Who SkillSpector is for#
Developers
Ensure the safety of third-party AI skills before integration.
Skip if:
Skip if only using trusted internal skills.
Security Analysts
Assess and report on the security posture of AI agent skills.
Skip if:
Skip if no AI skills are being evaluated.
The problem it solves#
The project helps solve the problem of identifying vulnerabilities and malicious patterns in AI agent skills before they are installed, ensuring user safety and security.
How it solves it#
Multi-format Input
Scan Git repos, URLs, zip files, directories, or single files.
64 Vulnerability Patterns
Detect vulnerabilities across 16 categories including prompt injection and data exfiltration.
Two-stage Analysis
Combine fast static analysis with optional LLM semantic evaluation.
Live Vulnerability Lookups
Query OSV.dev for real-time CVE data with automatic offline fallback.
Multiple Output Formats
Generate reports in Terminal, JSON, Markdown, and SARIF formats.
Risk Scoring
Provides a score from 0-100 with severity labels and recommendations.
Strengths and trade-offs#
Strengths
- Comprehensive DetectionIdentifies a wide range of vulnerabilities across various categories.
- Real-time DataUtilizes live vulnerability lookups for up-to-date security information.
- Flexible UsageSupports multiple input formats and output report types.
- User-friendlyProvides clear recommendations based on risk scoring.
Trade-offs
- -Static Analysis LimitationsStatic analysis may miss vulnerabilities that only manifest during runtime.
- -Language SupportMay not effectively analyze non-English content.
- -Image-based AttacksCannot analyze text in images or encrypted/binary code.
SkillSpector vs alternatives#
SkillSpector vs Snyk Code
SkillSpector and Snyk Code both scan for security risk, but they target different surfaces. SkillSpector focuses on AI agent skills before installation; Snyk Code focuses on application source code security across developer workflows.
| Feature | SkillSpector | Snyk Code |
|---|---|---|
| License | Apache-2.0 | Proprietary service with free and paid plans |
| Primary target | AI agent skills, skill repositories, zip files, directories, and individual skill files | Application source code in IDEs, repositories, and CI/CD workflows |
| Analysis style | Static checks plus optional LLM semantic analysis | SAST scanning with prioritization and automated fix guidance |
| Vulnerability data | Uses OSV lookups for live CVE data with offline fallback | Uses Snyk's managed code security engine and workflow integrations |
Choose SkillSpector when the risk is installing an agent skill that may contain prompt injection, data exfiltration, excessive agency, MCP tool abuse, or other agent-specific behavior. Its scanner understands the skill format and reports risk in terminal, JSON, Markdown, and SARIF formats.
Choose Snyk Code when the goal is broader application security scanning across a development team. It fits IDE, repository, and pull request workflows better than a skill-specific scanner. SkillSpector is the focused pre-installation check for agent skills; Snyk Code is the managed SAST workflow for application code.
Quick start#
# Local Python install
git clone https://github.com/NVIDIA/skillspector.git
cd skillspector
uv venv .venv && source .venv/bin/activate
make install
# Docker self-hosted scan
make docker-build
docker run --rm -v "$PWD:/scan" skillspector scan ./my-skill/ --no-llmWhat it's built on#
- Languages
- PythonTypeScript
- Infrastructure
- AWS
FAQ#
What types of files can SkillSpector scan?
SkillSpector can scan Git repositories, URLs, zip files, directories, and single files.
How does SkillSpector determine risk scores?
Risk scores are calculated based on the severity of detected vulnerabilities, with higher scores indicating greater risk.
Can I use SkillSpector without Python?
Yes, SkillSpector can be run using Docker without needing to install Python.
Similar open-source tools#
ASC
Zero-overhead Android decompiler for mobile security researchers
Claude-Red
78 offensive security skills for Claude, drop-in and MIT licensed
pentagi
Autonomous AI agents for hands-free penetration testing
reverse-skill
AI skill router for reverse engineering and penetration testing
Doberman-Core
Runtime guardrails that gate every AI agent tool call
jentic-one
Self-hosted API broker that keeps AI agent credentials secure