Agentic Awesome Skills (AAS Core)

1What it does

The biggest skills library in the ecosystem — 2,600+ installable SKILL.md playbooks — wrapped in an actual management layer called AAS Core rather than a bare folder dump. Core is a local, agent-first control plane: a CLI (npx agentic-awesome-skills), a local MCP server your agent can query to search the complete catalog, stack validation, and a plan-preview step that lets you inspect what the agent selected before anything changes your target project. The catalog itself is deliberately non-opinionated — Core does not rank or recommend skills, it indexes them and records what the agent chose. Install targets cover Antigravity, Claude Code, Cursor, Codex CLI and Gemini CLI from the same catalog, which makes it the closest thing to a cross-agent skill registry that exists. Apply and recovery operations are still flagged experimental in the README — the inspection and selection side is the production-ready half.

2When to use it

Reach for it when you want the search problem solved, not just the skills — at 2,600+ entries the bottleneck is discovery, and the local MCP server means your agent does the catalog search instead of you browsing a README. It fits projects that want a repeatable install workflow (select → validate → preview → apply) over ad-hoc folder copying, and teams evaluating skills across multiple agent hosts at once. Skip it if you want a small curated set — this is a warehouse, not a boutique; the breadth is the point and also the audit surface. Solo users who already know the three skills they want gain little over a manual copy.

3How to get started

The per-target install is one npx command: npx agentic-awesome-skills --gemini for Gemini CLI, --cursor for Cursor, --codex for Codex CLI, with Antigravity and Claude Code supported through the same CLI or the local MCP path. First use is conversational — ask the agent to choose and compose a skill stack (the README's example is @brainstorming help me plan a feature), and Core records the selection for review. Before any apply step, read the AAS Core preview guide in the repo — it documents the exact trust boundary: what the plan preview shows, what apply changes, and why recovery is still experimental.

4Configuration

The MCP server runs locally — catalog search happens on your machine against a bundled index, not a hosted API, which keeps skill selection private and offline-capable. Skills that Core installs land in the standard discovery locations for each host (.agents/skills/ for Antigravity), so they interoperate with skills installed any other way. The versioned guides in the repo matter: the README tracks main while published packages lag, so configure against the release you actually installed. Browser-local Workbench is a companion review surface — useful for humans inspecting the catalog, not required for the agent path.

5Things to know

• ~50k stars across the ecosystem (the biggest reach of any skill collection) — but the project's own positioning is anti-ranking: Core never tells you which skills are good, only which exist.
• Apply/recovery explicitly experimental — the safe posture today is inspect + select through Core, apply by hand or with review.
• A skill's install is only the start: 2,600 candidates means the description frontmatter quality varies; validate the specific skills you enable rather than trusting catalog breadth.
• Multi-target by design — the same skill folder serves Antigravity, Gemini CLI, Claude Code and friends, so a skills decision here travels with you between agents. - The catalog's breadth is also its audit burden — a selection-from-2600 workflow is only as good as the preview step; teams should treat the apply gate as a code review, not a formality.
• The npx agentic-awesome-skills install pattern means the CLI version is whatever npm serves at run time — teams pinning a catalog generation should record the package version in their own lockfile rather than treating the npx command as stable.

6Related on the map

References

• Repository
• Antigravity skills documentation