Question 1

How does SkillzBench measure whether an agent skill actually helps?

Accepted Answer

We benchmark every skill in a sandbox with the same model, running the agent with and without the skill, and publish the Lift in percentage points. Lift is always shown separately from our trust score, so performance evidence and quality ratings never blur together.

Question 2

What does the SkillzBench trust score actually measure?

Accepted Answer

The trust score runs 0-100 and is built from security, quality, license, maintenance, and community signals. It's computed independently of benchmark results, and money never touches scores — there's no way to pay for a better rating.

Question 3

Are script-bearing skills supported at launch?

Accepted Answer

Our launch scope is instruction-only skills — SKILL.md files plus text references — which covers the majority of the ecosystem. Script-bearing skills are listed link-out only so you can review them directly, without us implying they've passed our verification pipeline.

Question 4

When can I start using SkillzBench?

Accepted Answer

SkillzBench is pre-launch and actively being built. The waitlist and newsletter are live now, so you can join today and follow our methodology and early findings as they're published. Joining the waitlist is the fastest path to early access.

Question 5

How is SkillzBench different from just browsing GitHub for skills?

Accepted Answer

GitHub gives you raw, unverified SKILL.md files with no quality signal, no security scan, and no performance data. SkillzBench dedupes what's out there, pins versions to commit SHAs, runs security and license checks, and benchmarks each skill in a sandbox to publish real Lift — plus an MCP server with per-agent loadouts so agents only see what you enabled.

Agent Skill Benchmarking for Product Managers