Question 1

How does SkillzBench handle skills built for Claude Code when we also run Codex and Cursor?

Accepted Answer

Agent Skills use the SKILL.md open standard now supported by 25+ frameworks, so the same instruction package can serve multiple tools. SkillzBench verifies and benchmarks each skill once, then lets you attach it to any agent through per-agent loadouts on a single MCP server.

Question 2

What exactly does the Lift score tell us before we deploy a skill to a client agent?

Accepted Answer

Lift is the measured performance difference — in percentage points — from testing every skill in a sandbox with the same model, with and without the skill. It's always shown separately from ratings and trust scores, and money never touches the scores.

Question 3

Are script-bearing skills supported, or only instruction packages?

Accepted Answer

At launch, SkillzBench covers instruction-only skills — SKILL.md plus text references. Script-bearing skills are listed link-out only, so you always know exactly what execution surface a skill exposes before it goes anywhere near production.

Question 4

When can my agency actually start using SkillzBench?

Accepted Answer

SkillzBench is pre-launch and actively being built. The waitlist and newsletter are live now, so you can reserve early access and follow progress while the Discover, Verify, Benchmark, and Equip pillars come online.

Question 5

How is this different from just browsing Anthropic's public skills repo on GitHub?

Accepted Answer

A public repo gives you raw, unvetted packages — and published research found ~of skills carry at least one vulnerability. SkillzBench dedupes, pins versions to commit SHAs, runs security scans and license checks, benchmarks for Lift, and tracks upstream changes so you know when a skill you depend on has shifted.

Agent Skill Migration for Agencies