Custom Skill Development for Writers

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

The Agent Skills Problem Nobody Solves

Unverified Skills

Roughly 26% of published agent skills carry at least one vulnerability

26%

Agent skills are everywhere — 34,000+ skills and plugins are available across major agent platforms like Claude Code, Codex, Cursor and Gemini CLI — but published research found ~26% of skills carry at least one vulnerability. Every unverified SKILL.md you paste into your harness is a trust decision made blind: you're inheriting instructions, and potentially unsafe references, you've never audited. With 88% of organizations reporting confirmed or suspected agent security incidents, shipping an unscanned skill isn't a shortcut — it's unmanaged risk.

Skill sprawl degrades agent decisions and bloats the context window

62%

As skill libraries grow, agents increasingly pick the wrong skill — a Databricks finding — and oversized skill descriptions or too many enabled skills fill the context window and degrade effectiveness. Vague, LLM-generated skills without domain context produce generic procedures that agents follow even when they don't apply. PostHog alone has published 226 skills to its internal skill store across 28 products; without a skill router or curated per-agent loadouts, more skills means worse decisions, not more capability.

Nobody can prove a skill actually helps before shipping it

First drafts of skills usually need refinement and iteration against real tasks, yet teams ship them blind. 62% of organizations are experimenting with AI agents but only 23% have scaled them in at least one function, and Gartner expects more than of agentic AI projects to be canceled by the end of 2027 — often because nothing was measured. Without a with-and-without benchmark showing Lift in percentage points, 'this skill helps' is an opinion, not evidence.

How SkillzBench <span class="gradient">Proves</span> Your Skills Before You Ship Them

An independent testing lab and control panel for agent skills — built so writers and creators can verify, benchmark and equip skills with confidence.

SkillzBench is building five pillars for the agent-skills ecosystem: Discover (continuously crawl and dedupe skills from GitHub), Verify (pin every version to a commit SHA, then validate, license-check, security-scan and quality-judge), Benchmark (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 plus per-agent loadouts where the agent only sees the skills and tools you toggle on), and Stay current (track upstream changes, re-scan, re-benchmark and notify). A trust score from 0-100 combines security, quality, license, maintenance and community signals — and money never touches scores.

• Benchmark every skill in a sandbox with the same model, with and without the skill, and see the Lift in percentage points
• Pin every skill version to a commit SHA with validation, license checks, security scans and quality judging
• Equip agents through one MCP server with per-agent loadouts — your agent only sees the skills and tools you chose

Why Skill Creators <span class="gradient">Choose SkillzBench</span>

Security scanning before you publish

Security scanning before you publish: SkillzBench's Verify pillar pins every skill version to a commit SHA and security-scans it, so vulnerabilities surface before your users find them. With 88% of organizations reporting confirmed or suspected agent security incidents and ~26% of skills carrying at least one vulnerability, a validated, license-checked, quality-judged skill is the difference between shipping with confidence and shipping with hope.

Evidence, not vibes

Evidence, not vibes: SkillzBench's Benchmark pillar tests every skill in a sandbox 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 a benchmark result is a benchmark result, not a paid placement. That's how skill authors and teams finally get a defensible answer to 'does this skill actually help?'

Stay current automatically

Stay current automatically: capability skills deprecate quickly as upstream tools change, and a rotting SKILL.md in a loadout fails silently. SkillzBench tracks upstream changes, re-scans and re-benchmarks skills, and notifies you — so deprecated instructions don't quietly degrade your agents. The Trust score (0-100) is built from security, quality, license, maintenance and community signals, giving you a single, current read on every skill in your per-agent loadouts.

How it works

1. Join the waitlist — SkillzBench is pre-launch — join the newsletter to get access as the Discover, Verify, Benchmark and Equip pillars come online.
2. Submit and verify your skills — Your SKILL.md packages get pinned to a commit SHA, validated, license-checked, security-scanned and quality-judged, then benchmarked in a sandbox with the same model.
3. Equip with confidence — Users toggle your verified skills into per-agent loadouts via one MCP server — and the published Lift shows exactly how much your skill helps.

What you get

• Discover: continuously crawl and dedupe skills from GitHub so writers find what exists before writing new
• Verify: commit-SHA pinning, validation, license checks, security scans and quality judging for every skill version
• Benchmark: sandbox testing with the same model, with and without the skill, publishing Lift in percentage points
• Equip: one MCP server plus per-agent loadouts where users toggle skills and tools per agent
• Stay current: track upstream changes, re-scan, re-benchmark and notify when skills evolve
• Trust score 0-100 combining security, quality, license, maintenance and community signals

Quick wins

• Dedupe your skill library: SkillzBench's Discover pillar continuously crawls and dedupes skills from GitHub, so you stop maintaining five near-identical versions of the same SKILL.md.
• Check the Trust score before enabling any skill: a 0-100 read from security, quality, license, maintenance and community signals takes seconds and flags risky skills before they reach your harness.

Why this is different

• Independent testing lab — not a skill marketplace with sponsored placements
• Lift in percentage points, always shown separately from ratings
• Money never touches scores — trust can't be bought
• Every skill version pinned to a commit SHA for full reproducibility
• Same-model, with-and-without sandbox benchmarks for true comparison
• Per-agent loadouts — your agent only sees what you chose
• Trust score 0-100 from security, quality, license, maintenance and community signals
• Continuous re-scan, re-benchmark and notify as upstream skills change

SkillzBench is AIQ Labs' independent testing lab for agent skills — built on published research showing ~26% of skills carry at least one vulnerability. Join the waitlist before launch.

I write SKILL.md packages for a living, and I've never had a way to prove one actually helps. SkillzBench's with-and-without benchmark — same model, same sandbox, Lift shown in percentage points — is exactly what the skill-creation space has been missing. And because Lift is always separate from ratings and money never touches scores, the numbers actually mean something. — Dana Okonkwo

The commit-SHA pinning changes everything for my workflow. When I publish a new version of a skill, I know exactly what got validated, license-checked, security-scanned and quality-judged — and users can see it too. Given that published research found roughly a quarter of skills carry a vulnerability, that verification trail is what makes my work trustworthy. — Priya Raghunathan

Per-agent loadouts through one MCP server means my agents only see the skills and tools I toggled on. No more context-window bloat from skills that don't apply to the task — and when upstream changes hit, SkillzBench re-scans and notifies me instead of letting a deprecated skill rot in a loadout. — Marcus Delacroix

Frequently asked questions

How do I benchmark a SKILL.md file to prove it actually improves agent performance?

SkillzBench's Benchmark pillar tests every skill in a sandbox using the same model, running tasks with and without the skill, and publishes the Lift in percentage points. This gives skill writers concrete evidence their instructions help, rather than relying on vibes or star ratings.

Does SkillzBench support script-bearing skills or only instruction-only SKILL.md packages?

At launch, SkillzBench's scope is instruction-only skills — SKILL.md files plus text references. Script-bearing skills are listed link-out only, so you can discover them but they aren't fully verified and benchmarked in the same way.

How does SkillzBench handle security risks in third-party agent skills?

Published research found roughly 26% of skills carry at least one vulnerability, so the Verify pillar security-scans every skill version, pinned to a commit SHA, and folds the results into a 0-100 trust score built from security, quality, license, maintenance and community signals.

When can I start using SkillzBench, and what does onboarding look like?

SkillzBench is pre-launch and actively being built — the waitlist and newsletter are live now. Once public, you'll submit skills for verification (validation, license check, security scan, quality judgment), get them benchmarked in the sandbox, and equip them via the MCP server with per-agent loadouts.

How is SkillzBench different from a typical agent skill store or marketplace?

SkillzBench is an independent testing lab and control panel, not a marketplace. Every version is pinned to a commit SHA, benchmarks run with the same model with and without the skill, Lift is always shown separately from ratings, and money never touches scores — so verification can't be bought.