Traction scoreGitHub stars can be faked, so popularity alone can be misleading. Traction Score looks for broader signs of recent attention, adoption, and active maintenance.
SkillForge provides a structured framework for AI agent builders to develop and audit skills. It focuses on an evidence-driven approach where a skill is considered complete only when an agent demonstrably performs better with it than without it.
The system assists in the full lifecycle of skill management, from initial creation and testing to ecosystem health checks. It helps developers identify duplicate skills, resolve trigger collisions, and remove stale entries to maintain a clean operational environment.
WHO IT'S FOR
AI agent / automation builders
creating and validating evidence-driven agent skills
Compatible AgentsThe repository documents support for these agents. The skills may also work with other agents that can load SKILL.md files, but they may need some setup or small changes.
Follow the documented setup, then try a first task.
I’m using
Manual Claude Code installation
Installs: Installs the SkillForge skill bundle for Claude Code, with an optional /skillforge command and optional interactive advisor configuration. · Claude Code
Before you start
Claude Code
Python 3.8+
git
In a terminal, clone the SkillForge repository into the documented temporary location.
Installs: Installs the complete SkillForge skill bundle into the documented Claude Code, Codex, and shared .agents skill locations; it also installs the Claude Code /skillforge command and configures a safe proactivity level without hooks or Personal Context scanning. · Claude Code, Codex CLI
Before you start
Claude Code or Codex CLI
Python 3.8+
git
rsync
An existing SkillForge repository checkout containing scripts/install_workshop.sh
Run from: SkillForge repository root
In a terminal from the SkillForge repository root, run the workshop installer. Replace the literal [options] placeholder with any desired installation flags, or remove it to use the default balanced level. Supported installation substitutions are --level LEVEL, --source PATH, and --force; replace LEVEL with off, quiet, balanced, or active, and PATH with an existing SkillForge checkout path. Do not substitute -h or --help because those flags only display usage and exit.
bash scripts/install_workshop.sh [options]
Restart Claude Code and Codex after the installer completes so the new skill loads.
Replace <request> with the skill-related request you want assessed.
Use the skillforge skill to triage this skill-related request: "<request>". Refresh the skill index if needed, classify the result as a strong, moderate, or weak match, and recommend only the next action—use, improve, create, compose, or clarify. Do not create or modify a skill yet.
A SkillForge repository checkout containing scripts/tests/fixtures/sample-skill and its validator scripts.
Check the repository for this skill’s setup.
Use the sample-skill skill to validate the fixture at scripts/tests/fixtures/sample-skill. Report whether the validator's error list is empty, whether user-invocable remains a boolean, and whether allowed-tools parses as a three-item list. Do not modify any files.