Security audit

Pentest Workbench

Security checks across malware telemetry and agentic risk

Overview

This is a coherent penetration-testing skill, but it includes high-impact exploitation, credential-access, scanning, and post-exploitation guidance without clear authorization safeguards.

Install only if you are using it for authorized security testing, training labs, CTFs, or owned systems. Review the scripts before running them, do not aim scans or fuzzers at third-party systems without written permission, and treat the credential-dumping, persistence, AV-evasion, and lateral-movement sections as high-risk material that should be constrained to an approved engagement scope.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
YARA SignaturesMalware Match, Webshell Match, Cryptominer Match
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (8)

Vague Triggers

Medium

Confidence: 95% confidence
Finding: The skill declares very broad trigger phrases such as 'exploit this', 'run a pentest', 'rev shell', and 'test this target', which can cause the skill to activate in ambiguous or unauthorized contexts. Because the skill content contains offensive tradecraft, unintended activation materially increases the chance of unsafe guidance being surfaced to users without adequate authorization checks.

Missing User Warnings

High

Confidence: 98% confidence
Finding: This is a true security concern in the context of an agent skill because the file provides step-by-step exploit development guidance, including EIP control, JMP ESP discovery, shellcode generation, and reverse-shell payload creation, without any gating, safety framing, or authorization warnings. In an automation-oriented offensive security skill, this materially lowers the barrier to unauthorized exploitation and enables real-world abuse beyond purely educational discussion.

Missing User Warnings

High

Confidence: 98% confidence
Finding: This reference file provides detailed privilege-escalation, credential-access, and exploitation guidance, including shell escapes, SUID abuse, kernel exploits, and credential dumping commands, but it lacks any explicit authorization, safety, or anti-misuse warning. In an agent skill that is explicitly marketed for offensive security and exploitation, this materially increases the risk that the agent will facilitate unauthorized compromise rather than bounded defensive testing.

Missing User Warnings

Medium

Confidence: 92% confidence
Finding: This file is not merely mentioning security concepts; it inventories offensive tooling across reconnaissance, exploitation, credential access, command-and-control, AV evasion, and includes concrete payload-generation commands such as reverse-shell msfvenom examples. In an agent skill, that materially lowers the barrier to unsafe use because the content can be surfaced as ready-to-use operator guidance without any authorization boundary, safety gating, or restriction to lab-only contexts.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: This script deliberately sends increasingly large crafted inputs to a remote vulnserver instance in order to identify crash conditions. Even though this is a common penetration-testing workflow, it can disrupt availability or crash a service if pointed at a real or unauthorized target, and the file provides no safety warning, scope check, or confirmation step before transmitting the fuzz payloads.

YARA rule 'offensive_tool_references': References to well-known offensive security tools [hacktools]

High

Category: YARA Match
Content: - **Unconstrained Delegation**: Printer bug → DC compromise - **BloodHound**: Find shortest path to Domain Admin ## Quick Checklist - [ ] `sudo -l` — what can you run as root? - [ ] SUID binaries — `find / -perm -4000`
Confidence: 95% confidence
Finding: The file explicitly references BloodHound for identifying attack paths to Domain Admin, which is a high-value offensive technique for Active Directory compromise. In the context of a pentest/exploitation skill, inclusion of such tooling without guardrails can enable lateral movement planning and privilege escalation in unauthorized environments.

YARA rule 'privilege_escalation_tools': Privilege escalation tools and techniques [hacktools]

High

Category: YARA Match
Content: ``` Known exploits (verify before running): - `CVE-2022-0847` (DirtyPipe) — Linux 5.8+ - `CVE-2021-4034` (PwnKit) — polkit < 0.120 - `CVE-2019-13272` (PTRACE_TRACEME) — <= 5.1.17 - `CVE-2017-16995` (eBPF) — <= 4.14
Confidence: 97% confidence
Finding: This section names concrete privilege-escalation exploits and commands, including DirtyPipe, Dirty COW, PwnKit, and SUID enumeration, providing actionable paths to root compromise. Because the skill is explicitly designed for offensive testing and exploitation, these references substantially increase operational misuse risk when not constrained by authorization and safety controls.

YARA rule 'info_stealer': Information stealer patterns (credential harvesting, browser data theft) [malware]

High

Category: YARA Match
Content: **Mimikatz** (Windows): ``` privilege::debug sekurlsa::logonpasswords sekurlsa::tickets kerberos::list ```
Confidence: 99% confidence
Finding: The document includes credential-access tooling and commands for Mimikatz and LaZagne, including LSASS password and ticket extraction, which directly enable credential theft and follow-on compromise. In an agent skill that supports exploitation and privilege escalation, this is highly dangerous because it facilitates theft of secrets, account takeover, and domain-wide compromise.

VirusTotal

1/65 vendors flagged this skill as malicious, and 64/65 flagged it as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.