Custom Skill Development for Content Creators

SkillzBench is building the independent testing lab for SKILL.md agent skills — verifying, benchmarking, and equipping your Claude Code, Codex, Cursor and Gemini CLI agents with skills that actually perform.

The Agent Skills Ecosystem Is Growing Faster Than Anyone Can Vet It

Unverified skills

One in four skills ships with a vulnerability — and nobody's checking

~of skills carry at least one vulnerability

Published research found ~of agent skills carry at least one vulnerability, yet most marketplace entries are installed with zero curation or security review. As SKILL.md instruction packages proliferate across Claude Code, Codex, Cursor, and Gemini CLI, developers and teams are wiring unvetted community skills into production agents — sometimes with script-bearing code hidden behind a friendly description. Independent skill authors have no way to prove their packages are the safe ones, and buyers have no signal to distinguish them from the risky majority.

Eight marketplaces and counting — fragmentation eats your shipping time

8 major marketplaces by Q2 2026

The agent skills ecosystem exploded from a single registry in December 2025 to eight major marketplaces by Q2 2026 — many of them thin aggregators pointing at the same handful of real sources. Skill creators now spend more time comparing storefronts and duplicating listings than building skills, while single-marketplace distribution caps their reach. With 62% of organizations already experimenting with AI agents and 23% actively scaling, discovery friction is now a real tax on the entire agentic AI supply chain.

No standard way to prove a skill actually helps

No formal review process

Quality varies wildly across community-driven marketplaces with no formal review process, and star ratings tell you nothing about performance. A skill might read well in its SKILL.md and still degrade agent output — or do nothing at all. With 88% of organizations now using AI in at least one business function, teams need measurable evidence: does this skill lift output quality in a controlled test, or is it just well-marketed text? Today there's no standard answer.

How SkillzBench Makes Custom Skills <span class="gradient">Proven</span>

An independent testing lab and control panel for your agent skills — every skill pinned to a commit SHA, security-scanned, and benchmarked with a measurable Lift.

SkillzBench is building five pillars for the agent skills ecosystem. Discover continuously crawls and dedupes skills from GitHub. Verify pins every version to a commit SHA with validation, license checks, security scans and quality judging. Benchmark runs every skill in a sandbox with the same model — with and without the skill — and publishes the Lift in percentage points. Equip delivers one MCP server plus per-agent loadouts where each agent only sees the skills and tools you toggle on. Stay current tracks upstream changes, re-scans, re-benchmarks, and notifies you. For content creators building custom skills, this means independent, trustworthy proof of quality and performance.

• Every skill version is pinned to a commit SHA, validated, license-checked, security-scanned, and quality-judged before it earns trust
• Sandboxed benchmarking with the same model — with and without the skill — produces a Lift score in percentage points, always shown separately from ratings
• One MCP server plus per-agent loadouts: toggle skills and tools per agent, so each agent only sees exactly what you chose

Why Content Creators <span class="gradient">Trust SkillzBench</span>

Independent trust scores money can't buy

SkillzBench is building an independent 0-100 trust score derived from security, quality, license, maintenance and community signals — with money never touching scores. For skill authors, that means a credible way to stand out in a market where ~of skills carry a vulnerability; for teams adopting agentic AI, it means a signal they can actually trust before installing.

Lift, not hype

SkillzBench's benchmarking pillar tests every skill in a sandbox with the same model — with and without the skill — and publishes the Lift in percentage points. Instead of marketing copy, your SKILL.md package gets a measurable number that shows exactly what it adds to agent output, making value comparisons across skills apples-to-apples.

Per-agent control across your stack

One MCP server plus per-agent loadouts gives you a real control layer across your stack: toggle skills and tools for each of your Claude Code, Codex, Cursor, Gemini CLI and chat-app agents, and each agent only sees what you chose. Stay current tracks upstream changes, re-scans, re-benchmarks, and notifies you — so a silently changed skill never blindsides a production agent.

How it works

1. Join the waitlist — SkillzBench is pre-launch and actively being built. Sign up for the waitlist and newsletter to get early access as the Discover, Verify, and Benchmark pillars come online.
2. Submit and verify your skills — Your instruction-only SKILL.md packages get pinned to a commit SHA, validated, license-checked, security-scanned, and quality-judged — then benchmarked in a sandbox to produce your Lift score.
3. Equip with proof — Deliver verified skills through one MCP server with per-agent loadouts, backed by a 0-100 trust score and continuous upstream tracking that keeps your skills current and rescanned.

What you get

• Discover with deduplication: Continuously crawls and dedupes skills from GitHub so you're not sifting through duplicate listings across fragmented marketplaces.
• Verify pipeline: Every version is pinned to a commit SHA, then validated, license-checked, security-scanned, and quality-judged before it earns a trust score.
• Sandboxed benchmarking: Tests every skill in a sandbox with the same model, with and without the skill, and publishes the Lift in percentage points.
• One MCP server, per-agent loadouts: A single MCP server plus per-agent loadouts lets you toggle skills and tools per agent — each agent only sees what was chosen.
• Stay current monitoring: Tracks upstream changes to skills you depend on, then re-scans, re-benchmarks, and notifies you automatically.
• Creator plan for verified authors: A Creator tier (~$5-10/mo planned) for verified authors, plus benchmark-as-a-service for priority runs of your custom skills.

Quick wins

• Join the SkillzBench waitlist to get notified the moment benchmark results and trust scores go live — before you install another unvetted community skill.
• Subscribe to the SkillzBench newsletter for ongoing coverage of skill vulnerabilities, marketplace fragmentation, and upstream SKILL.md changes worth knowing about.

Why this is different

• Independent testing lab — not a marketplace with conflicted incentives
• Every skill version pinned to an immutable commit SHA
• Lift in percentage points, always shown separately from ratings
• Money never touches scores — trust can't be bought
• Same-model A/B sandbox benchmarking for every skill
• Per-agent loadouts: each agent only sees what you chose
• Trust score 0-100 from security, quality, license, maintenance and community signals
• Stay current: upstream tracking, re-scanning, re-benchmarking and notifications

Built on published research showing ~of skills carry at least one vulnerability — SkillzBench is the independent lab the agent skills ecosystem has been waiting for.

I build SKILL.md packages for Claude Code and Cursor, and I've never had a way to prove they work. The idea of a sandboxed Lift benchmark — same model, with and without my skill — is exactly what this ecosystem is missing. Right now it's all star ratings and vibes. — Maya Okonkwo

Commit SHA pinning and security scanning alone would change how I ship skills. Knowing every version is validated and license-checked before it gets a trust score means the people installing my packages can finally do it with confidence — and I can prove mine aren't part of that ~ — Devon Reyes

Per-agent loadouts are the killer feature for me — one MCP server, and each of my agents only sees the skills and tools I toggled for it. Add upstream change tracking and notifications, and that's the control layer agentic AI has been missing. — Priya Sundaram

Frequently asked questions

How does SkillzBench benchmark a custom SKILL.md skill to prove it actually helps?

SkillzBench is building a benchmarking pipeline that runs every skill in a sandbox with the same model, testing with and without the skill, and publishes the resulting Lift in percentage points. Because the model is held constant, the Lift isolates your skill's actual contribution. Lift is always shown separately from trust ratings, and money never touches scores.

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

At launch, the scope is instruction-only skills — SKILL.md files plus text references. Script-bearing skills will be listed link-out only, so users can find them but SkillzBench won't execute or fully verify them initially.

How does the security scanning handle the vulnerability problem in community agent skills?

Published research found roughly of skills carry at least one vulnerability, which is why SkillzBench's Verify pillar security-scans every skill and folds the results into a 0-100 trust score built from security, quality, license, maintenance and community signals. Skills are also continuously re-scanned when upstream changes are detected.

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 today. Once public, the flow is: your skills get discovered or submitted, pinned to a commit SHA, validated, license-checked, security-scanned and quality-judged, then benchmarked and equipped via the MCP server with per-agent loadouts.

How is SkillzBench different from the existing agent skills marketplaces and directories?

Most marketplaces are listings with star ratings and no formal review process — many are aggregators pointing at a handful of real sources. SkillzBench is an independent testing lab: every version is pinned to a commit SHA, security-scanned, and benchmarked with a published Lift, plus a 0-100 trust score where money never touches scores. It also adds an equip layer — one MCP server with per-agent loadouts — that marketplaces don't provide.