Question 1

How does SkillzBench detect prompt injection and malicious payloads in SKILL.md files?

Accepted Answer

Every skill in the Verify pillar is pinned to a commit SHA, validated, license-checked, security-scanned, and quality-judged before it earns a trust score. That security scan is designed to surface the prompt injection and credential-leak patterns that Snyk's ToxicSkills audit found in over a third of published skills. Skills that fail scrutiny simply don't get a clean score.

Question 2

What does the Lift number mean, and how is it different from the trust score?

Accepted Answer

Lift is the measured performance improvement a skill provides, expressed in percentage points, from benchmarking every skill in a sandbox with the same model — with and without the skill. The trust score (0-100) measures security, quality, license, maintenance, and community signals. SkillzBench always shows Lift separately from ratings, and money never touches scores.

Question 3

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

Accepted Answer

At launch, SkillzBench's verified scope is instruction-only skills — SKILL.md files plus text references. Script-bearing skills are listed link-out only, since executable code carries a far larger blast radius. This is a deliberate safety boundary while the verification pipeline matures.

Question 4

When can I actually start using SkillzBench?

Accepted Answer

SkillzBench is pre-launch and actively being built. The waitlist and newsletter are live now, and joining puts you first in line as the Discover, Verify, Benchmark, and Equip pillars roll out. Nothing about live skill counts or published benchmarks is being claimed yet — the product is being built in the open.

Question 5

How is SkillzBench different from browsing GitHub or ClawHub directly for Claude Code skills?

Accepted Answer

Public registries and GitHub offer raw distribution with little security vetting — ClawHub alone saw 800+ malicious packages and 76 confirmed malicious payloads during its growth. SkillzBench adds an independent verification layer: SHA-pinned versions, security scans, license checks, sandbox benchmarks with published Lift, and per-agent loadouts so each agent only sees what you've chosen and verified.

Verified Agent Skills for Researchers