glance?	`summary`	`TOTAL` first, then a count for each of `CRITICAL`, `HIGH`, `MEDIUM`, `LOW`, `INFO`, `UNREADABLE`. `UNREADABLE` and `INFO` are not issues.
What software is installed, for matching elsewhere?	`inventory`	Raw `name\|version` pairs. No CVE or configuration logic.

What a clean result means. A scan with no cve rows says only that none of the 41 built-in rules matched the installed software. It does not say the host is patched. The table covers 21 product names (OpenSSL, curl, sudo, polkit, Log4j, Apache, OpenSSH, Python, Node.js, Chrome, Firefox, .NET, OpenJDK, Windows, nginx, PostgreSQL, Git, 7-Zip, WinRAR, PuTTY, PHP) with one to four rules each, and no rule is for a CVE newer than 2024. Software that is not in the software list the leg reads (per-user installs of other accounts on Windows, anything on Linux when none of the four package managers is present) is never examined. Operating-system patch level, kernel, firmware and distribution back-ports are not assessed at all. For a real patch-compliance answer use a patch-state source, not this plugin.

What a cve row means. It is a lead to check, not a confirmed exposure. Matching is by substring of the product name and by a simple dotted-number version comparison, so some rows are false positives (see Caveat 2). The two Windows Print Spooler rules in particular compare an application’s own version against a Windows build number and never read the real OS build or installed hotfixes, so they cannot tell you whether PrintNightmare is fixed; in the Windows sample every Print Spooler row names a .NET runtime, a Windows update package or a Windows utility, not the spooler itself.

Configuration checks by OS (severity is fixed per check; a passing check is INFO):

OS	Checks that always report	Checks that report only in some cases	Known ways the answer can be wrong
Windows	UAC (`EnableLUA`), SMBv1 (`SMB1`), Auto-Logon, Defender real-time flag (`DisableRealtimeMonitoring`), Firewall for the Domain, Private and Public profiles	RDP Network Level Authentication, only when Remote Desktop is enabled	A missing SMBv1 value is read as enabled (`CRITICAL`); a missing Defender flag is read as enabled protection; the Defender row is one registry flag, not the live service state; an unreadable UAC or firewall value reads as disabled.
macOS	Gatekeeper (`spctl`), FileVault (`fdesetup`), SIP (`csrutil`), Application Firewall (`socketfilterfw`), Remote Login (`systemsetup`), Automatic Software Updates (`defaults`)	none	`systemsetup` needs administrator rights; without them its error text lacks the word `Off`, so Remote Login reads `Enabled`. The updates row treats any `1` in the `defaults` output as enabled, including inside an error message. The risky-port check is not expected to produce rows: the plugin runs `ss -tlnH
Linux	ASLR (`randomize_va_space`), SUID core dumps (`suid_dumpable`), SSH Root Login, Firewall	SSH Password Authentication, only if `sshd_config` sets it (it also reports `INFO` “not applicable” when the file is absent and `UNREADABLE` when it cannot be read); `/tmp noexec`, only if `/tmp` is a separate mount; World-Writable PATH, only if the agent’s own `PATH` has such a directory; Open Port rows for FTP 21, Telnet 23, SMB 445, RDP 3389, VNC 5900 bound on all interfaces	Only `/etc/ssh/sshd_config` is read (no `Include`d files, no `Match` blocks, no effective-config query); a plain `PermitRootLogin prohibit-password` reads as “not explicitly set”. The firewall check reads `iptables`, `nft` and `ufw` output, which normally need root, so an unprivileged agent reports `No firewall rules detected` (`HIGH`) even on a protected host.

A Linux file the agent cannot read (SSH config, ASLR, suid_dumpable, the mount table) is reported as UNREADABLE, never as a weak setting. An absent sshd_config is INFO, “not applicable”.

Diagrams

flowchart TD
  S["scan"] --> INV["Read installed software: name and version"]
  INV --> M{"Does a rule's product text appear in the name?"}
  M -- "no" --> N1["No row"]
  M -- "yes" --> V{"Is the installed version below the rule's fixed version?"}
  V -- "no" --> N1
  V -- "yes" --> CVE["cve row, severity from the rule"]
  S --> CFG["Run the fixed checks for this OS"]
  CFG --> R{"Could the check read its input?"}
  R -- "yes" --> F["INFO if passing, else the check's fixed severity"]
  R -- "no, Linux file read" --> U["UNREADABLE row, not an issue"]
  R -- "no, registry or command" --> D["Check's default branch, no error"]

The CVE branch and the configuration branch are independent: neither can suppress the other, and scan simply concatenates them. A software entry with no version is skipped, so it can never match. The last two configuration branches are the important difference: a failed Linux file read is surfaced as UNREADABLE, while a failed registry or command read silently falls through to that check’s default result.

flowchart LR
  ROWS["scan rows: cve and config"] --> CNT["Count rows per severity"]
  CNT --> TOT["TOTAL: all rows, and issues = rows that are not INFO or UNREADABLE"]
  CNT --> PER["One row each: CRITICAL, HIGH, MEDIUM, LOW, INFO, UNREADABLE"]

summary counts rows, not distinct problems. One installed product that matches two rules, or several installed products that match one rule, each add to the issue count, so the number is an upper bound on real exposures and is inflated by false-positive cve rows.

OS capability

Action	Windows	macOS	Linux
`config_scan`	✅ supported · rung 1 · win32_registry	✅ supported · rung 3 · popen(spctl/fdesetup/csrutil/socketfilterfw)	✅ supported · rung 3 · popen(iptables/nft/ufw)+procfs
`cve_scan`	✅ supported · rung 1 · win32_registry	✅ supported · rung 3 · popen(system_profiler/brew)	✅ supported · rung 3 · popen(dpkg-query/rpm/pacman/apk)
`inventory`	✅ supported · rung 1 · win32_registry	✅ supported · rung 3 · popen(system_profiler/brew)	✅ supported · rung 3 · popen(dpkg-query/rpm/pacman/apk)
`scan`	✅ supported · rung 1 · win32_registry	✅ supported · rung 3 · popen(system_profiler/brew+security-checks)	✅ supported · rung 3 · popen(pkg-manager+iptables/nft/ufw)
`summary`	✅ supported · rung 1 · win32_registry	✅ supported · rung 3 · popen(system_profiler/brew+security-checks)	✅ supported · rung 3 · popen(pkg-manager+iptables/nft/ufw)

Privileges and prerequisites

OS	Runs as	Extra grant needed	Measured	If the read is refused
Windows	agent daemon identity (captured as `SYSTEM`)	None established; the plugin reads `HKLM` Uninstall and policy values and the agent account’s own `HKCU` Uninstall list. No row in `docs/agent-privilege-model.md`	2026-09-07, bare metal, as SYSTEM	not observed; a denied `RegOpenKeyExW` or `RegQueryValueExW` returns an empty string or the check’s default, so a blocked read shows up as that check’s default result, not an error
macOS	agent daemon; the shipped LaunchDaemon runs as root today (`docs/agent-privilege-model.md`)	`systemsetup -getremotelogin` needs administrator rights (root has them). No other grant. No row for this plugin	2026-09-07, bare metal, as an unprivileged user (euid 501), not as root	not observed for a denial; a failed command returns empty or error text, which each check parses as its negative or disabled branch (Remote Login reads `Enabled` when `systemsetup` is refused)
Linux	agent daemon, dedicated unprivileged account by design (`docs/agent-privilege-model.md`)	None for the package list, `/proc` or world-readable files. `iptables`, `nft` and `ufw status` normally need root or `CAP_NET_ADMIN`; a mode-0600 `sshd_config` (the RHEL-family default) needs read access. No row for this plugin	2026-09-06, container, euid 0	a failed file read is `UNREADABLE` (both SSH checks on a 0600 `sshd_config`); a failed firewall command returns empty output, read as “no firewall rules detected”

Windows uses no subprocess: software enumeration and every Windows check are in-process registry calls. Linux and macOS run commands through popen and system() with fixed strings and no caller input: Linux dpkg-query, rpm, pacman, apk (the first one present), iptables, nft, ufw, ss or netstat; macOS system_profiler, brew when installed, spctl, fdesetup, csrutil, socketfilterfw, systemsetup, defaults. There is no network access; the firewall and listener commands read local state only.

Data contract

Inputs

No action takes parameters.

Outputs

Pipe-delimited rows, one per finding (or per installed package for inventory). scan, cve_scan and config_scan share the 4-field row shape severity|category|title|detail. An action that finds nothing still emits one placeholder row, INFO|scan|No vulnerabilities|No issues detected; in practice only cve_scan reaches it, because config_scan and scan always report at least one configuration check. summary emits a different, 3-field shape: a literal summary discriminator, then a severity name (or the literal TOTAL), then a count, which is one field more than the two result columns (severity, count) the definition declares. inventory emits raw name|version pairs with no discriminator, and version is empty when the source gives none.

A Linux config check whose file read fails (EACCES, EIO, …) emits UNREADABLE|config|<title>|<path>: <cause> (for example eacces) rather than a finding about an empty value. UNREADABLE means the check could not run; it is not an issue, so summary leaves it out of the TOTAL row’s issue count and reports it on its own summary|UNREADABLE|<n> row after summary|INFO|<n>. An absent /etc/ssh/sshd_config is not a weak setting: both SSH checks emit INFO with “/etc/ssh/sshd_config not present; check not applicable”.

security.vuln_scan.config_scan — severity|category|title|detail

Field	Type	Values	Available	Example	Description
`severity`	string	`CRITICAL` `HIGH` `MEDIUM` `LOW` `INFO` `UNREADABLE`	Windows, Linux, macOS	`HIGH`	Finding severity assigned by the compliance check that produced this row. UNREADABLE means the check could not run; not an issue.
`category`	string	`config` `scan`	Windows, Linux, macOS	`config`	Always “config” for this action; “scan” only on the placeholder row emitted when there is nothing to report.
`title`	string	-	Windows, Linux, macOS	`SMBv1 Protocol`	The name of the compliance check, e.g. the setting it inspects. Values: free text.
`detail`	string	-	Windows, Linux, macOS	`Enabled - vulnerable to EternalBlue/WannaCry (MS17-010)`	The check’s human-readable result text, including why a failing check failed. Values: free text.

security.vuln_scan.cve_scan — severity|category|title|detail

Field	Type	Values	Available	Example	Description
`severity`	string	`CRITICAL` `HIGH` `MEDIUM` `LOW` `INFO`	Windows, Linux, macOS	`HIGH`	Finding severity assigned by the CVE rule that matched.
`category`	string	`cve` `scan`	Windows, Linux, macOS	`cve`	Always “cve” for a match; “scan” on the single placeholder row emitted when nothing matches.
`title`	string	-	Windows, Linux, macOS	`CVE-2023-24329: urllib.parse URL parsing bypass via leading whitespace`	”: ” from the matched rule, or “No vulnerabilities” on the placeholder row. Values: free text.
`detail`	string	-	Windows, Linux, macOS	`Python 3.9.6 (fixed in 3.11.4)`	(fixed in <fixed_ver>). Values: free text.

security.vuln_scan.inventory — name|version

Field	Type	Values	Available	Example	Description
`name`	string	-	Windows, Linux, macOS	`curl`	Installed application or package name, as reported by the OS’s own inventory source. Values: free text.
`version`	string	-	Windows, Linux, macOS	`8.14.1-2+deb13u4`	Installed version string, or empty when the source reports none (e.g. a Windows uninstall entry with no DisplayVersion). Values: free text, or empty.

security.vuln_scan.scan — severity|category|title|detail

Field	Type	Values	Available	Example	Description
`severity`	string	`CRITICAL` `HIGH` `MEDIUM` `LOW` `INFO` `UNREADABLE`	Windows, Linux, macOS	`CRITICAL`	Finding severity assigned by the check that produced this row. UNREADABLE means the check could not run; not an issue.
`category`	string	`cve` `config` `scan`	Windows, Linux, macOS	`cve`	Which half of the scan produced this row.
`title`	string	-	Windows, Linux, macOS	`CVE-2021-34527: PrintNightmare: RCE via Windows Print Spooler`	For a cve row, ”: ” from the matched rule; for a config row, the name of the compliance check; “No vulnerabilities” on the single placeholder row emitted when a run finds nothing. Values: free text.
`detail`	string	-	Windows, Linux, macOS	`Microsoft Windows Desktop Runtime - 6.0.11 (x64) 6.0.11.31823 (fixed in KB5004945)`	For a cve row, ” (fixed in <fixed_ver>)”; for a config row, the check’s human-readable result text. Values: free text.

security.vuln_scan.summary — severity|count

Field	Type	Values	Available	Example	Description
`severity`	string	`TOTAL` `CRITICAL` `HIGH` `MEDIUM` `LOW` `INFO` `UNREADABLE`	Windows, Linux, macOS	`CRITICAL`	The severity this row counts, or the literal “TOTAL” for the first row emitted by every run.
`count`	int32	-	Windows, Linux, macOS	`17`	The number of findings at this severity. The TOTAL row is an exception: despite the declared int32 type, that row’s cell holds the formatted string ” findings ( issues)”, not an integer. Values: integer, except a formatted string on the TOTAL row.

Result status

Status	Completeness	Provenance	When
`UNDECLARED`	`UNKNOWN`	(empty)	every action on every OS: the plugin does not set a typed result status, so a consumer must read the rows. A failed read is therefore never visible in the status; it shows only as an `UNREADABLE` row (Linux file checks) or as a default-branch finding

Where the data goes

Instruction result only. Rows travel over the agent’s mTLS gRPC channel as the command response and land in the ResponseStore, rendered by the dashboard and REST with the plugin’s default column schema (Agent, Severity, Category, Title, Detail) or, when the request carries a definition id, that definition’s own result-column schema.
Server-side NVD matching. The inventory definition says its rows are for server-side NVD matching and centralized correlation; this README does not describe that server feature. The server’s version comparison notes that it duplicates the algorithm in cve_rules.hpp.
Not consumed by daily-sync inventory, Echo, DEX or metrics; nothing runs on a schedule and every action is on-demand.
Sensitivity. cve rows embed installed application names and versions in detail, and inventory returns that list directly: an installed-software inventory by another route. Configuration rows reveal the host’s security posture (for example a disabled firewall or SMBv1 enabled), which is useful to an attacker as well as an administrator. Nothing in either row shape names a person.
Siblings: installed_apps is a separate installed-software collector with overlapping data (it shares the “installed, held” package-presence convention on Linux); a caller working from software identity may have two plugins to reconcile.
MCP / REST. Discover: discover_plugins (summary) then yuzu://plugin-docs (this page as data) then discover_instructions or get_definition("security.vuln_scan.scan"). Run: execute_instruction {definition_id, parameters}. Read: /api/responses/{id}.

Sample output

Windows — captured: windows Microsoft Windows NT 10.0.26200.0 x64 · bare-metal · 2026-09-07 · SYSTEM · leg-hash 20c66f1c0618

== action=scan
CRITICAL|cve|CVE-2021-34527: PrintNightmare: RCE via Windows Print Spooler|Microsoft Windows Desktop Runtime - 6.0.11 (x64) 6.0.11.31823 (fixed in KB5004945)
CRITICAL|cve|CVE-2021-34527: PrintNightmare: RCE via Windows Print Spooler|Microsoft Windows Desktop Runtime - 6.0.20 (x86) 6.0.20.32621 (fixed in KB5004945)
CRITICAL|cve|CVE-2021-34527: PrintNightmare: RCE via Windows Print Spooler|Microsoft Windows Desktop Runtime - 8.0.28 (x64) 8.0.28.36119 (fixed in KB5004945)
CRITICAL|cve|CVE-2021-34527: PrintNightmare: RCE via Windows Print Spooler|Microsoft Windows Desktop Runtime - 8.0.28 (x86) 8.0.28.36119 (fixed in KB5004945)
CRITICAL|cve|CVE-2021-34527: PrintNightmare: RCE via Windows Print Spooler|Update for x64-based Windows Systems (KB5001716) 8.94.0.0 (fixed in KB5004945)
CRITICAL|cve|CVE-2021-34527: PrintNightmare: RCE via Windows Print Spooler|Windows 11 Installation Assistant 1.4.19041.5003 (fixed in KB5004945)
CRITICAL|cve|CVE-2021-34527: PrintNightmare: RCE via Windows Print Spooler|Windows PC Health Check 3.6.2204.08001 (fixed in KB5004945)
CRITICAL|cve|CVE-2021-34527: PrintNightmare: RCE via Windows Print Spooler|Windows Subsystem for Linux 2.6.3.0 (fixed in KB5004945)
CRITICAL|cve|CVE-2021-1675: Print Spooler privilege escalation|Microsoft Windows Desktop Runtime - 6.0.11 (x64) 6.0.11.31823 (fixed in KB5003637)
CRITICAL|cve|CVE-2021-1675: Print Spooler privilege escalation|Microsoft Windows Desktop Runtime - 6.0.20 (x86) 6.0.20.32621 (fixed in KB5003637)
CRITICAL|cve|CVE-2021-1675: Print Spooler privilege escalation|Microsoft Windows Desktop Runtime - 8.0.28 (x64) 8.0.28.36119 (fixed in KB5003637)
CRITICAL|cve|CVE-2021-1675: Print Spooler privilege escalation|Microsoft Windows Desktop Runtime - 8.0.28 (x86) 8.0.28.36119 (fixed in KB5003637)
… 12 of 24 rows shown
[result_status] UNDECLARED / UNKNOWN

== action=cve_scan
CRITICAL|cve|CVE-2021-34527: PrintNightmare: RCE via Windows Print Spooler|Microsoft Windows Desktop Runtime - 6.0.11 (x64) 6.0.11.31823 (fixed in KB5004945)
CRITICAL|cve|CVE-2021-34527: PrintNightmare: RCE via Windows Print Spooler|Microsoft Windows Desktop Runtime - 6.0.20 (x86) 6.0.20.32621 (fixed in KB5004945)
CRITICAL|cve|CVE-2021-34527: PrintNightmare: RCE via Windows Print Spooler|Microsoft Windows Desktop Runtime - 8.0.28 (x64) 8.0.28.36119 (fixed in KB5004945)
CRITICAL|cve|CVE-2021-34527: PrintNightmare: RCE via Windows Print Spooler|Microsoft Windows Desktop Runtime - 8.0.28 (x86) 8.0.28.36119 (fixed in KB5004945)
CRITICAL|cve|CVE-2021-34527: PrintNightmare: RCE via Windows Print Spooler|Update for x64-based Windows Systems (KB5001716) 8.94.0.0 (fixed in KB5004945)
CRITICAL|cve|CVE-2021-34527: PrintNightmare: RCE via Windows Print Spooler|Windows 11 Installation Assistant 1.4.19041.5003 (fixed in KB5004945)
CRITICAL|cve|CVE-2021-34527: PrintNightmare: RCE via Windows Print Spooler|Windows PC Health Check 3.6.2204.08001 (fixed in KB5004945)
CRITICAL|cve|CVE-2021-34527: PrintNightmare: RCE via Windows Print Spooler|Windows Subsystem for Linux 2.6.3.0 (fixed in KB5004945)
CRITICAL|cve|CVE-2021-1675: Print Spooler privilege escalation|Microsoft Windows Desktop Runtime - 6.0.11 (x64) 6.0.11.31823 (fixed in KB5003637)
CRITICAL|cve|CVE-2021-1675: Print Spooler privilege escalation|Microsoft Windows Desktop Runtime - 6.0.20 (x86) 6.0.20.32621 (fixed in KB5003637)
CRITICAL|cve|CVE-2021-1675: Print Spooler privilege escalation|Microsoft Windows Desktop Runtime - 8.0.28 (x64) 8.0.28.36119 (fixed in KB5003637)
CRITICAL|cve|CVE-2021-1675: Print Spooler privilege escalation|Microsoft Windows Desktop Runtime - 8.0.28 (x86) 8.0.28.36119 (fixed in KB5003637)
… 12 of 16 rows shown
[result_status] UNDECLARED / UNKNOWN

== action=config_scan
INFO|config|UAC (User Account Control)|Enabled
CRITICAL|config|SMBv1 Protocol|Enabled - vulnerable to EternalBlue/WannaCry (MS17-010)
INFO|config|Auto-Logon|Disabled
INFO|config|RDP Network Level Authentication|Enabled
INFO|config|Windows Defender Real-Time Protection|Enabled
INFO|config|Windows Firewall|Domain profile: Enabled
INFO|config|Windows Firewall|Private profile: Enabled
INFO|config|Windows Firewall|Public profile: Enabled
[result_status] UNDECLARED / UNKNOWN

== action=summary
summary|TOTAL|24 findings (17 issues)
summary|CRITICAL|17
summary|HIGH|0
summary|MEDIUM|0
summary|LOW|0
summary|INFO|7
[result_status] UNDECLARED / UNKNOWN

== action=inventory
7-Zip 26.02 (x64)|26.02
Age of Empires II: Definitive Edition|
Application Verifier x64 External Package (DesktopEditions)|10.1.26100.7705
Application Verifier x64 External Package (OnecoreUAP)|10.1.26100.7705
Baldur's Gate 3|
Battle.net|
CCleaner 7|7.10.1464.1889
CMake|4.3.3
Cities: Skylines II|
Crusader Kings III|
DayZ|
Defraggler|2.22
… 12 of 226 rows shown
[result_status] UNDECLARED / UNKNOWN

macOS — captured: macos macOS 26.6.2 arm64 · bare-metal · 2026-09-07 · euid 501 (jsmith) · leg-hash 20c66f1c0618

== action=scan
HIGH|cve|CVE-2023-24329: urllib.parse URL parsing bypass via leading whitespace|Python 3.9.6 (fixed in 3.11.4)
MEDIUM|cve|CVE-2024-0450: zipfile quoted-overlap zipbomb protection bypass|Python 3.9.6 (fixed in 3.12.2)
INFO|config|Gatekeeper|Enabled - only verified apps can run
INFO|config|FileVault Disk Encryption|Enabled
INFO|config|System Integrity Protection (SIP)|Enabled
MEDIUM|config|Application Firewall|Disabled
MEDIUM|config|Remote Login (SSH)|Enabled - SSH access is open
INFO|config|Automatic Software Updates|Enabled
[result_status] UNDECLARED / UNKNOWN

== action=cve_scan
HIGH|cve|CVE-2023-24329: urllib.parse URL parsing bypass via leading whitespace|Python 3.9.6 (fixed in 3.11.4)
MEDIUM|cve|CVE-2024-0450: zipfile quoted-overlap zipbomb protection bypass|Python 3.9.6 (fixed in 3.12.2)
[result_status] UNDECLARED / UNKNOWN

== action=config_scan
INFO|config|Gatekeeper|Enabled - only verified apps can run
INFO|config|FileVault Disk Encryption|Enabled
INFO|config|System Integrity Protection (SIP)|Enabled
MEDIUM|config|Application Firewall|Disabled
MEDIUM|config|Remote Login (SSH)|Enabled - SSH access is open
INFO|config|Automatic Software Updates|Enabled
[result_status] UNDECLARED / UNKNOWN

== action=summary
summary|TOTAL|8 findings (4 issues)
summary|CRITICAL|0
summary|HIGH|1
summary|MEDIUM|3
summary|LOW|0
summary|INFO|4
[result_status] UNDECLARED / UNKNOWN

== action=inventory
App Store|3.0
Apps|1.0
Automator|2.10
Books|8.5
Calculator|12.0
Calendar|16.0
Chess|3.18
Clock|1.1
Contacts|14.0
Dictionary|2.3.0
FaceTime|36
Find My|4.0
… 12 of 387 rows shown
[result_status] UNDECLARED / UNKNOWN

Linux — captured: linux Debian GNU/Linux 13 (trixie) aarch64 · container · 2026-09-06 · euid 0 · leg-hash 20c66f1c0618

== action=scan
HIGH|cve|CVE-2023-24329: urllib.parse URL parsing bypass via leading whitespace|python3-wheel 0.46.1-2 (fixed in 3.11.4)
MEDIUM|cve|CVE-2024-0450: zipfile quoted-overlap zipbomb protection bypass|python3-wheel 0.46.1-2 (fixed in 3.12.2)
CRITICAL|cve|CVE-2024-32002: RCE via crafted repositories with submodules|git 1:2.47.3-0+deb13u1 (fixed in 2.45.1)
CRITICAL|cve|CVE-2024-32002: RCE via crafted repositories with submodules|git-man 1:2.47.3-0+deb13u1 (fixed in 2.45.1)
HIGH|cve|CVE-2023-25652: git apply --reject writes outside worktree|git 1:2.47.3-0+deb13u1 (fixed in 2.40.1)
HIGH|cve|CVE-2023-25652: git apply --reject writes outside worktree|git-man 1:2.47.3-0+deb13u1 (fixed in 2.40.1)
MEDIUM|config|SSH Root Login|PermitRootLogin not explicitly set - may default to prohibit-password
INFO|config|ASLR (Address Space Layout Randomization)|Full randomization enabled (value=2)
INFO|config|SUID Core Dumps|Restricted (suid_dumpable=0)
HIGH|config|Firewall|No firewall rules detected - host may be unprotected
[result_status] UNDECLARED / UNKNOWN

== action=cve_scan
HIGH|cve|CVE-2023-24329: urllib.parse URL parsing bypass via leading whitespace|python3-wheel 0.46.1-2 (fixed in 3.11.4)
MEDIUM|cve|CVE-2024-0450: zipfile quoted-overlap zipbomb protection bypass|python3-wheel 0.46.1-2 (fixed in 3.12.2)
CRITICAL|cve|CVE-2024-32002: RCE via crafted repositories with submodules|git 1:2.47.3-0+deb13u1 (fixed in 2.45.1)
CRITICAL|cve|CVE-2024-32002: RCE via crafted repositories with submodules|git-man 1:2.47.3-0+deb13u1 (fixed in 2.45.1)
HIGH|cve|CVE-2023-25652: git apply --reject writes outside worktree|git 1:2.47.3-0+deb13u1 (fixed in 2.40.1)
HIGH|cve|CVE-2023-25652: git apply --reject writes outside worktree|git-man 1:2.47.3-0+deb13u1 (fixed in 2.40.1)
[result_status] UNDECLARED / UNKNOWN

== action=config_scan
MEDIUM|config|SSH Root Login|PermitRootLogin not explicitly set - may default to prohibit-password
INFO|config|ASLR (Address Space Layout Randomization)|Full randomization enabled (value=2)
INFO|config|SUID Core Dumps|Restricted (suid_dumpable=0)
HIGH|config|Firewall|No firewall rules detected - host may be unprotected
[result_status] UNDECLARED / UNKNOWN

== action=summary
summary|TOTAL|10 findings (8 issues)
summary|CRITICAL|2
summary|HIGH|4
summary|MEDIUM|2
summary|LOW|0
summary|INFO|2
[result_status] UNDECLARED / UNKNOWN

== action=inventory
apt|3.0.3
autoconf|2.72-3.1
automake|1:1.17-4
autotools-dev|20240727.1
base-files|13.8+deb13u6
base-passwd|3.6.7
bash|5.2.37-2+b9
binutils|2.44-3
binutils-aarch64-linux-gnu|2.44-3
binutils-common|2.44-3
bison|2:3.8.2+dfsg-1+b2
bsdutils|1:2.41.5-0+deb13u1
… 12 of 206 rows shown
[result_status] UNDECLARED / UNKNOWN

Caveats and known gaps

A clean scan is not “patched”. CVE matching uses a fixed, built-in table of 41 rules over 21 product names, and nothing outside it is reported: other products, CVEs after 2024, operating-system patches, kernels and distribution back-ports. Software the leg cannot list (other accounts’ per-user installs on Windows, any host with none of the four Linux package managers) is not examined.
Substring matching produces false positives, and the version test is naive. A rule fires on any installed name that contains its product text, so python matches python3-wheel, git matches git-man, and windows matches .NET runtimes and Windows utilities; the version is compared as dotted numbers, and a Debian epoch such as 1:2.47.3 is compared as text and reads as lower than 2.45.1. The samples show all three, so some cve rows are not exposures, and the issue count in summary is inflated by them.
A failed read usually becomes the check’s default result, not an error. Windows registry reads fall back to a default (a missing SMBv1 value reads as enabled; an unreadable UAC or firewall value reads as disabled), a failed popen command returns empty output, and only the Linux file checks surface UNREADABLE. No capture observed a real permission denial, so this is read from the code, not measured; the World-Writable PATH walk also skips an entry whose stat() fails (#5515).
summary does not match its declared result columns. It emits three pipe fields per row (summary|<severity or TOTAL>|<count>) against two declared columns, severity and count, and the TOTAL row’s count cell is the string <n> findings (<n> issues) rather than the declared int32.
Configuration checks are single-signal heuristics, not effective policy. Each reads one registry value, one command’s text or one file, so a setting applied another way (an Included SSH file, a managed firewall, a third-party antivirus) can read differently from the host’s real state. There is no docs/agent-privilege-model.md row for this plugin, so the Privileges table records only what each capture measured.

Source and tests

Plugin: agents/plugins/vuln_scan/src/config_checks.hpp · agents/plugins/vuln_scan/src/cve_rules.hpp · agents/plugins/vuln_scan/src/vuln_scan_plugin.cpp
Definitions: content/definitions/vuln_scan.yaml
Capability rows: server/core/src/capability_decls/plugin_action_catalogue_a.hpp
Tests: none found by name
Privilege row: docs/agent-privilege-model.md (no row yet)