agentic-awesome-skills

AAS Core (Agentic Awesome Skills) is an open-source repository that provides a local, agent-first control plane for working with reusable agent skill instructions. Its stated goal is to let a coding agent inspect a project, search a complete local catalog, choose exact skill IDs, and turn that choice into reproducible, reviewable state rather than an opaque one-shot install. The project describes a workflow in which Codex or Claude inspects the project, searches and reads the complete local catalog through a read-only local MCP (stdio), selects exact skill IDs, and passes them to a read-only compose_stack tool that validates the selection in memory. A client or the aas CLI can then persist the selection as aas-stack.json, optionally alongside a separate selection-evidence sidecar, and produce an immutable plan before any target change. The CLI offers stack validate and stack plan commands; Workbench is a browser-local surface for importing and reviewing stack and plan JSON without filesystem access or installation. The MCP exposes tools such as search_skills, get_skill, list_skill_files, read_skill_file, compose_stack, inspect_stack, diff_stack, plus read-only export_selection_evidence and inspect_selection_evidence. The README states that Core does not rank, recommend, exclude or hide skills, and that structural and identity validity does not certify semantic fit, compatibility, setup correctness or operational safety. Apply and recovery are described as experimental and outside the supported preview path. The repository also retains distribution surfaces: direct installers via npx, specialized plugins for domains such as web app building, product design, security, documents, data analytics, agent/MCP building, QA, DevOps, accessibility, API platforms, SaaS launch and AI evaluation ops, plus bundles and workflows. The catalog is described as containing 2,115+ skills spanning development, testing, security, infrastructure, product and marketing. Supported hosts mentioned include Claude Code, Cursor, Codex CLI, Autohand Code, Gemini CLI, Antigravity, Kiro, OpenCode, Copilot and others. The README notes an audit command that reads selected skill directories without executing them and reports command, network, credential, filesystem, privileged, destructive, symlink and binary signals, described as a review aid rather than a safety certificate. It also warns that a full Antigravity install may exhaust context or cause crash loops, so the installer requires explicit selectors or an --all override for that target. The project is MIT-licensed and states it is an independent community project not affiliated with Google.