skillab

Tools for working with Agent Skills (SKILL.md files)

Skill Input

Load one or more SKILL.md files — from a GitHub repo, a folder of skills, or paste directly

Supports: owner/repo · https://github.com/owner/repo · .../tree/main/skills · .../blob/main/SKILL.md

Top repos on skills.sh:

Tool

Test how well models trigger and follow the skill instructions

Provider

API Key

Keys are sent directly to the provider, never stored

Test Models

Each selected model will be evaluated independently against the skill

1 selected: llama-3.3-70b-versatile

Advanced Options— 5+5 prompts | Generator: llama-3.3-70b-versatile | Judge: llama-3.3-70b-versatile | Tools: 14

Prompts per type (positive = should trigger, negative = should not)

5 + 5 = 10 total

Verbose output

Generator Model

Generates test prompts from the skill definition

llama-3.3-70b-versatile

Judge Model

Evaluates trigger accuracy and instruction compliance

llama-3.3-70b-versatile

Available Tools

Tools provided to the model during compliance testing. Auto-detected from the skill content. The model can make structured tool calls against these (with mock responses).

14 tools enabled

Custom Test Prompts

Optional — override auto-generated prompts with your own. Leave empty to let the generator model create them.