Add eval runner script

[mattpodwysocki]

Summary

• Adds scripts/run-evals.js — runs evals for any skill against Claude with and without the SKILL.md as system context, grades each expectation using Claude as a judge, and reports pass rates and delta
• Adds npm run eval <skill-name> script
• Adds @anthropic-ai/sdk as a dev dependency
• Updates CONTRIBUTING.md to replace the skill-creator eval command (not yet implemented in any published version) with the actual npm run eval command, and clarifies the difference between knowledge evals and tool-execution evals

Usage

export ANTHROPIC_API_KEY=your-key-here
npm run eval mapbox-location-grounding

Output

Running evals for: mapbox-location-grounding
Model: claude-sonnet-4-6
Evals: 8

Eval 1: What restaurants are near -87.6298, 41.8781?
  Without skill: 17%  |  With skill: 50%  |  Delta: +33pp
  ...

Overall Results:
  Without skill (baseline): 21.6%
  With skill:               59.5%
  Delta:                    +37.8pp

  ✅ Strong skill (+20pp target met)

🤖 Generated with Claude Code