Apply AI
SkillScore: scoring AI agent skills
AI agents are steered by SKILL.md files, but nothing checked their quality. So I built a scorer.
- Problem
- Agent skill files (SKILL.md) vary wildly in quality, with no objective way to judge them or check them against the vendors' authoring guidance.
- Decision
- Built SkillScore, an offline Dart CLI that lints a SKILL.md and scores it 0 to 100 against the authoring guides for Claude, Codex, and Google Antigravity, with no network call.
- Result
- Shipped open source on pub.dev under Apache-2.0. Pointed it at a 52k-star repository to see what a high score looks like in the wild.
- Dart
- CLI
- AI agents
- pub.dev
As agent skills multiply, “is this skill any good” becomes a real question with no objective answer. SkillScore gives one: an offline Dart CLI that reads a SKILL.md, lints it, and scores it 0 to 100 against the authoring guides for Claude, Codex, and Google Antigravity, so the quality conversation moves from taste to numbers.
It is open source on pub.dev under Apache-2.0. I pointed it at a 52k-star repository to see what a high score looks like in the wild, and wrote up the result.