Purpose
Agentic Awesome Skills (AAS for short) is a skill library that provides 2,690+ installable SKILL.md instruction playbooks. These skills give AI coding agents reusable instructions for completing specific tasks in a project. AAS Core is a local, agent-first control plane that helps Codex or Claude search the complete local catalog, record the skills the agent selects, and preview an inspectable plan before modifying the target.
AAS explicitly states: Core does not rank or recommend skills.
Main Capabilities
- Local catalog viewing: Search and read the complete local skill catalog. Each skill in the current directory can be searched individually, read, and selected by the agent.
- Agent autonomous selection: Codex or Claude inspects the project and selects the exact skills.
- Tech stack validation: A read-only
compose_stacktool validates the selected IDs and structure in memory; the client oraasCLI can persistaas-stack.jsonand optional selection evidence.aas stack validatevalidation manifest,aas stack planwrites an immutable preview for review. - Plan preview: Review the generated plan before installation. The technical upper limit of the manifest is 128 skills.
- Components: Includes CLI, local MCP, catalog, plugins, and Workbench.
- Workbench: Review the tech stack and plan artifacts in browser memory, without accessing the local file system.
Usage
Core Setup and Trust Boundary
Start with AAS Core in Codex or Claude. Configure the local MCP using the Codex or Claude guide. Once MCP is available, ask the agent to inspect the project, compare relevant skills, and save the exact selections. Then, after validating its manifest and reviewing the generated plan, proceed with any installation. The first configuration command previews the changes and returns an approval digest. Claude requires --host claude and its configuration path.
Directly Install Selected Skills
If the skill ID is already known, you can preview installation to the host skill directory:
aas stack install-preview
After reviewing the preview, remove --dry-run when ready and run it again. The direct installer does not consume or apply Core plans. Antigravity's watched skill directory may overload its context, so its default target requires a selected set, a filter, or an explicit --all override.
Plugins and Packaging
Choose focused plugins for the domain you are working in; these packages are available for Claude Code and Codex, and compatible bundles also have standard Agent Plugins manifests.
Bundles group skills by role or goal (such as Web Wizard, Security Engineer, OSS Maintainer). Workflows provide ordered playbooks for planning, delivery, testing, and auditing. They are only guidance for selecting and running skills, not additional packages to install.
Browsing and Discovery
You can browse the complete library in the hosted catalog or CATALOG.md . The canonical manual is located at skills/,skills_index.json Provides a machine-readable discovery method.
Applicable scenarios
- Projects that want AI coding agents such as Codex and Claude to execute tasks according to an exact selection.
- Need a reviewable, reusable skill set, and preview the plan before installation.
- Need to retrieve, compare, and validate selected skills as a tech stack within a local directory.
Limitations
- The current release version is V19.7.0, supporting local directory viewing, agent autonomous selection, tech stack validation, and plan preview; Apply and recovery are still experimental, require explicitly enabling the experimental option, and are not within the supported preview path.
- Core does not rank or recommend candidates in the supported preview path, nor does it install skills.
- The validity of structure and identity does not equate to semantic fit, compatibility, correct configuration, operational safety, or safe applicability.
- The manifest technical upper limit is 128 skills.
- Workbench reviews tech stacks and plan artifacts in browser memory and does not access the file system.
- This project is an independent community project, not affiliated with or endorsed by Google; Google, Antigravity, Gemini, and related names are used only to describe compatibility and installation targets.
- The materials do not provide specific information about fees or pricing (except that individual skill sources are labeled as paid hosting services).
Other notes
- New skills are added under
skills/<skill-name>/SKILL.mdand followCONTRIBUTING.md. Start from the skill template, and before submitting a PR runnpm run validate. - Source code PRs should not include generated registry artifacts. Skill content and high-risk guidance require manual logical and security review beyond automated checks.
- Official source: the GitHub repository sickn33/agentic-awesome-skills is the canonical source; the hosted directory and the browser-local Workbench are the accompanying discovery and review interfaces.