SkillGuard
Scans AI agent skills for malicious code
2 followers
Scans AI agent skills for malicious code
2 followers
We built a zero-config npm/pip scanner that audits AI agent skills before installation. We expose security gaps by diffing declared permissions against real-world behavior across 10 rule categories. Unlike single-file scanners, we catch combined multi-file risks: privilege chaining, prompt injections, undeclared network calls, and credential harvesting. SkillGuard is open-source, self-hosted, and licensed with Apache 2.0.
Free
Launch Team
World model AI : Sharm SEKAISimulate your idea on a living world.
Promoted
As AI agents become more autonomous, incorporating third-party skills via npm or pip introduces significant security blind spots. We noticed that traditional, single-file scanners struggle to detect complex, multi-file vulnerabilities. Security teams need a reliable way to audit what an AI skill actually does at runtime versus what it claims to do.
We wanted to expose these security gaps before a package is ever installed. The inspiration came from seeing how easily malicious actions such as credential harvesting, undeclared network calls, cross-skill privilege chaining, or hidden prompt injections could bypass standard checks by spreading across hooks, scripts, and SKILL.md files.
What started as a concept for basic permission diffing evolved into a rigorous behavioral auditing tool. We realized that to be truly effective for fast-moving development teams, the tool needed to be completely frictionless. That realization drove us to build a zero-config, self-hosted solution that validates actual code behavior across 10 comprehensive rule categories. We also ensured it could be seamlessly deployed anywhere: via CLI, library, MCP server, or GitHub Action, without any authentication requirement.
Repo:
https://github.com/RudrenduPaul/skillguard
MCP Servers:
https://mcpservers.org/servers/rudrendupaul/skillguard
https://glama.ai/mcp/servers/RudrenduPaul/skillguard
NPM:
https://www.npmjs.com/package/skillguard-cli
PyPI:
https://pypi.org/project/skillguard-cli
We have open-sourced the project under Apache 2.0 and would love your feedback on the rule categories and threat detection capabilities.