Question 1

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

Accepted Answer

SkillzBench benchmarks every skill in a sandbox using the same model, running the task twice — once with the skill and once without — and publishes the difference as a Lift in percentage points. Lift is always shown separately from ratings, so you see measured performance rather than a blended opinion score.

Question 2

What's the difference between a trust score and a Lift number?

Accepted Answer

The trust score (0-100) aggregates security, quality, license, maintenance, and community signals — it tells you whether a skill is safe and well-maintained. Lift is a benchmark result showing how much the skill improved task performance in a controlled test. Keeping them separate means a well-marketed skill can't hide behind a good rating, and money never touches scores.

Question 3

Does SkillzBench support script-bearing skills and all agent frameworks?

Accepted Answer

Launch scope is instruction-only skills — SKILL.md packages plus text references. Script-bearing skills are listed link-out only, so you know they exist but SkillzBench doesn't serve them. Skills are built for the SKILL.md format used by Claude Code, Codex, Cursor, Gemini CLI, and chat-app agents.

Question 4

How do per-agent loadouts work for a small team managing multiple agents?

Accepted Answer

You equip agents through one MCP server. For each agent, you toggle which skills and tools it gets, and the agent only sees what was chosen — no context bloat, no accidental tool access. When an upstream skill changes, SkillzBench re-scans, re-benchmarks, and notifies you so loadouts stay current.

Question 5

Why not just install skills directly from GitHub like we do now?

Accepted Answer

Direct installs mean no version pinning, no security scan, and no idea whether the skill helps or hurts — published research found roughly of skills carry at least one vulnerability. SkillzBench dedupes the GitHub sprawl, pins every version to a commit SHA, security-scans and license-checks each skill, and benchmarks it before you equip it, turning ad-hoc installs into a governed process.

Curated Agent Loadouts for Small Businesses