Question 1

How does SkillzBench handle prompt injection and other LLM security risks hidden in skill instructions?

Accepted Answer

SkillzBench's Verify pillar security-scans every skill version, validates it, and quality-judges it before assigning a trust score. Because instruction-only skills (SKILL.md plus text references) are the launch scope, the full instruction text is inspectable and scannable for injection-style directives — and the 0-100 trust score surfaces security signals so risky skills are visible before you load them.

Question 2

What does the trust score actually measure, and can vendors pay to improve it?

Accepted Answer

The trust score runs 0-100 and combines security, quality, license, maintenance, and community signals. No — money never touches scores, and Lift is always published separately from ratings. SkillzBench is an independent testing lab, not a marketplace, so there's no incentive to inflate a score.

Question 3

Can I control which skills each of my team's agents can use?

Accepted Answer

Yes — that's the Equip pillar. Users toggle skills and tools per agent through one MCP server, and the agent only sees what was chosen. For organizations, the Teams plan adds org loadouts, allowlists, and audit, giving security teams a policy enforcement point over agent capabilities.

Question 4

When can I start using SkillzBench, and what's the rollout timeline?

Accepted Answer

SkillzBench is pre-launch and actively being built — the waitlist and newsletter are live now, but the product is not yet public. Joining the waitlist gets you early access as the pillars ship, starting with Discover (crawling and deduping skills from GitHub) and Verify (scanning, pinning, and scoring).

Question 5

How is this different from just reading a skill's SKILL.md myself before installing it?

Accepted Answer

Manual review doesn't scale and doesn't watch for change. SkillzBench continuously crawls and dedupes skills from GitHub, pins every version to a commit SHA, runs security scans and license checks, and re-scans and re-benchmarks when upstream changes — so a skill that was safe when you reviewed it can't silently become a risk later. Benchmarking also shows you whether the skill actually helps, via measured Lift.

Agent Security Scanning for Researchers