SkillProof tested Claude skills

Skill Evaluator

Betygsätter en SKILL.md mot Anthropic:s bästa praxis med ett valideringsskript och en viktad bedömningsmatris

av gotalab · gotalab/skillport

Testad · Fungerar ★ 8.4/10

Skill Evaluator — Betygsätter en SKILL.md mot Anthropic:s bästa praxis med ett valideringsskript och en viktad bedömningsmatris

Vad den gör

Utvärderar an agent skill's SKILL.md against Anthropic's authoring best practices, combining an automated Python validator (naming, description length, body length, reference depth) with a six-dimension weighted rubric that yields a 1-5 score and a templated report. Utlöses when the user asks to review, evaluate, assess, or audit a skill for quality. Producerar per-dimension scores, detected anti-patterns, and prioritized fix recommendations.

Testrapport

Klonade repot och körde scripts/validate_skill.py against a deliberately weak test skill (name "pdf", vague description); it emitted valid JSON flagging the non-gerund name and missing activation triggers, exit 0. Jag märkte a real gap: the path_style check only inspects markdown-link references, so the bare Windows path "references\extra.md" was NOT flagged. Sedan producerade jag two artifacts: a baseline free-form review (6 lines, no score) vs a skill-followed report (weighted 1.50/5.0 with a per-dimension table, reproduced validator warnings, and a "Poor" recommendation band) - the skill demonstrably adds a reproducible number and forces the checks to run. Alla 3 refererade filer returnerade HTTP 200; inga security smells.

Testad: 2026-07-21 · Claude Code 2.x (agent harness)

Installation

git clone --depth 1 https://github.com/gotalab/skillport.git /tmp/skill-evaluator-src
mkdir -p ~/.claude/skills
cp -R /tmp/skill-evaluator-src/.skills/experimental/skill-evaluator ~/.claude/skills/skill-evaluator
# No deps: scripts/validate_skill.py is pure Python 3 stdlib (argparse, re, json, pathlib).
# Run the validator: python3 ~/.claude/skills/skill-evaluator/scripts/validate_skill.py <path/to/skill> --json

Kommandon och exempelprompter

  • /skill-evaluatorBetygsätter en SKILL.md mot Anthropic:s bästa praxis med ett valideringsskript och en viktad bedömningsmatris

Skills triggas av vanliga förfrågningar — inga kommandon att memorera. Efter installationen aktiverar prompter som dessa skillen (på engelska):

  • Evaluate this skill's SKILL.md for quality
  • Review my skill description for clarity
  • Audit this skill for structural anti-patterns

Läs hela rapporten →