Question 1

How does SkillzBench measure whether an agent skill actually improves performance?

Accepted Answer

SkillzBench is building a benchmarking pillar that tests every skill in a sandbox with the same model, running the same tasks with and without the skill. The result is published as a Lift in percentage points, so you see exactly how much a skill adds. Lift is always shown separately from ratings, and money never touches scores — the benchmark stays independent.

Question 2

What kinds of agent skills will SkillzBench support at launch?

Accepted Answer

Launch scope is instruction-only skills — SKILL.md packages plus text references — for agents like Claude Code, Codex, Cursor, Gemini CLI and chat-app agents. Script-bearing skills will be listed link-out only, so you can find them but they won't be benchmarked in the sandbox at launch.

Question 3

How do you handle skill security, given reports that a quarter of skills carry vulnerabilities?

Accepted Answer

Every skill goes through the Verify pillar: each version is pinned to a commit SHA, then validated, license-checked, security-scanned and quality-judged. Skills also receive a 0-100 Trust score built from security, quality, license, maintenance and community signals, so you can filter by risk before loading anything into a production agent.

Question 4

Can I control which skills each of my agents uses?

Accepted Answer

Yes — the Equip pillar is built around one MCP server plus per-agent loadouts. You toggle skills and tools per agent, and the agent only sees what was chosen. When upstream skills change, SkillzBench tracks those changes, re-scans, re-benchmarks and notifies you, so your approved loadout never drifts silently.

Question 5

How is SkillzBench different from just browsing GitHub for popular skills?

Accepted Answer

GitHub shows stars, not evidence. SkillzBench continuously crawls and dedupes skills from GitHub, then does what GitHub can't: pins every version to a commit SHA, runs license checks and security scans, quality-judges each skill, benchmarks it with and without the skill in a sandbox, and publishes the Lift. You get a Trust score and a measured Lift instead of a popularity contest — and ratings are never influenced by money.

Agent Skill Benchmarking for Founders