Question 1

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

Accepted Answer

SkillzBench benchmarks every skill in a sandbox using the same model, running tasks with and without the skill, then publishes the Lift in percentage points. Because the comparison is apples-to-apples with a fixed model, the Lift reflects what the skill itself contributes — not model variance or a gamable task setup.

Question 2

What does the 0-100 Trust score actually evaluate?

Accepted Answer

The Trust score combines security, quality, license, maintenance and community signals into a single 0-100 number. It's always shown separately from Lift, and money never touches scores — so a skill's rating reflects real signals, not promotion.

Question 3

Can SkillzBench help with the security risk in open-source agent skills?

Accepted Answer

Yes. Published research found roughly 26% of skills carry at least one vulnerability, which is why SkillzBench security-scans and license-checks every skill as part of its Verify pillar, pinning each version to a commit SHA so you know exactly what was scanned. Script-bearing skills are listed link-out only at launch, keeping the risk surface focused on instruction-only skills.

Question 4

Is SkillzBench available now, and when can my team start using it?

Accepted Answer

SkillzBench is pre-launch and actively being built — the waitlist/newsletter is live, but the product is not yet public. Join the waitlist for early access as the Discover, Verify, Benchmark, Equip and Stay-current pillars come online. Launch scope covers instruction-only skills (SKILL.md plus text references).

Question 5

How is SkillzBench different from just browsing GitHub for agent skills?

Accepted Answer

GitHub shows you stars and READMEs; SkillzBench shows you evidence. It continuously crawls and dedupes skills, validates and security-scans them, benchmarks them with and without the skill to publish Lift, and lets you equip verified skills through one MCP server with per-agent loadouts. It's an independent testing lab — not a marketplace where rankings can be bought.

Agent Skill Benchmarking for Startup Teams