Developer Tools

SkillzBench: Free Security Scans and Benchmarks for AI Agent Skills

SkillzBench is a newly launched free directory by AIQ Labs that security-scans AI agent skills, benchmarks each against the same model, and publishes the…

AIQ Labs of Halifax, Nova Scotia, launched SkillzBench on October 23, 2026: a free, independent directory of AI agent skills that security-scans every skill, benchmarks it, and publishes the result, including when the result is negative.

Agent skills are packages of instructions, usually a SKILL.md file, that teach AI agents such as Claude Code, Codex, Cursor and Gemini CLI how to handle a specific kind of work. Thousands are shared publicly on GitHub with little vetting. A January 2026 study cited in the launch analyzed 31,132 public skills and found that 26.1 percent contained at least one security vulnerability, with about 5.2 percent showing patterns suggesting malicious intent. SkillzBench exists to answer a simple question before you hand a skill to your agent: does it help, hurt, or do something it should not.

Graphic: GPTOnline, using SkillzBench assets.

How it works

Security scans on every skill: scan gates check for leaked secrets, data exfiltration patterns and risky install hooks before a skill is listed.
Honest benchmarks: each skill is tested against the same AI model working without it, and the difference is published. Negative results are published too, not hidden.
One MCP connector: a single connector gives each agent only the skills its owner selects, with loadouts per agent.
Ongoing monitoring: skills are pinned to specific commits, with alerts and a weekly digest when things change.
Scale at launch: the site reports 799 skills published, 135 benchmarked, 15,419 benchmark trials and 1,476 security scans (figures shown on skillzbench.com, October 2026).

“SkillzBench runs the test and publishes the number, including when the number is negative,” said Pat Riley, founder of AIQ Labs, in the launch announcement.

Who it is for

SkillzBench targets agent developers and teams using Claude Code, Codex, Cursor, Gemini CLI or any MCP client who want a vetting layer between public skill repositories and their production agents.

Pricing

Free, all of it ($0). The official pricing page states “No plans, no trials, no card”; the service is supported by labeled advertising that does not affect scores. Pricing may change, so check skillzbench.com for current terms.