Test a skill in Skill Studio

You can run a skill's test with and without the skill in Skill Studio and tell which answer passed which checks.

Studio01:06Recorded 18 Oct 2026

You need: Node.js 20 or later, for npx vskill studio; A model provider: Claude Code, an Anthropic or OpenAI API key, OpenRouter, Ollama or LM Studio

Steps

  1. Put the skill at skills/<name>/SKILL.md and its tests at skills/<name>/evals/evals.json.
  2. Run npx vskill studio in the repo and open the localhost address it prints.
  3. Pick the skill, open Tests and read the prompt and assertions.
  4. Choose the model in the model picker.
  5. On Run, click Run A/B Test.
  6. Read the assertion results, Skill vs Baseline, and the blind comparison. Results vary, so run it again before you change the skill.
Transcript

Test a skill in Skill Studio.

This sample repo has one skill, `commit-message`, and its test file. Start Skill Studio in the repo with `npx vskill studio`. It runs on this machine, at a localhost address.

Pick the skill in the sidebar, then open Tests. The test is a real task: write a commit message for this diff. Four checks in plain English say what a good answer looks like.

Choose the model that runs the test. Here, Claude Opus 4.8 through the Anthropic API.

On Run, click Run A/B Test. Studio answers the task twice: with the skill, and without it. The same model then checks both answers against every assertion.

With the skill, all four checks pass. Without it, two of four. A blind comparison of the two answers also preferred the skill. Results vary between runs, so run it again before you change the skill.

Next: read the Skill Studio docs.

The narration as voiced in the video.

Learning path

Write and ship a skill

Install skills with the CLI, scan them, then test yours in Studio.

  1. Find and install a verified skill

    01:13
  2. Understand what the security scan checks

    01:18
  3. Test a skill in Skill Studio

    You are here
  4. Publish a skill, public or private

    01:03

Plan the work that uses your skills: SpecWeave tutorials on spec-weave.com.

Test a skill in Skill Studio | Verified Skill