AI Agent Skills Security Scanner
Scan AI Agent Skills files and repositories for prompt injection, supply chain vulnerabilities, privilege escalation, and data exfiltration risks. Free from Backslash Security.
What this scanner checks
- Prompt injection - hidden instructions in SKILL.md or bundled files that could hijack the agent
- Supply chain vulnerabilities - CVEs in dependencies the Skill pulls in
- Privilege escalation - paths to broader system access than the Skill advertises
- Data exfiltration - network calls or file reads that leak data outside the agent context
What is an AI Agent Skill?
An AI Agent Skill is a modular capability package (prompt instructions, code, config) that an AI agent such as Claude loads at runtime to perform a specific task. Because Skills can call tools, access files, and execute code, each one is a new attack surface — and most organizations don't have a way to vet them before they ship.
Frequently asked questions
What is an AI Agent Skill?
An AI Agent Skill is a modular capability package (prompt instructions, code, config) that an AI agent such as Claude loads at runtime to perform a specific task. Skills can call tools, access files, and execute code, which makes them a new attack surface.
What risks does the Backslash Skills Scanner detect?
Prompt injection, software supply chain dependency vulnerabilities, privilege escalation paths, and data exfiltration risks hidden in Skills files, SKILL.md manifests, and bundled scripts.
Is the scanner free?
Yes. The AI Agent Skills Security Scanner is free to use for grading individual Skills files and public repos.
Who built this scanner?
Backslash Security built it as part of its Agentic AI Endpoint Security platform, which secures developer use of AI coding agents, MCP servers, and AI Skills.
What does Backslash Security do?
Backslash is the Agentic AI Endpoint Security platform - we secure developer use of AI coding agents (Cursor, Claude Code, Copilot), MCP servers, and AI Skills across the enterprise.
How is this different from a code scanner?
Traditional SAST looks at your code. The Skills scanner looks at the agent capability layer - prompts, manifests, and the runtime behavior Skills unlock when loaded.
Learn more about Backslash Security · MCP Server Security Hub · Blog