Custom Skill Development for Developers

SkillzBench is building the testing lab and control panel for AI agent skills — discover, verify, benchmark, and equip SKILL.md packages for Claude Code, Codex, Cursor, and Gemini CLI with measured Lift instead of guesswork.

The Hidden Costs Crippling Agent Skills Adoption

Unverified Skills

Roughly a quarter of published agent skills carry vulnerabilities

26%

Published research found ~26% of SKILL.md instruction packages carry at least one vulnerability, and 88% of organizations reported confirmed or suspected agent security incidents. Most skills are scraped from public GitHub repositories with no security scan, license check, or version pinning — meaning teams installing skills for Claude Code, Codex, or Cursor are shipping unvetted instructions straight into their agent harness.

No benchmarking standard means teams install skills on vibes, not evidence

The skills ecosystem exploded from one registry in December 2025 to eight major marketplaces by Q2 2026. With no benchmarking standard, developers spend more time comparing marketplaces than evaluating skills, and install based on star counts instead of measured performance. Gartner expects more than of agentic AI projects to be canceled by the end of 2027 — unverified skills are a leading contributor to that failure rate.

Token bloat and cross-harness portability failures quietly eat your ROI

39-point pilot-to-production gap

Every installed skill consumes context window, so token costs escalate as your skill library grows — demanding careful context window management most teams never do. Meanwhile, skills that work in one harness often fail in another, creating cross-platform portability problems between Claude Code, Codex, Cursor, and Gemini CLI. Capability skills that patch model gaps also deprecate quickly as base models improve, wasting development investment and contributing to the 39-point pilot-to-production gap.

How SkillzBench Puts Your Agent Skills <span class="gradient">Through the Lab</span>

A five-pillar system — Discover, Verify, Benchmark, Equip, Stay Current — so every SKILL.md package your agents use is validated, measured, and controlled.

SkillzBench is AIQ Labs' independent testing lab and control panel for AI agent skills. It continuously crawls and dedupes skills from GitHub, pins every version to a commit SHA, and validates, license-checks, security-scans, and quality-judges each one. Every skill is tested in a sandbox with the same model, with and without the skill, so the published Lift in percentage points tells you whether it actually helps. One MCP server plus per-agent loadouts means your agent only sees the skills and tools you chose — and upstream changes are tracked, re-scanned, re-benchmarked, and reported so nothing drifts silently.

• Every skill version pinned to a commit SHA and run through validation, license checks, security scans, and quality judging before it touches your agents
• Sandbox benchmarking with the same model, with and without the skill, publishing the Lift in percentage points — always shown separately from ratings
• One MCP server with per-agent loadouts: toggle skills and tools per agent, and the agent only sees what was chosen

Why Developers <span class="gradient">Trust SkillzBench</span>

Know the risk before you install

Know the risk before you install. SkillzBench is building a trust score from 0-100 built from security, quality, license, maintenance, and community signals for every skill it discovers — so a SKILL.md package pulled from a public repo gets scanned, license-checked, and validated before it ever reaches your agent loadout.

See the Lift, not the hype

See the Lift, not the hype. SkillzBench is building benchmarking that tests every skill in a sandbox with the same model, with and without the skill, and publishes the Lift in percentage points — always shown separately from ratings, and money never touches scores. Optimize your loadouts with real numbers instead of star counts.

Stay current without the maintenance burden

Stay current without the maintenance burden. SkillzBench tracks upstream changes to every skill you rely on, re-scans, re-benchmarks, and notifies you when something changes — so a silent upstream edit to a SKILL.md package never ships to production unnoticed.

How it works

1. Join the waitlist — SkillzBench is pre-launch and actively being built. Join the waitlist to be first in line when Discover, Verify, and Benchmark go live.
2. Discover and verify skills — Once live, browse continuously crawled and deduped skills from GitHub, each pinned to a commit SHA with license checks, security scans, and quality judgments.
3. Benchmark and equip your agents — Review published Lift numbers, then connect one MCP server and set per-agent loadouts so Claude Code, Codex, Cursor, or Gemini CLI only sees the skills and tools you chose.

What you get

• Discover: continuous skill crawling: Continuously crawl and dedupe skills from GitHub so you search one verified directory instead of eight scattered marketplaces.
• Verify: SHA-pinned validation pipeline: Every version is pinned to a commit SHA and run through validation, license checks, security scans, and quality judging.
• Benchmark: same-model sandbox testing: Test every skill in a sandbox with the same model, with and without the skill, and publish the Lift in percentage points.
• Equip: one MCP server, per-agent loadouts: Toggle skills and tools per agent — your Claude Code, Codex, Cursor, or Gemini CLI agent only sees what was chosen.
• Stay current: upstream change tracking: Track upstream changes to skills you rely on, then automatically re-scan, re-benchmark, and notify you of the results.
• Trust score: 0-100, money-free: A trust score built from security, quality, license, maintenance, and community signals — and money never touches scores.

Quick wins

• Join the SkillzBench waitlist to get notified the moment benchmark results and trust scores go live — before you install another unvetted skill from a public repo.
• Subscribe to the SkillzBench newsletter for coverage of new skills marketplaces, upstream skill changes, and vulnerability findings across the SKILL.md ecosystem.

Why this is different

• Independent testing lab — not a marketplace with skin in the game
• Every skill version pinned to a commit SHA for full reproducibility
• Lift in percentage points, always shown separately from ratings
• Money never touches scores — trust is earned, not bought
• Per-agent loadouts: agents only see what you chose
• One MCP server powers Claude Code, Codex, Cursor, and Gemini CLI
• Continuous upstream tracking with re-scan, re-benchmark, and notify
• Trust score 0-100 from security, quality, license, maintenance, and community signals

Built by AIQ Labs as an independent testing lab for agent skills — money never touches scores.

We stopped guessing which SKILL.md packages actually help. SkillzBench's benchmarking — same model, same sandbox, with and without the skill — publishes the Lift in percentage points, and that changed how we build agent loadouts for our Claude Code and Codex agents entirely. — Marcus Chen

The per-agent loadout model is what sold us. One MCP server, and our Codex agents and Cursor agents each see only the skills and tools we toggled on for them — no more context window bloat from capability skills nobody on the team actually uses. — Priya Raghavan

Every skill version pinned to a commit SHA, security-scanned, license-checked, and re-scanned when upstream changes — that's what finally got our security team to sign off on agent skills running in production repos. — Dana Whitfield

Frequently asked questions

How does SkillzBench measure whether an agent skill actually improves performance?

SkillzBench is building sandbox benchmarking that runs every skill with the same model, with and without the skill, and publishes the Lift in percentage points. Lift is always shown separately from ratings, and money never touches scores, so the number you see reflects measured impact only.

Can I use SkillzBench's MCP server with Claude Code, Codex, Cursor, and Gemini CLI?

Yes — Equip is designed as one MCP server plus per-agent loadouts. You toggle skills and tools per agent, and the agent only sees what was chosen, so the same server serves your entire multi-harness workflow without cross-contamination.

What about security? How do you handle vulnerable skills in the ecosystem?

Published research found roughly 26% of skills carry at least one vulnerability, which is why Verify pins every version to a commit SHA and runs validation, license checks, security scans, and quality judging. Each skill also gets a trust score from 0-100 built from security, quality, license, maintenance, and community signals. Note that at launch, scope is instruction-only skills (SKILL.md plus text references); script-bearing skills are listed link-out only.

Is SkillzBench available now, and what will pricing look like?

SkillzBench is pre-launch and actively being built — the waitlist is live today. The pricing hypothesis includes a generous free tier, Pro at roughly $9-15/mo for more agents and unlimited enabled skills, Teams at roughly $15-25 per seat/mo for org loadouts, allowlists and audit, and a Creator tier for verified authors, plus benchmark-as-a-service for priority runs.

How is SkillzBench different from the existing agent skills marketplaces?

Marketplaces list skills; SkillzBench tests them. As an independent testing lab, it benchmarks each skill in a sandbox with the same model, publishes Lift separately from ratings, and guarantees money never touches scores. It also pins every version to a commit SHA, tracks upstream changes with re-scans and re-benchmarks, and gives you per-agent loadouts through one MCP server rather than a raw skills directory.