Rubric Evaluator

Rubric Evaluator is a plugin that grades a Claude Code / Codex skill directory (one containing a SKILL.md) against a 6-section, 31-item rubric. Structural and safety checks run deterministically as a stdlib-only Python script, while semantic checks are applied from a bundled rubric reference. The headline grade (S/A/B/C/F) is computed deterministically — never eyeballed.

The rubric is based on Toss Tech’s article Designing a Rubric for Skill Quality Management.

The same skill bundle ships for both Claude Code and Codex from a single source of truth (plugins/rubric-evaluator/), installable in one line from each host’s marketplace.

Features

• Grades skills against a 6-section, 31-item rubric
• Deterministic structural and safety checks via a stdlib-only Python script (safety violations like secret leaks grade F immediately)
• Per-item report explaining why each failing item matters and how to fix it
• Dual-host support — one source ships both Claude Code and Codex plugin manifests
• Triggers on natural-language requests like “grade ./my-skill”