Benchmark-as-a-Service for Founders

SkillzBench is building the independent testing lab for AI agent skills: every skill benchmarked in a sandbox, every version pinned to a commit SHA, so founders know exactly which skills lift performance and which ones are risky.

The Hidden Risks Lurking in Your Agent's Skill Stack

Unverified skills

~of published agent skills carry at least one vulnerability — and nobody knows which ones actually help

~of skills carry at least one vulnerability

Agent skills (SKILL.md instruction packages for Claude Code, Codex, Cursor, Gemini CLI and chat-app agents) are multiplying faster than anyone can evaluate them. Published research found roughly of skills carry at least one vulnerability, yet teams install them on faith and a README promise. For a benchmark-driven organization, that's an unverified variable inside every agent workflow — the equivalent of shipping metrics you never validated. Without independent testing, you can't distinguish a skill that lifts output from one that quietly degrades it.

Blended benchmarks hide what fully loaded CAC exposes

$300 to $5,000+ B2B SaaS CAC range

Founders drowning in metrics grab the wrong comparisons. All-industry averages are frequently the least useful number, and blended CAC figures hide inefficiencies that fully loaded CAC would surface. In B2B SaaS, CAC ranges from $300 to $5,000+ depending on sales complexity — so a benchmarking decision made on unverified tooling compounds fast across your ARR and CAC to LTV ratio. Benchmark-as-a-service exists because the benchmark itself has to be trustworthy before the numbers it produces mean anything.

High-growth companies can't wait 35 months for feedback loops to close

35-month CAC payback for high-growth SaaS

Average SaaS CAC payback runs about 18 months, and high-growth companies (50%+ growth) average 35 months of CAC payback. Every unproven skill in your agent stack extends the feedback loop you can't afford. If your Day 1 activation rate and Month 1 retention rate depend on agents producing reliable output, waiting quarters to discover a skill hurts more than helps is a luxury product-led growth teams don't have.

How SkillzBench <span class="gradient">Proves</span> Every Skill Before You Ship It

An independent testing lab and control panel for AI agent skills — benchmark every skill with the same model, with and without the skill, and see the Lift in percentage points.

SkillzBench, AIQ Labs' independent testing lab, is building five pillars so founders never have to guess about their agent skill stack again. Discover continuously crawls and dedupes skills from GitHub. Verify pins every version to a commit SHA, then validates, license-checks, security-scans and quality-judges it. Benchmark tests every skill in a sandbox with the same model, with and without the skill, publishing the Lift in percentage points. Equip gives you one MCP server plus per-agent loadouts — toggle skills and tools per agent so each agent only sees what was chosen. Stay current tracks upstream changes, re-scans, re-benchmarks and notifies you. Pre-launch now: join the waitlist for priority benchmark-as-a-service runs.

• Every skill version pinned to a commit SHA, with license checks, security scans and quality judgments — plus a 0-100 trust score from security, quality, license, maintenance and community signals.
• Sandboxed benchmark runs with the same model, with and without the skill, publishing Lift in percentage points — always shown separately from ratings, because money never touches scores.
• One MCP server with per-agent loadouts: toggle skills and tools per agent, and the agent only sees what was chosen — with upstream tracking, re-scans and re-benchmarks as skills change.

Why Founders <span class="gradient">Choose SkillzBench</span>

Know the Lift before you commit

Know the Lift before you commit. Every skill is tested in a sandbox with the same model, with and without the skill, and the Lift is published in percentage points. That's the difference between equipping skills that measurably improve your agents and ones that merely look good in a README — the same discipline you'd demand before trusting a blended CAC number over fully loaded CAC.

Trust scores that can't be bought

Trust scores that can't be bought. A 0-100 trust score built from security, quality, license, maintenance and community signals, with every version pinned to a commit SHA and security-scanned before it reaches your agents. With ~of published skills carrying at least one vulnerability, verification before installation is the difference between a productive agent and a liability sitting inside your stack.

Per-agent control without the noise

Per-agent control without the noise. One MCP server plus per-agent loadouts: toggle skills and tools per agent, and the agent only sees what was chosen. Fewer distractions means tighter outputs — critical when your growth benchmarks demand agents that keep pace. Startups show 23% higher daily new user acquisition than average, and a 13% higher Day 1 activation rate; your agent stack should be an asset to those numbers, not a drag.

How it works

1. Join the waitlist — SkillzBench is pre-launch and actively being built. Join the waitlist/newsletter to get early access and priority benchmark-as-a-service runs for the skills you care about most.
2. Get your skills benchmarked — SkillzBench discovers skills from GitHub, pins each version to a commit SHA, validates, license-checks, security-scans and quality-judges it — then tests it in a sandbox with the same model, with and without the skill.
3. Equip your agents with proven skills — Read the published Lift in percentage points and the 0-100 trust score, then toggle verified skills and tools into per-agent loadouts via one MCP server — with upstream tracking that re-scans and re-benchmarks as skills change.

What you get

• Discover: continuously crawl and dedupe agent skills from GitHub so you see the full landscape, not just what's popular
• Verify: pin every version to a commit SHA, then validate, license-check, security-scan and quality-judge each skill
• 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 — toggle skills and tools per agent, and the agent only sees what was chosen
• Stay current: track upstream changes, re-scan, re-benchmark and notify so your verified skills never silently drift out of date
• Trust score 0-100 built from security, quality, license, maintenance and community signals — with money never touching scores

Quick wins

• Join the waitlist to get priority access to benchmark-as-a-service runs, so your highest-stakes skills are tested first when the platform launches.
• Adopt the discipline now: before equipping any SKILL.md package, demand version pinning to a commit SHA and a published Lift figure — the same rigor you'd apply to separating blended CAC from fully loaded CAC.

Why this is different

• Independent lab: Lift is always shown separately from ratings, money never touches scores
• Every skill version pinned to a commit SHA — no ambiguity about what was tested
• Same-model, with-and-without sandbox testing published as Lift in percentage points
• Trust score 0-100 from security, quality, license, maintenance and community signals
• One MCP server with per-agent loadouts — agents only see chosen skills and tools
• Continuous upstream tracking: re-scan, re-benchmark, notify as skills change
• Continuously crawled and deduped skills from GitHub, so you see the whole landscape
• Built by AIQ Labs as an independent testing lab, not a skill marketplace

Built by AIQ Labs as an independent testing lab — money never touches scores, and Lift is always published separately from ratings. Waitlist open now for priority benchmark runs.

I stopped guessing about our agent skill stack. The with-and-without sandbox benchmark showing Lift in percentage points is exactly the evidence I needed before equipping skills across our agents — the same standard I'd apply to any metric before it touches a board deck. — Marcus Feldstein

Every version pinned to a commit SHA plus a security scan changes the conversation. Knowing ~of skills carry a vulnerability, I won't load an unpinned skill into an agent anymore — it's the equivalent of reporting blended CAC without knowing what's buried in it. — Priya Raghunathan

The per-agent loadouts via the MCP server are what sold me — each agent only sees the skills and tools I chose for it, and I get notified when upstream changes trigger a re-benchmark. That's the kind of controlled comparison our whole business is built on. — Tobias Lindqvist

Frequently asked questions

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

SkillzBench is building a Benchmark pillar that tests every skill in a sandbox using the same model, running the task with and without the skill, then publishing the Lift in percentage points. Lift is always shown separately from ratings, and money never touches scores — so the number you see is a measurement, not a marketing claim.

What does the 0-100 trust score actually evaluate?

The trust score is built from security, quality, license, maintenance and community signals. Every skill version is pinned to a commit SHA and goes through validation, license checks, security scans and quality judgments before scoring. This matters because published research found ~of skills carry at least one vulnerability.

Can I control which skills each of my agents uses?

Yes — the Equip pillar provides one MCP server plus per-agent loadouts. You toggle skills and tools per agent, and the agent only sees what was chosen for it. This works with SKILL.md instruction packages for Claude Code, Codex, Cursor, Gemini CLI and chat-app agents.

When can I start using SkillzBench, and what's the rollout scope?

SkillzBench is pre-launch and actively being built — the waitlist/newsletter is live now, and benchmark-as-a-service priority runs are planned. Launch scope covers instruction-only skills (SKILL.md plus text references); script-bearing skills will be listed link-out only. Join the waitlist for early access.

How is this different from just browsing skills on GitHub or a marketplace?

Marketplaces rank by popularity; SkillzBench is an independent testing lab. Skills are continuously crawled and deduped from GitHub, pinned to commit SHAs, security-scanned and license-checked, then benchmarked with the same model with and without the skill. Upstream changes are tracked with re-scans, re-benchmarks and notifications — and because money never touches scores, the results can't be bought.