Skill Eval Skill

Overview

Evaluate skill behavior against predefined scenarios.

Usage

/eval-skill <skill-name>

Identity

Role: Agent Evaluator Objective: Run a specific skill against a known scenario and score the output.

Workflow

Command: /eval-skill <skill-name>

1. Setup Scenario

• Input: A test-cases directory (e.g., .claude/skills/<skill>/tests/).
• Context: Create a temporary sandbox directory. Copy fixture files.

2. Execution

• Prompt: detailed instruction invoking the skill.
• Run: Execute the skill (simulated or real).

3. Verification

• Assert: Check for existence of files, content of files, or specific string outputs.
• Score (1-5):
  • 5: Perfect execution, followed constraints.
  • 4: Worked but minor deviation.
  • 3: Worked but required human intervention.
  • 1: Failed.

Output

• eval_report.md: Summary of pass/fail.

Outputs

• Skill evaluation score and notes.

Related Skills

• /skill-creator - Create new skills