From idea to published, installable AI skill — in one pipeline.
Say "publish this skill" to your AI coding assistant and Skill Forge handles the rest — from writing SKILL.md to pushing a ready-to-install repo to GitHub.
Agent Skills compatible — works with Claude Code, Cursor, Codex, Windsurf, Gemini CLI, GitHub Copilot, and more.
You wrote a great skill. It works locally. But:
- It's trapped in one project's skills directory — not a proper, installable package
- You can't share it across machines, let alone with other people
- It doesn't meet any platform's install standard — no README, no LICENSE, no proper structure
- There's no versioning, no discoverability, no community presence
- Publishing means figuring out GitHub repo setup, README conventions, symlink registration, and community platform requirements — all manually
Your skill is trapped. It can't be maintained globally, iterated on independently, or shared with the community.
Takes a skill idea (or an existing project-local skill) and runs the full pipeline:
Config → Gather → Create → Validate → Publish
- Config — Set up
~/skills/root, detect your GitHub org and preferences (auto-defaults, minimal questions) - Gather — Auto-detect existing skill content from project and conversation. Detect what the skill needs (first-use setup? persistent state? enforceable constraints?)
- Create — Write SKILL.md following the Agent Skills standard, bake detected capabilities in
- Validate — Structure, frontmatter, content quality, optional community readiness checks
- Publish —
git init, register symlinks, push to GitHub. Community platforms auto-index public repos
The result: a standalone repo that anyone can install with one command.
Skill Forge also guides you through organizing multiple skills — whether as independent repos bundled into a themed Kit (authoring-kit, quality-kit), or as a single-repo Collection. See the publishing strategy reference for details.
Say any of:
- "Create a skill for X and publish it"
- "Publish this skill to GitHub"
- "Forge a skill from my notes"
- "Turn this project-local skill into a repo"
$ "Publish self-review to GitHub"
Step 0: Config
✓ ~/.config/skill-forge/config.md found
✓ skill_root: ~/skills/, github_org: motiful
Step 1: Gather
✓ Existing skill detected at ~/skills/self-review/SKILL.md
✓ Capabilities: none needed (pure methodology, no state or onboarding)
Step 2: Create
✓ SKILL.md already exists — using as-is
Step 3: Validate
✓ name: self-review (kebab-case, 11 chars)
✓ description: single-line, 133 chars
✓ body: 226 lines (< 500)
✓ references/dimensions.md exists and is linked
✓ no junk files in skill content
Step 4: Publish
✓ git init + initial commit
✓ ~/.claude/skills/self-review → ~/skills/self-review/
✓ ~/.agents/skills/self-review → ~/skills/self-review/
✓ gh repo create motiful/self-review --public --source=. --push
✓ Published — install with: npx skills add motiful/self-review
- Git (required)
- Node.js (required for
npx skills add) - GitHub CLI (
gh) — recommended for one-command publishing. Without it, you'll set up the remote manually
npx skills add motiful/skill-forgeOr manually:
git clone https://github.com/motiful/skill-forge ~/skills/skill-forge
# Claude Code
ln -sfn ~/skills/skill-forge ~/.claude/skills/skill-forge
# Other platforms (Cursor, Codex, Windsurf, Gemini CLI, Copilot)
ln -sfn ~/skills/skill-forge ~/.agents/skills/skill-forgeSKILL.md — Full creation + publishing pipeline
references/
├── skill-format.md — How to write a valid SKILL.md
├── publishing-strategy.md — One skill, multiple skills, or a themed Kit
├── skill-composition.md — Why skills need composition (like npm needs dependencies)
├── platform-registry.md — Where each platform looks for skills
├── onboarding-pattern.md — Adding first-use setup to a skill
├── state-management.md — Persistent config and state across sessions
├── constraint-companion.md — Separating enforceable rules into a companion skill
└── templates.md — README, LICENSE, .gitignore templates