Question 1

What exactly is an agent skill, and why does it need verification?

Accepted Answer

Agent skills are SKILL.md instruction packages that tell agents like Claude Code, Codex, Cursor, and Gemini CLI how to behave. They're spreading fast on GitHub, but published research found roughly of skills carry at least one vulnerability. SkillzBench independently verifies each one with security scans, license checks, and version pinning before it earns a trust score.

Question 2

How does the Lift benchmark work?

Accepted Answer

Every skill is tested in a sandbox with the same model, once with the skill and once without, and the difference is published as the Lift in percentage points. This tells you whether a skill genuinely improves agent performance on relevant tasks. Lift is always shown separately from ratings, so a high score can never buy a good benchmark.

Question 3

Can schools meet AI governance and transparency requirements with SkillzBench?

Accepted Answer

SkillzBench gives you the evidence layer: security scans, license checks, pinned versions, and explainable benchmark results for every skill. As compliance pressure from EU and US AI acts requires transparency and explainability of agent decisions, having documented verification per skill version helps your institution show exactly what each agent was equipped with and why.

Question 4

When can I actually start using SkillzBench?

Accepted Answer

SkillzBench is pre-launch and actively being built — the waitlist and newsletter are live now, and the product itself is not yet public. Join the waitlist to be notified at launch, when you'll get access to verified skills, published benchmarks, and the MCP server with per-agent loadouts.

Question 5

Why not just pick skills straight from GitHub or a marketplace?

Accepted Answer

GitHub listings tell you a skill exists, not whether it's safe or effective. Marketplaces often blur ratings with promotion. SkillzBench is an independent testing lab: we crawl and dedupe skills from GitHub, verify every version, benchmark the Lift with the same model, and keep money entirely separate from scores — so the evidence isn't for sale.

Verified Agent Skills for Teachers