Agent SecuritySandbox

Security checks across malware telemetry and agentic risk

Overview

This skill describes a Docker sandbox for unsafe code, but the package is missing the runner and Dockerfile needed to prove or provide that protection.

Review before installing. The sandbox idea is reasonable, but do not rely on this package to run untrusted code unless the missing runner script and Dockerfile are provided and audited for no network, dropped capabilities, non-root execution, seccomp use, no host-sensitive mounts, and disposable containers.

SkillSpector

By NVIDIA

Vulnerability Patterns

Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (1)

Vague Triggers

Medium

Confidence: 91% confidence
Finding: The README states that agents can use the sandbox automatically when they 'identify untrusted code execution tasks,' but it does not define explicit trigger conditions, user-consent requirements, or safety boundaries. Ambiguous auto-activation can cause an agent to invoke this skill too broadly or inappropriately, leading to unintended execution of attacker-supplied commands even if they run inside a container.

VirusTotal

63/63 vendors flagged this skill as clean.

View on VirusTotal