v2.0.2: Anthropic-Aligned Rubric Redesign

What's New since v1.2.1

Breaking: New Scoring Rubric (v2.0.0)

• Completely redesigned scoring to align with Anthropic's official skill documentation
• New rubric categories reflect real-world skill quality signals

Fixes

• Excluded plm-cli/ from npm package (v2.0.1)
• Removed dead code, simplified async, fixed scoreScope budget
• Fixed incorrect test:ui command in README

Docs

• Updated README with real-world scores, working GitHub examples, and API field docs
• Full changelog: v1.2.1...v2.0.2

v1.2.1: Slim package

Added .npmignore to exclude source, tests, and assets from the published npm package.

• Package size: 4MB → 41KB (99% reduction)
• Published files: 35 (dist + README + LICENSE only)

Full Changelog: v1.2.0...v1.2.1

v1.2.0: CLI Quality Improvements

What's New

CLI Improvements

• Variadic paths — skillscore <path...> now accepts multiple paths natively, auto-entering batch mode when 2+ paths are given
• Explicit GitHub flag — Shorthand GitHub references (e.g. user/repo/skill) now require -g/--github flag, eliminating false positives from local paths that look like GitHub shorthands
• Testable error handling — CliError class replaces raw process.exit(1) calls, enabling programmatic testing and better error messages

Architecture

• Shared Reporter interface — All three reporters (TerminalReporter, JsonReporter, MarkdownReporter) now implement a common Reporter interface, exported for programmatic use
• Weight validation — Runtime assertion ensures scoring category weights always sum to 1.0, catching drift immediately on module load

Testing

• 6 new programmatic CLI tests — End-to-end tests covering error paths, valid skills, JSON output, verbose mode, and batch mode (60 total tests)

Docs

• Updated README with new CLI options, -g flag examples, and Reporter interface in API usage section

Full Changelog: v1.1.0...v1.2.0

v1.1.0 — Routing, Templates & Containment Scoring

What's New

5 new scoring sub-criteria inspired by OpenAI's Skills + Shell + Compaction blog and production data from Glean:

🎯 Scope — Routing Quality

• Negative routing examples (2pts) — Does the skill say when not to use it? Glean found triggering dropped ~20% without these.
• Routing-quality description (1pt) — Concrete routing signals (tool names, I/O, "use when") vs vague marketing copy.

📄 Documentation — Embedded Templates

• Embedded templates & worked examples (2pts) — Real output templates inside the skill, not just descriptions. Glean reported biggest quality gains from this pattern.

🔒 Safety — Network Containment

• Network containment (1pt) — If a skill uses HTTP/curl/fetch, does it mention allowlists or scoping? Flags the tools + networking combo risk.

📁 Structure — Artifact Output

• Artifact output spec (1pt) — Does the skill define where outputs/artifacts go?

No Breaking Changes

• All 8 categories unchanged
• All weights unchanged
• Point rebalancing within categories only
• 56 tests passing

npm install -g skillscore@1.1.0

SkillScore v1.0.0 — Initial Release

🚀 SkillScore v1.0.0

The universal quality standard for AI agent skills.

Features

• 8 weighted scoring categories: Structure, Clarity, Safety, Dependencies, Error Handling, Scope, Documentation, Portability
• Deterministic analysis — no API keys required
• GitHub URL support — score skills directly from GitHub repos
• Batch comparison mode — evaluate multiple skills side by side
• 3 output formats — Terminal (colorful), JSON, Markdown
• 56 passing tests with comprehensive coverage

Installation

npm install -g skillscore

Quick Start

# Local skill
skillscore ./my-skill/

# From GitHub
skillscore vercel-labs/skills/find-skills

# Batch compare
skillscore ./skill1 ./skill2 --batch

Scoring Methodology

Category	Weight
Safety	20%
Clarity	20%
Structure	15%
Dependencies	10%
Error Handling	10%
Scope	10%
Documentation	10%
Portability	5%

Works with any SKILL.md-based skill — skills.sh, ClaHub, GitHub, or local.