Question 1

How does SkillzBench handle MCP tool poisoning and prompt injection risks in agent skills?

Accepted Answer

SkillzBench's Verify pillar security-scans every skill and feeds those signals into a 0-100 trust score alongside quality, license, maintenance, and community signals. Launch scope is instruction-only skills (SKILL.md plus text references); script-bearing skills are listed link-out only, precisely because they carry higher execution risk. SkillzBench is pre-launch, so scanning capabilities are what we're building and will offer at launch.

Question 2

Can SkillzBench replace my bespoke MCP registry or gateway for skill discovery?

Accepted Answer

SkillzBench isn't a gateway — it's an independent testing lab and control panel that sits alongside your MCP setup. The Discover pillar continuously crawls and dedupes skills from GitHub, Verify pins every version to a commit SHA, and Equip delivers one MCP server with per-agent loadouts. You get a curated, verified catalog instead of raw registry listings where more than half of entries can be invalid or low-value.

Question 3

What does the benchmark actually measure, and can I trust the results?

Accepted Answer

Benchmark tests every skill in a sandbox with the same model, with and without the skill, and publishes the Lift in percentage points. Lift is always shown separately from trust ratings, and money never touches scores — so a skill's measured contribution is never conflated with popularity or sponsorship. Note that SkillzBench is pre-launch; benchmarking is what we're building, not yet a published results library.

Question 4

When can I actually use SkillzBench, and what will it cost?

Accepted Answer

SkillzBench is pre-launch and actively being built — the waitlist/newsletter is live now. Our pricing hypothesis is a generous free tier, Pro at roughly $9-15/mo for more agents and unlimited enabled skills, Teams at roughly $15-25/seat/mo for org loadouts, allowlists and audit, and Creator at roughly $5-10/mo for verified authors. Join the waitlist for launch timing and early access.

Question 5

How is SkillzBench different from just searching GitHub or existing MCP registries myself?

Accepted Answer

Raw discovery is the problem: of 17,630 raw crawled MCP entries in one measurement study, only 8,401 were valid projects. SkillzBench dedupes continuously, pins every skill version to a commit SHA, license-checks and security-scans each one, quality-judges it, benchmarks it in a sandbox, and re-scans and re-benchmarks when upstream changes — then notifies you. A registry lists; SkillzBench verifies and measures, and keeps your per-agent loadouts current.

MCP Server Discovery for Founders