Question 1

What are agent skills, and why do they need governance in an enterprise?

Accepted Answer

Agent skills are SKILL.md instruction packages that extend coding and chat agents like Claude Code, Codex, Cursor, and Gemini CLI. Because skills inherit the host agent's permissions, an unvetted skill can read internal data or invoke tools without extra authorization. SkillzBench is building an independent testing lab and control panel that verifies, benchmarks, and curates skills before they reach your agents.

Question 2

How does SkillzBench detect malicious agent skills before they reach production?

Accepted Answer

SkillzBench's Verify pillar pins every skill version to a commit SHA, then runs validation, license checks, security scans, and quality judging. Each skill also receives a 0-100 trust score built from security, quality, license, maintenance, and community signals. Script-bearing skills are listed link-out only at launch, so instruction-only SKILL.md packages get the deepest inspection.

Question 3

How does the benchmark actually prove a skill helps my agent?

Accepted Answer

SkillzBench's Benchmark pillar tests every skill in a sandbox using the same model, running tasks both with and without the skill, and publishes the Lift in percentage points. Lift is always shown separately from ratings and trust scores, and money never touches scores — so you're seeing measured impact, not sponsored placement.

Question 4

Can we control which skills each individual agent can use?

Accepted Answer

Yes — that's the Equip pillar. SkillzBench provides one MCP server plus per-agent loadouts: your team toggles which skills and tools each agent gets, and the agent only sees what was chosen. This gives you least-privilege access per agent rather than an all-or-nothing skill install. Teams-tier features like org loadouts, allowlists, and audit are planned as part of the pricing roadmap.

Question 5

How is SkillzBench different from just installing skills from GitHub or a marketplace directly?

Accepted Answer

Marketplaces distribute skills; SkillzBench tests them. Instead of trusting a README, you get version pinning to a commit SHA, license checks, security scans, quality judging, a 0-100 trust score, and measured Lift benchmarks in a sandbox with the same model. SkillzBench is also independent — it doesn't sell skills, and money never touches scores.

Curated Agent Loadouts for Enterprise Teams