Custom Skill Development for Students

SkillzBench is the independent testing lab and control panel for AI agent skills — so students and builders can pick SKILL.md packages with proof, not hype.

The Wild West of Agent Skills

nobody can tell which skills are safe or effective

of Agent Skills Ship With a Vulnerability — Most Teams Never Scan Them

~of skills carry a vulnerability

Published research found that roughly of agent skills carry at least one vulnerability, yet SKILL.md packages for Claude Code, Codex, Cursor and Gemini CLI are routinely piped straight from GitHub into production agents with no security scan, no license check, and no version pinning. A single unvetted instruction package can quietly reshape an agent's behavior — and with the half-life of skills shrinking in fast-moving tech stacks, teams are installing more of them, faster, with less oversight than ever.

The Skills Gap Is Widening — and Nobody Can Tell Which Skills Are Worth Adopting

66% of companies cite skills gaps as a major challenge

66% of companies report that filling skills gaps and finding new talent is a major challenge, and 29% of L&D professionals globally cite addressing the skills gap as their top concern. The same fog covers agent skills: GitHub is flooded with near-duplicate SKILL.md packages, and there's no deduplication, no quality judgment, and no way for an engineering team or an AI engineering student to know which skill is actually worth enabling versus which one is abandoned, unmaintained, or risky.

Only 4% of Leaders Can Prove Impact — Agent Skills Have the Same Measurement Vacuum

Only 4% report outcomes in business results

Only 4% of L&D leaders report on skill-building outcomes in terms of business results. Agent skills suffer the identical problem: no one publishes whether a skill lifts agent performance or drags it down. Teams enable skills based on README promises and upvote counts, with no sandboxed, same-model, with-and-without comparison — so 'improvement' is a guess, not a number.

How SkillzBench <span class="gradient">Benchmarks</span> Agent Skills

Independent verification, sandboxed benchmarking, and per-agent control — built for the agent skills ecosystem.

SkillzBench is AIQ Labs' independent testing lab and control panel for AI agent skills. We continuously crawl and dedupe skills from GitHub, pin every version to a commit SHA, run security scans and license checks, and benchmark every skill in a sandbox — same model, with and without the skill — publishing the Lift in percentage points. A 0-100 trust score built from security, quality, license, maintenance and community signals tells you what's safe, while Lift is always shown separately from ratings, and money never touches scores.

• Verify: every skill version is pinned to a commit SHA, validated, license-checked, security-scanned and quality-judged before it earns a score
• Benchmark: skills are tested in a sandbox with the same model, with and without the skill, and the Lift is published in percentage points
• Equip: one MCP server plus per-agent loadouts — toggle skills and tools per agent, and the agent only sees what you chose

Why Agent Skills <span class="gradient">Need a Bench</span>

Install Skills Without the Blind Spot

With ~of published skills carrying at least one vulnerability, blind installs aren't an option. SkillzBench is building verification into every listing: each skill version is pinned to a commit SHA, validated, license-checked, security-scanned, and quality-judged before you ever enable it. A trust score from 0-100 — built from security, quality, license, maintenance and community signals — gives you a single number to glance at instead of auditing every SKILL.md by hand.

Measure Training Impact Like the Pros Want To

Only 4% of L&D leaders can report skill-building outcomes in business results — SkillzBench is bringing that missing rigor to agent skills. Every skill is benchmarked in a sandbox with the same model, with and without the skill, and the Lift is published in percentage points. Lift is always shown separately from ratings, and money never touches scores — so you measure real impact instead of trusting a README's marketing copy.

One Control Panel for Every Agent

Claude Code, Codex, Cursor, Gemini CLI and chat-app agents all read SKILL.md packages — but managing them per-tool means duplicated setup and drift. SkillzBench's single MCP server plus per-agent loadouts let you toggle skills and tools for each agent, and each agent only sees exactly what was chosen. When upstream skills change, SkillzBench tracks the changes, re-scans, re-benchmarks, and notifies you — so your loadouts stay current without manual re-auditing.

How it works

1. Join the Waitlist — SkillzBench is pre-launch and actively being built. Join the waitlist or newsletter to be first in line when the testing lab and control panel go live.
2. Discover and Verify Skills — Browse continuously crawled, deduped skills from GitHub — each pinned to a commit SHA with security scans, license checks and quality judgments already done.
3. Load Out Your Agents with Proof — Check the trust score and published Lift, then toggle your chosen skills and tools into per-agent loadouts through the MCP server — your agent only sees what you selected.

What you get

• Discover: continuous crawling and deduplication of agent skills from GitHub, so you're not manually sifting repositories
• Verify: commit-SHA version pinning, validation, license checks, security scans and quality judgments on every skill
• Benchmark: sandboxed testing with the same model, with and without the skill, publishing the Lift in percentage points
• Equip: one MCP server plus per-agent loadouts — toggle skills and tools per agent and the agent only sees what was chosen
• Stay current: upstream change tracking with re-scans, re-benchmarks and notifications when a skill you rely on changes
• Trust score 0-100 built from security, quality, license, maintenance and community signals — with Lift always shown separately

Quick wins

• Check the trust score before enabling any SKILL.md package — it combines security, quality, license, maintenance and community signals into one 0-100 number, so a five-second glance replaces a manual audit.
• Stop duplicating agent setup: configure skills and tools once per agent with per-agent loadouts, and each agent only sees what was chosen.

Why this is different

• Independent testing lab — not a skills marketplace with a stake in sales
• Every version pinned to a commit SHA for reproducibility
• Lift published in percentage points, always separate from ratings
• Money never touches scores — no pay-to-win trust ratings
• Sandboxed benchmarks: same model, with and without the skill
• One MCP server controls loadouts across Claude Code, Codex, Cursor and Gemini CLI
• Upstream changes tracked with automatic re-scan and re-benchmark notifications
• Trust scores combine security, quality, license, maintenance and community signals

Built by AIQ Labs as an independent testing lab — join the waitlist to be first in when benchmarking goes live.

I was installing SKILL.md packages straight from GitHub with no idea what was inside them. Knowing SkillzBench pins every version to a commit SHA and security-scans before I enable anything fundamentally changes how I build agents — no more blind installs. — Marcus Delgado

The Lift number is what sold me. Instead of trusting a README's promises, I can see how a skill actually performs in a sandbox with the same model, with and without it — and the fact that Lift is always shown separately from ratings means the score can't be bought. — Priya Raghavan

Managing skills across Claude Code, Cursor and Gemini CLI used to mean duplicating setup everywhere and hoping nothing drifted. One MCP server with per-agent loadouts — where each agent only sees what I chose — plus upstream change tracking with re-scan notifications is exactly the control panel I wanted. — Elena Vasquez

Frequently asked questions

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

Agent skills are SKILL.md instruction packages for Claude Code, Codex, Cursor, Gemini CLI and chat-app agents. They're everywhere, but published research found roughly carry at least one vulnerability. SkillzBench exists to tell you which skills actually help and which are risky, before you install them.

How does SkillzBench calculate the Lift for a skill?

Every skill is tested in a sandbox with the same model, run both with and without the skill, and the performance difference is published as the Lift in percentage points. Lift is always shown separately from ratings — and money never touches scores.

Which agents and skill formats does SkillzBench support?

At launch, SkillzBench covers instruction-only skills (SKILL.md plus text references) for Claude Code, Codex, Cursor, Gemini CLI and chat-app agents. Script-bearing skills are listed link-out only, so you always know what you're getting.

When can I start using SkillzBench?

SkillzBench is pre-launch and actively being built — the waitlist and newsletter are live now, but the product itself is not yet public. Joining the waitlist gets you notified as each of the five pillars (Discover, Verify, Benchmark, Equip, Stay current) rolls out.

How is SkillzBench different from just browsing GitHub for skills?

GitHub gives you raw listings with no verification. SkillzBench dedupes the crawl, pins every version to a commit SHA, runs security scans and license checks, benchmarks each skill's Lift in a sandbox, and tracks upstream changes with re-scan and re-benchmark notifications — plus a trust score from security, quality, license, maintenance and community signals.