Question 1

How does SkillzBench detect MCP tool poisoning in agent skills?

Accepted Answer

SkillzBench's Verify pillar security-scans every skill version, pinned to a commit SHA, and combines those results into a 0-100 Trust score alongside quality, license, maintenance, and community signals. Because SkillzBench is pre-launch, these scans describe what the platform is building and will offer at launch — join the waitlist for early access.

Question 2

What's the difference between SkillzBench's Trust score and its benchmark Lift?

Accepted Answer

The Trust score (0-100) reflects security, quality, license, maintenance, and community signals. Lift is a separate benchmark result: the same model tested in a sandbox with and without the skill, expressed in percentage points. The two are always shown separately, and money never touches scores.

Question 3

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

Accepted Answer

Launch scope covers instruction-only skills — SKILL.md packages plus text references. Script-bearing skills will be listed link-out only, so you can discover them but SkillzBench won't benchmark or scan them at launch.

Question 4

When can I actually start using the MCP server and per-agent loadouts?

Accepted Answer

SkillzBench is pre-launch and actively being built — the waitlist and newsletter are live today, the product is not yet public. Waitlist members get early access as the Discover, Verify, Benchmark, and Equip pillars come online, with a generous free tier planned at launch.

Question 5

How is this different from browsing a public MCP server directory or GitHub directly?

Accepted Answer

Directories list; SkillzBench proves. Instead of self-reported READMEs, you get commit-SHA-pinned versions, security scans, license checks, quality judging, a 0-100 Trust score, and measured Lift from sandbox benchmarking — plus per-agent loadouts through one MCP server rather than raw links you have to vet yourself.

MCP Server Discovery for Researchers