Jq Hardened

Security checks across malware telemetry and agentic risk

Overview

This is an instruction-only jq helper with disclosed safety guardrails, though its supplemental safety examples contain mistakes users should not copy blindly.

Install only if you want a lightweight jq reference with safety reminders. Treat SAFETY.md as imperfect evaluation documentation: follow the main guardrails, avoid copying its xargs/sh example, and do not use its curl examples to transmit secrets from local files.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (3)

Intent-Code Divergence

Medium

Confidence: 98% confidence
Finding: The safety guidance presents `xargs -I {} sh -c 'echo "Hello {}"'` as a safe fix even though the document itself later acknowledges this remains injectable because xargs interpolates `{}` before the shell runs. In a safety file for a hardened jq skill, documenting an unsafe pattern as a recommended defense is especially dangerous because users and downstream agents may copy it as trusted remediation.

Intent-Code Divergence

High

Confidence: 99% confidence
Finding: The purported "secure alternatives" in the PASS section still instruct the user to extract an API key from `secrets.json` and transmit it over the network via `curl` headers or body. That directly contradicts the stated guardrail of never piping or sending credentials to network-transmitting commands, weakening the protection and normalizing exfiltration.

Natural-Language Policy Violations

High

Confidence: 99% confidence
Finding: The documented PASS example undermines its own refusal by offering follow-on commands that still place the API key into a network request. This creates a confused-deputy problem: the text appears safety-compliant at first glance but still enables the prohibited behavior through "alternatives."

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal