Question 1

How do I know if an agent skill is safe to run in my SOC environment?

Accepted Answer

Published research found roughly of agent skills carry at least one vulnerability. SkillzBench verifies every skill version by pinning it to a commit SHA, then running validation, license checks, security scans, and quality judgment. Each skill receives a trust score from 0-100 built from security, quality, license, maintenance, and community signals.

Question 2

Can SkillzBench benchmark whether a skill actually improves our alert triage workflow?

Accepted Answer

Yes — that's the core of the Benchmark pillar. Every skill is tested in a sandbox with the same model, once with the skill enabled and once without, and the Lift is published in percentage points. Lift is always shown separately from ratings, and money never touches scores, so you get an honest measure of impact.

Question 3

We run agents across Claude Code, Cursor, and Gemini CLI — does SkillzBench cover all of them?

Accepted Answer

SkillzBench targets agent skills (SKILL.md instruction packages) for Claude Code, Codex, Cursor, Gemini CLI, and chat-app agents. At launch the scope is instruction-only skills; script-bearing skills are listed link-out only. Distribution to your agents happens through one MCP server with per-agent loadouts.

Question 4

SkillzBench isn't launched yet — how do we get the skill audit and risk report?

Accepted Answer

SkillzBench is pre-launch and actively being built. The waitlist and newsletter are live now, so joining is the fastest path to early access. Once available, the audit process identifies your skills, verifies each version, and delivers trust scores plus benchmarked Lift in a risk report.

Question 5

How is this different from just reviewing SKILL.md files ourselves or relying on GitHub stars?

Accepted Answer

Manual review doesn't scale and GitHub stars say nothing about security or actual performance. SkillzBench provides continuous discovery with dedupe, commit-pinned verification with security scanning, sandboxed benchmarks that quantify Lift, and per-agent loadouts for least-privilege control — plus upstream change tracking with re-scan and re-benchmark notifications when skills change under you.

Skill Audit & Risk Report for Data Analysts