Security audit

Web Shells

Security checks across malware telemetry and agentic risk

Overview

This skill openly packages web-shell samples for authorized security testing and does not show evidence that it runs or installs them automatically.

Install only if you intentionally need live web-shell specimens for authorized security research, detection testing, or education. Keep the files isolated, do not deploy them to any reachable web server, and do not let an agent copy, package, or execute the samples unless you are in a controlled lab with explicit permission.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
Behavioral ASTexec() Call, eval() Call, Dynamic Import
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands

Findings (13)

Context-Inappropriate Capability

High

Confidence: 99% confidence
Finding: This code enumerates configured ColdFusion datasources, decrypts stored datasource passwords, and prints usernames, passwords, and connection URLs directly to the page. That is explicit credential theft behavior with no legitimate need in the context of a web-facing skill, and it enables immediate database compromise and lateral movement.

Intent-Code Divergence

Medium

Confidence: 96% confidence
Finding: The comments downplay the borrowed code as formatting-related, but the implementation actually decrypts and exposes datasource passwords. This mismatch is deceptive and increases the likelihood that dangerous credential-harvesting behavior could evade review and be deployed unnoticed.

Intent-Code Divergence

Low

Confidence: 99% confidence
Finding: The file is explicitly a CGI-style web shell: it presents an HTML form that accepts a `cmd` parameter, decodes attacker-controlled input from `QUERY_STRING`, and executes it with `eval $VALUE`. This enables arbitrary operating system command execution by anyone who can reach the endpoint, which is a classic remote code execution backdoor.

Intent-Code Divergence

High

Confidence: 99% confidence
Finding: This script is not just a directory lister: it accepts untrusted input from QUERY_STRING, runs shell commands with it, and if the path appears to be a file, it reads and returns the file contents with `cat`. In a CGI/web-shell context, that enables arbitrary file disclosure and supports unauthorized browsing of the filesystem, which is a classic web shell capability and substantially more dangerous than the header comment suggests.

Intent-Code Divergence

Medium

Confidence: 97% confidence
Finding: This script is a classic web-shell style upload utility: it accepts arbitrary file content from an HTTP request, writes it to a predictable location in /tmp, and then echoes the uploaded content back to the requester. In the context of a web-accessible CGI shell script under a Web-Shells reference path, that combination enables unauthorized file staging and data disclosure, making it far more dangerous than a benign upload demo.

Missing User Warnings

High

Confidence: 99% confidence
Finding: The file upload handler accepts user-supplied files and writes them directly to C:\ with no authentication, validation, filename restrictions, or storage isolation. In a web-shell context, this allows attackers to plant tools, overwrite files, or stage further compromise on the host.

Missing User Warnings

High

Confidence: 100% confidence
Finding: The form passes user-controlled values directly into CFEXECUTE, allowing arbitrary OS command execution with attacker-supplied arguments and timeout. This is a classic web shell capability that provides remote code execution on the server and can lead to full system takeover.

Missing User Warnings

Critical

Confidence: 100% confidence
Finding: The script accesses datasource configurations, decrypts passwords, and displays the resulting credentials to any user who can reach the page. Combined with the surrounding upload and command-execution features, this materially increases attacker capability by exposing high-value secrets for database compromise and pivoting.

Missing User Warnings

High

Confidence: 100% confidence
Finding: The script takes web-supplied input, reconstructs URL-decoded characters, echoes the command, and then executes it directly via `eval $VALUE` with no authentication, validation, or confirmation. This is unauthenticated arbitrary command execution from HTTP input, allowing full compromise of the host under the web server's privileges and often escalation through local misconfiguration.

Missing User Warnings

Medium

Confidence: 98% confidence
Finding: The script executes `ls $QUERY_STRING` using untrusted CGI input and writes the result to `/tmp/test`, allowing shell metacharacter injection and unsafe file-system interaction. In addition to command injection risk, using a fixed temporary file in /tmp is unsafe and can be abused via symlink/race attacks, making this far more dangerous in a web-exposed context.

Missing User Warnings

Medium

Confidence: 97% confidence
Finding: The `cat $QUERY_STRING` behavior returns the contents of any attacker-supplied path directly to the HTTP response, creating arbitrary file read/disclosure. In the context of a web-shell collection, this is especially dangerous because it can expose credentials, keys, configuration files, application source, and other sensitive data from the host.

Missing User Warnings

Medium

Confidence: 98% confidence
Finding: The script writes attacker-controlled request bodies into fixed filenames under /tmp (/tmp/test, /tmp/test2, /tmp/upload) without validation, isolation, or secure temporary-file handling. Predictable temporary paths create overwrite, race-condition, symlink, and cross-request data exposure risks, especially when run by a web server account in a multi-user or shared environment.

Missing User Warnings

High

Confidence: 99% confidence
Finding: The README explicitly documents an installable Vtiger extension that provides arbitrary command execution on the web server and even gives a ready-to-use invocation example. In context, this is not a neutral mention of security research; it is operational guidance for deploying a web shell, which would enable full remote code execution and likely complete server compromise.

VirusTotal

14/65 vendors flagged this skill as malicious, and 51/65 flagged it as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.