Breaking

NVIDIA Open-Sources SkillSpector, a Security Scanner for AI Agent Skills

July 18, 2026 at 12:45 EDT

Security
AI Agents
Open Source

NVIDIA has released SkillSpector, an open-source security scanner that inspects AI agent "skills" for credential theft, data leaks, prompt injection and vulnerable dependencies before they are installed. The tool, published under an Apache 2.0 license on GitHub, is designed to vet the portable instruction-and-code bundles that power agents such as Claude Code, Codex CLI and Gemini CLI.

Open Source · NVIDIA

SkillSpector: A Security Scanner for the AI Agent Era

An open-source, Apache 2.0 tool that vets AI agent "skills" for credential theft, data leaks, prompt injection and vulnerable dependencies — before they are ever installed.

26.1%

of public skills contain vulnerabilities

5.2%

show signs of malicious intent

0–100

risk score per scan

Skills that ship executable scripts carry higher risk

Relative risk level — proportional comparison

1×

Instruction-only skill

2.12×

Skill with executable scripts

A two-stage scanning pipeline

Input

Git repo · URL · ZIP · directory · SKILL.md

→

Static analysis

Fast, deterministic default pass

→

Optional LLM layer

Semantic intent comparison

→

Output

Terminal · JSON · Markdown · SARIF

What it checks for

Prompt injection Data exfiltration Privilege escalation Supply chain risk Excessive agency System prompt leakage Memory poisoning Tool misuse MCP tool poisoning

Dependency checks query the OSV.dev database live with an offline fallback. Design references the OWASP LLM Top 10, OWASP Agentic AI risks and MITRE ATLAS.

Positive reception

Seen as a timely control for the agent era
Real cases of flagged suspicious skills
SARIF output wires into GitHub Actions CI

Still to be proven

Detection coverage untested at scale
False-positive behavior as adoption grows
New tool — no long track record yet

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.