Why Skillry
Most Claude Code resource repos are link lists (you still copy files by hand) or are
Claude-only. Skillry is different on five axes:
| Skillry | Typical "awesome" list | Typical CLI installer |
|---|
| Installs actual skill/agent files (not links) | ✅ | ❌ | ✅ |
| Multi-platform (Claude + Codex + Copilot + Gemini) | ✅ | ❌ | ❌ (Claude only) |
Per-agent permission boundaries (least-privilege tools) | ✅ | ❌ | ⚠️ |
| Validation harness (structure + frontmatter lint + permission + lockfiles) | ✅ | ❌ | ⚠️ |
| Skill-sync: discover (license + risk scan), normalize (frontmatter + provenance), vet (staged, attributed, never auto-enabled) | ✅ | ❌ | ❌ |
| Native plugin marketplace (sha-pinned, reproducible) | ✅ | ❌ | ⚠️ |
| Full upstream attribution for redistributed content | ✅ | n/a | ⚠️ |
These directly reflect Anthropic's own guidance: least-privilege tools, single-responsibility
subagents, and auditing third-party skills before use.
Quickstart
Claude Code (native plugin marketplace — recommended)
# In Claude Code:
/plugin marketplace add FluxonLab/Skillry
/plugin install core-operations@skillry
Browse all departments with /plugin marketplace after adding.
npm / npx (any platform, no clone, no publish needed)
# Runs straight from GitHub — no global install, no npm account required:
npx github:FluxonLab/Skillry install # dry-run, all platforms
npx github:FluxonLab/Skillry install --apply --targets claude # or: codex copilot antigravity
npx github:FluxonLab/Skillry install --apply --targets claude --community # include attributed 3rd-party skills
# Or install the CLI globally:
npm install -g skillry
skillry install --apply --targets claude codex
skillry validate
Portable installer (from a clone)
git clone https://github.com/FluxonLab/Skillry
cd Skillry
python3 tools/install.py # dry-run: shows what will be installed
python3 tools/install.py --apply --targets claude # or: codex copilot antigravity
The installer rewrites machine-specific paths to your $HOME and backs up any existing
config (*.bak-skillry) before writing. Dry-run is the default. Verify with
python3 tools/validate.py.
What's inside
- 125 original skills across 18 departments — concrete procedures, real commands,
checklists, and safety rules (not generic templates); median ~150 lines of substance.
- 73 original subagents with explicit
tools allowlists and read-only vs. write scopes.
- A curated
community/ set — 98 skills + 49 agents from 6 permissively-licensed sources,
redistributed with full attribution (MIT/ISC only — see NOTICE and
THIRD-PARTY-NOTICES).
- A flagship
CLAUDE.md — a project-agnostic engineering operating manual
(inspect-first, surgical changes, security, i18n + theme parity, dev-launch defaults, honest
verification). It governs this repo and is written to be copied into your own project as a
strong default: cp CLAUDE.md /path/to/your/repo/CLAUDE.md.
Departments (18)
Core Operations · Runtime & Local App · Backend & API · Frontend & Web Design ·
Mobile & Desktop · Gaming & Interactive Media · Database & Data · AI & Agent Systems ·
Security · Testing & QA · DevOps & Release · Product, Docs & Research ·
Documentation & Tech Writing · Data & ML / AI Engineering · Performance & Cost ·
Cloud & Infrastructure · Skill Library & Installation · Optional Specialists
Multi-platform
