skill-forge

A toolkit for authoring and evaluating Claude Code skills: create-skill scaffolds a new skill test-first against a failing baseline, and eval-skill runs the Tier 0-3 harness that gates it before shipping.

Install

Install at project scope so the skills are checked into .claude/settings.json and shared with every collaborator who clones the repo.

Option 1: Claude Code /plugin

This repo is both a Claude Code plugin and its own marketplace, so no separate distribution repo is needed.

/plugin marketplace add neokn/skill-forge
/plugin install skill-forge@skill-forge --scope project

After installing, the two skills appear namespaced as skill-forge:create-skill and skill-forge:eval-skill.

The plugin source is the whole repo root, so installing also downloads runbooks/, src/, and the ~486MB of compiled binaries under skills/*/bin/ for all six supported platforms, even though any given install only uses one of them.

Option 2: npx skills

skills CLI is a general-purpose skill installer that doesn’t depend on the Claude Code plugin system, and works across Claude Code, Codex, Cursor, and dozens of other agents. This repo is a monorepo with two skills (skills/create-skill, skills/eval-skill) and no root-level SKILL.md, so target them with --skill or a direct subdirectory path. Project scope is the tool’s default, so no extra flag is needed:

# See what's available in the repo
npx skills add neokn/skill-forge --list

# Install both skills to Claude Code
npx skills add neokn/skill-forge --skill '*' -a claude-code

# Install just one
npx skills add neokn/skill-forge --skill eval-skill -a claude-code

# Or target a single skill's subdirectory directly
npx skills add https://github.com/neokn/skill-forge/tree/main/skills/eval-skill

Contents

• skills/create-skill/ — scaffold a new skill test-first
• skills/eval-skill/ — run the Tier 0-3 evaluation harness against an existing skill