Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Adversaries may attempt to get a listing of security software, configurations, defensive tools, and sensors that are installed on a system or in a cloud environment.
.claude/skills/cyberstrikeus-t1518-001-security-software-discovery/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 4 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-14 | ✗→✓ | ▲ Improved | 196% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 139% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 19% | 0% |
| case-01 | ✓→✓ | = Same ✓ | 258% | 0% |
| case-02 | ✓→✓ | = Same ✓ | 51% | 0% |
> Sub-technique of: T1518
Adversaries may attempt to get a listing of security software, configurations, defensive tools, and sensors that are installed on a system or in a cloud environment. This may include things such as cloud monitoring agents and anti-virus. Adversaries may use the information from Security Software Discovery during automated discovery to shape follow-on behaviors, including whether or not the adversary fully infects the target and/or attempts specific actions.
Example commands that can be used to obtain security software information are netsh, <code>reg query</code> with Reg, <code>dir</code> with cmd, and Tasklist, but other indicators of discovery behavior may be more specific to the type of software or security system the adversary is looking for. It is becoming more common to see macOS malware perform checks for LittleSnitch and KnockKnock software.
Adversaries may also utilize the Cloud API to discover cloud-native security software installed on compute infrastructure, such as the AWS CloudWatch agent, Azure VM Agent, and Google Cloud Monitor agent. These agents may collect metrics and logs from the VM, which may be centrally aggregated in a cloud-based monitoring platform.
Platforms: IaaS, Linux, macOS, Windows
The following tests are from Atomic Red Team and provide actionable ways to test this technique:
Methods to identify Security Software on an endpoint
when sucessfully executed, the test is going to display running processes, firewall configuration on network profiles and specific security software.
Supported Platforms: windows
cmdnetsh.exe advfirewall show allprofiles netsh.exe advfirewall firewall dump netsh.exe advfirewall show currentprofile netsh.exe advfirewall firewall show rule name=all netsh.exe firewall show state netsh.exe firewall show config sc query windefend powershell.exe /c "Get-Process | Where-Object { $_.ProcessName -eq 'Sysmon' }" powershell.exe /c "Get-Service | where-object {$_.DisplayName -like '*sysm*'}" powershell.exe /c "Get-CimInstance Win32_Service -Filter 'Description = ''System Monitor service'''" tasklist.exe tasklist.exe | findstr /i virus tasklist.exe | findstr /i cb tasklist.exe | findstr /i defender tasklist.exe | findstr /i cylance tasklist.exe | findstr /i mc tasklist.exe | findstr /i "virus cb defender cylance mc"
Methods to identify Security Software on an endpoint
when sucessfully executed, powershell is going to processes related AV products if they are running. Note that, depending on the privilege of current user, get-process | ?{$_.Description -like ""} may not return the processes related to AV products of the check. For instance, only with Administrator right, you can see the process description of McAffee processes. Hence, it is better to use get-process | ?{$\_.ProcessName -like "\"}, if you know the name of those processes.
Supported Platforms: windows
powershellget-process | ?{$_.Description -like "*virus*"} get-process | ?{$_.Description -like "*carbonblack*"} get-process | ?{$_.Description -like "*defender*"} get-process | ?{$_.Description -like "*cylance*"} get-process | ?{$_.Description -like "*mc*"} get-process | ?{$_.ProcessName -like "*mc*"} get-process | Where-Object { $_.ProcessName -eq "Sysmon" }
Methods to identify Security Software on an endpoint when sucessfully executed, command shell is going to display AV/Security software it is running.
Supported Platforms: macos
bashps aux | egrep 'Little\ Snitch|CbOsxSensorService|falcond|nessusd|santad|CbDefense|td-agent|packetbeat|filebeat|auditbeat|osqueryd|BlockBlock|LuLu'
Methods to identify Security Software on an endpoint when sucessfully executed, command shell is going to display AV/Security software it is running.
Supported Platforms: linux
bashps aux | egrep 'falcond|nessusd|cbagentd|td-agent|packetbeat|filebeat|auditbeat|osqueryd'
Methods to identify Security Software on an endpoint when sucessfully executed, command shell is going to display AV/Security software it is running.
Supported Platforms: linux
bashpgrep -l 'bareos-fd|icinga2|cbagentd|wazuh-agent|packetbeat|filebeat|osqueryd'
If Atomic Red Team tests are not applicable, manually verify the technique by:
No specific mitigations documented for this technique.
| Finding | Severity | Impact | | ------------------------------------------------ | -------- | --------- | | Security Software Discovery technique applicable | Medium | Discovery |
| CWE ID | Title | | ------- | --------------------------------- | | CWE-200 | Exposure of Sensitive Information |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→pass | 8,623 | 9,818 | +14% | 1 | 1 | 0% | 651 | 2,329 | +258% | 0 | 0 | — |
case-02 | pass→pass | 8,086 | 3,743 | -54% | 1 | 1 | 0% | 1,390 | 2,098 | +51% | 0 | 0 | — |
case-03 | fail→fail | 13,278 | 2,161 | -84% | 1 | 1 | 0% | 744 | 1,852 | +149% | 0 | 0 | — |
case-04 | pass→pass | 5,685 | 4,236 | -25% | 1 | 1 | 0% | 732 | 2,321 | +217% | 0 | 0 | — |
case-05 | pass→pass | 8,962 | 8,547 | -5% | 1 | 1 | 0% | 1,209 | 2,154 | +78% | 0 | 0 | — |
case-06 | pass→pass | 12,992 | 9,567 | -26% | 1 | 1 | 0% | 1,535 | 2,177 | +42% | 0 | 0 | — |
case-07 | pass→pass | 9,952 | 10,167 | +2% | 1 | 1 | 0% | 1,615 | 2,264 | +40% | 0 | 0 | — |
case-08 | pass→pass | 6,240 | 3,535 | -43% | 1 | 1 | 0% | 1,048 | 2,102 | +101% | 0 | 0 | — |
case-09 | pass→pass | 8,301 | 4,271 | -49% | 1 | 1 | 0% | 488 | 1,774 | +264% | 0 | 0 | — |
case-10 | pass→pass | 8,825 | 9,698 | +10% | 1 | 1 | 0% | 697 | 1,882 | +170% | 0 | 0 | — |
case-11 | pass→pass | 8,834 | 10,989 | +24% | 1 | 1 | 0% | 1,556 | 1,804 | +16% | 0 | 0 | — |
case-12 | pass→pass | 14,234 | 9,829 | -31% | 1 | 1 | 0% | 1,264 | 2,137 | +69% | 0 | 0 | — |
case-13 | pass→pass | 8,767 | 7,431 | -15% | 1 | 1 | 0% | 1,417 | 2,666 | +88% | 0 | 0 | — |
case-14 | fail→pass | 4,553 | 4,075 | -10% | 1 | 1 | 0% | 731 | 2,162 | +196% | 0 | 0 | — |
case-15 | fail→fail | 10,269 | 3,438 | -67% | 1 | 1 | 0% | 677 | 2,086 | +208% | 0 | 0 | — |
case-16 | fail→pass | 19,053 | 3,936 | -79% | 1 | 1 | 0% | 921 | 2,198 | +139% | 0 | 0 | — |
case-17 | pass→pass | 8,678 | 5,109 | -41% | 1 | 1 | 0% | 562 | 2,344 | +317% | 0 | 0 | — |
case-18 | pass→pass | 8,823 | 3,147 | -64% | 1 | 1 | 0% | 413 | 2,097 | +408% | 0 | 0 | — |
case-19 | pass→pass | 12,917 | 11,297 | -13% | 1 | 1 | 0% | 1,304 | 1,944 | +49% | 0 | 0 | — |
case-20 | fail→pass | 8,929 | 7,411 | -17% | 1 | 1 | 0% | 1,624 | 1,935 | +19% | 0 | 0 | — |
case-21 | pass→pass | 46,826 | 13,842 | -70% | 1 | 1 | 0% | 7,372 | 2,792 | -62% | 0 | 0 | — |
case-22 | pass→pass | 10,759 | 10,510 | -2% | 1 | 1 | 0% | 1,190 | 2,546 | +114% | 0 | 0 | — |
case-23 | pass→pass | 9,615 | 5,346 | -44% | 1 | 1 | 0% | 779 | 2,251 | +189% | 0 | 0 | — |
case-24 | pass→pass | 8,439 | 2,705 | -68% | 1 | 1 | 0% | 658 | 2,066 | +214% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 24 cases were attempted, and 23 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +13 percentage points is the difference between those two pass rates over the 23 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.