Question 1

How does SkillzBench measure whether an agent skill actually works?

Accepted Answer

Every skill is tested in a sandbox with the same model, run both with and without the skill. The difference is published as the Lift in percentage points, always shown separately from ratings so the two can never be conflated.

Question 2

What does the 0-100 trust score actually include?

Accepted Answer

The trust score is built from five signal categories: security, quality, license, maintenance and community signals. Critically, money never touches scores — paid placement or sponsorship cannot move a skill's rating.

Question 3

Are script-bearing skills supported?

Accepted Answer

At launch, SkillzBench covers instruction-only skills — SKILL.md packages plus text references. Script-bearing skills are listed link-out only, so you can find them but they aren't sandbox-verified in the initial scope.

Question 4

When can I get access to benchmark-as-a-service priority runs?

Accepted Answer

SkillzBench is pre-launch and actively being built. The waitlist and newsletter are live now — joining puts you in line for early access and priority benchmark runs once the service opens.

Question 5

How is this different from just browsing GitHub for skills myself?

Accepted Answer

GitHub gives you raw listings with no verification, no version pinning and no performance data. SkillzBench dedupes the landscape, pins every version to a commit SHA, security-scans and license-checks it, and benchmarks it in a sandbox so you know the Lift before you install.

Benchmark-as-a-Service for Developers