Guide · Open source

How to use agentic-awesome-skills: install, configure and preview

Updated 2026-10-29 · written with AI from the project's own documentation and checked against our catalogue data

Agentic Awesome Skills is a library of 2,877+ installable SKILL.md playbooks, which are instruction files that AI coding agents such as Codex or Claude can read and follow. This guide explains how to set up the AAS Core local tool, let your agent pick and save a skill selection, and preview an install before anything changes. It is written for developers who are comfortable with a terminal and config files.

Agentic Awesome Skillssickn33/agentic-awesome-skills · ★ 49k · MIT Install options & downloads

Quick start

1Check prerequisites. Confirm Node.js is installed and that you use Codex or Claude.
2Configure the MCP. Run the aas mcp configure command with your host, config path and cache path, and review the returned approval digest.
3Select skills with the agent. Ask the agent to inspect your project and save the exact skill selection it chooses.
4Validate the manifest. Run aas stack validate to check the saved manifest structure.
5Write a plan preview. Run aas stack plan to create an immutable preview you can review.
6Preview a direct install. Run the agentic-awesome-skills command with --dry-run to see what would be installed.

What you need before you start

You need Node.js, because the install commands run through npm exec. The README excerpt does not state a minimum Node.js version, so check the project documentation for that. You also need a client that the project supports: Codex or Claude.

Node.js installed (npm exec is used)
Codex or Claude client
Familiarity with editing config files and running terminal commands

Configure the local MCP for Codex

The first configuration command previews a change and returns an approval digest. Replace the two absolute paths with locations on your machine. Use --host claude and a Claude configuration path if you use Claude instead.

The setup guide for AAS Core explains approval, reconnection, validation and planning in more detail.

Replace /absolute/path/to/codex/config.toml with your Codex config file path
Replace /absolute/path/to/aas-cache with a cache directory you choose

npm exec --yes --ignore-scripts --package=agentic-awesome-skills@20.2.0 -- aas mcp configure \
  --host codex \
  --scope user \
  --config /absolute/path/to/codex/config.toml \
  --cache-root /absolute/path/to/aas-cache

Let the agent select and save a skill set

With the MCP available, ask the agent to inspect your project, compare relevant skills and save the exact selection. Core searches the full local catalog and records the chosen skills. It does not rank or recommend skills itself, so the choice comes from the agent and you.

Then validate the manifest with aas stack validate, which checks the manifest structure. Run aas stack plan to write an immutable preview you can review before any change. Manifests have a technical maximum of 128 skills.

aas stack validate checks the manifest
aas stack plan writes a preview for review
The read-only compose_stack tool validates chosen IDs in memory

Preview a direct install of selected skills

If you already know the skill IDs, you can skip Core and preview a focused install into your host's skill directory. The --dry-run flag shows the result without writing files. Review the preview, then run the same command without --dry-run when you are ready.

The direct installer does not consume or apply a Core plan. Antigravity's skill directory has a default that requires a selected set, a filter, or an explicit --all override, because a large directory can overload its context.

--release sets the release tag
--path sets the target skill directory
--skills takes a comma-separated list of IDs
--dry-run previews without applying

npm exec --yes --ignore-scripts --package=agentic-awesome-skills@20.2.0 -- \
  agentic-awesome-skills --release 20.2.0 --path .agents/skills \
  --skills brainstorming,systematic-debugging --dry-run

Know the limits of the preview

Structural and identity validity does not certify semantic fit, compatibility, setup correctness, operational safety, or safety to apply. The README says apply and recovery remain experimental and sit outside the supported preview.

The Workbench on the hosted site reviews stack and plan artifacts in browser memory without accessing your filesystem.

Apply and recovery are experimental
Validation does not prove a skill is a good fit
Core does not rank or recommend skills

What people use it for

A developer using Codex asks the agent to pick skills for a specific project and saves the list for review.
A team reviews an exact, saved skill list before installing anything into a shared skill directory.
A developer previews an install of brainstorming and systematic-debugging skills into .agents/skills.

Frequently asked questions

Is agentic-awesome-skills free and what is its license?

The project is licensed under MIT, according to the README badge. The README does not describe pricing for the tool itself, so check the repository for any paid services.

What do I need to install agentic-awesome-skills?

The install commands run through npm exec, so you need Node.js. The README excerpt does not state a required Node.js version. You also need Codex or Claude as your client.

Does agentic-awesome-skills recommend which skills I should use?

No. The README says Core does not rank or recommend skills. The agent chooses skills from the full local catalog, and you review the saved list and plan.

Can I use agentic-awesome-skills without a terminal?

Not easily. Setup involves MCP configuration and terminal commands, so the project is aimed at users comfortable editing config files and running CLI commands.

How is agentic-awesome-skills different from a hosted skill service?

The project describes Core as a local, inspectable workflow. The GitHub repository is the canonical source, and the hosted catalog and Workbench are companion discovery and review surfaces.

Why does my install or apply step not work in production?

The README says apply and recovery features remain experimental and outside the supported preview. Use the --dry-run preview and the supported validation and plan commands instead.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Agentic Awesome Skills suits developers who use Codex or Claude and want agent-chosen instructions that they can inspect, save and preview before installing. It is a poor fit for non-technical users, for anyone who wants the tool to recommend skills, or for teams that need apply and recovery features in production.

Install options & downloads