~/skill-gate — zsh — 96×40

skill-gate — the marketplace for Claude Code skills

the marketplace for Claude Code skills — browse, search, install. or just ask Claude to pick.

⌕ browse the marketplace → ✦ or let Claude pick ★ github
$ marketplace live catalogue

A curated library
for Claude Code skills.

Every skill is AI-reviewed for safety, scored 0–10, and admin-accepted before it goes public. Install with one git clone. Or just ask: /skill-gate find me a skill for <X>.

⌕ ⌘ K
  • — curated skills
  • — domains
  • 100% AI-reviewed
  • 0 source uploaded

01 browse by domain

click a card to filter the catalogue · 7 domains

02 all skills

—
sort by

$ let Claude pick the skills for you

Install one skill, type one prompt. /skill-gate reads your project + recent prompts and picks the matches from the catalogue above.

~/your-app $
thinking
  • scannext.js · tailwind · 14 components · Dockerfile · CI on vercel
  • behaviorrecent prompts: "polish the UI", "test the app", "shrink the image"
  • match20 candidates → top 4 (frontend, dx, devops)
  • installfrontend-design · webapp-testing · glance-gate · gstack
✓ installed 4 skills — usage explainers below
  • /frontend-design design layouts + components + responsive variants. Tailwind-aware. Use when you’re iterating on UI.
  • /webapp-testing end-to-end browser testing — flows, assertions, screenshots, regressions. Use before shipping.
  • /glance-gate rewrites your Dockerfile to a 5–60 MB image with a CVE scan and SBOM. Use when the image is over 100 MB.
  • /gstack headless browser for QA + dogfooding. ~100ms per command. Use when you want to verify a deploy.
→ need something else? /skill-gate find me a skill for ...
install in 10 seconds: git clone https://github.com/amineorion/skill-gate-skill.git ~/.claude/skills/skill-gate

$ how it works

01

read the repo

skill-gate scans your project — languages, frameworks, recent commits, what you keep asking Claude about. No upload; everything happens in your Claude Code session.

02

match the marketplace

Workflow signals go to /v1/recommend. The recommender ranks accepted skills by domain match, dependency overlap, and observed success rate across similar repos.

03

install + explain

The top picks are cloned into ~/.claude/skills/ and Claude prints a one-line usage explainer per skill. Nothing runs without your okay.

04

AI-reviewed in

Every skill in the marketplace was reviewed by Claude on the way in: scored, summarized, domain-tagged, safety-flagged. Admins accept before anything goes public.

05

local-first trust

The skill makes one call to the marketplace API: recommend. Everything else (scan, install, run) happens against your filesystem. Your code never leaves your machine.

06

community submissions

Built a skill? Submit it. The AI reviewer runs first, then an admin accepts or rejects. Accepted skills appear in the marketplace with attribution.

$ install

Three commands. The whole skill is <80 KB. The marketplace lives in the cloud — the skill talks to it via one API call.

git clone https://github.com/amineorion/skill-gate-skill.git ~/.claude/skills/skill-gate
# restart Claude Code, then type /skills to verify

Then in any project:

/skill-gate set me up based on what I'm working on

Or, for a specific domain:

/skill-gate find me a skill for marketing copy

$ trust model

  • ▸The skill never reads files outside the current project working directory.
  • ▸All AI inference runs in your Claude Code session against your Anthropic key.
  • ▸One outbound call: POST /v1/recommend with anonymized workflow signals (language, framework names, public dep names). Source is never uploaded.
  • ▸Skill installs are git clone from the canonical URL in the marketplace. You can pin or fork.
  • ▸Every accepted skill was AI-reviewed for prompt-injection patterns, shell-out safety, and outbound-network surface. The review is published on the skill's marketplace page.
~/skill-gate $ _