FAQ
Quick answers — or ask the assistant anything.
What is SkillzBench?
The independent testing lab and control panel for agent skills. Every skill is discovered from GitHub, pinned to a commit, security-scanned, quality-judged, and benchmarked with and without itself — then served to your agent through one MCP you control.
What is Lift?
Pass-rate with the skill minus pass-rate without it, in percentage points, with a 95% confidence interval. It's the headline number: evidence, not opinion. Negative Lift is published with the same prominence as positive.
When does it launch?
We're in the lab now. The waitlist is open — join it on the home page and you'll be first through the door when the bench opens.
Which agents are supported?
Anything that speaks MCP: Claude Code, Codex, Cursor, Gemini CLI, and chat apps with remote MCP connectors. One loadout, every client.
Is it free?
The plan is a generous free tier: browse, search, one agent, up to 15 enabled skills. Paid plans add more agents, org loadouts and author tools — and money never touches scores.
How do you keep skills safe?
Static and semantic security scans on every pinned version, findings published with file and line, canary-credential tests in benchmark sandboxes, and instant delisting when something fails. Only scanned versions are ever served.