sickn33/agentic-awesome-skills

What it is

Agentic Awesome Skills collects 2,884+ installable SKILL.md playbooks aimed at Claude Code, Codex CLI, Cursor, Gemini CLI and other AI coding agents. Each skill is a reusable instruction file: a workflow, checklist or domain playbook the agent can load and follow. The repository also ships AAS Core, a local MCP interface that lets Codex or Claude search the full catalog, pick exact skill IDs, and write a reviewable selection manifest before anything is installed. Core deliberately does not rank or recommend skills.

It is for developers who already drive an AI coding agent and want to extend it on specific tasks — brainstorming, debugging, security review, test audits — without writing every instruction themselves. A hosted catalog and browser-local Workbench provide discovery and review surfaces; the GitHub repo is the canonical source.

Quick start

Requirements: Node.js and npm, plus a supported agent host. The pinned release used below is 20.3.0.

1. Configure the local MCP for Codex. This command previews the change and returns an approval digest rather than silently editing the config:

npm exec --yes --ignore-scripts --package=agentic-awesome-skills@20.3.0 -- aas mcp configure \
  --host codex \
  --scope user \
  --config /absolute/path/to/codex/config.toml \
  --cache-root /absolute/path/to/aas-cache

For Claude, use --host claude with its configuration path. Approval, reconnection and validation steps are in the AAS Core guide.

2. With the MCP available, ask the agent to inspect your project, compare relevant skills, and save its selection. Validate the manifest with aas stack validate, then run aas stack plan to write an immutable preview.

3. If you already know the skill IDs, preview a direct install instead:

npm exec --yes --ignore-scripts --package=agentic-awesome-skills@20.3.0 -- \
  agentic-awesome-skills --release 20.3.0 --path .agents/skills \
  --skills brainstorming,systematic-debugging --dry-run

Review the output, then repeat without --dry-run. Per-host install commands for Cursor, Gemini CLI, Antigravity, Kiro, Copilot and others are listed in Choose Your Tool.

Constraints and limits

• Core's supported path covers discovery, selection, validation and plan preview. Apply and recovery are experimental, require opt-in, and sit outside the supported preview; aas stack install-preview only prepares a --dry-run and does not apply a Core plan.
• Validation checks structure and identity, not semantic fit, compatibility, setup correctness or operational safety.
• A stack manifest has a technical maximum of 128 skills.
• Antigravity watches a skill directory that can overload its context, so its default target requires a selected set or filter; other hosts may need --risk, --category or --tags to narrow installs.
• Third-party skills come from many upstream projects with their own licenses and hosted services. Original code and tooling are MIT; documentation is CC BY 4.0 unless an upstream notice says otherwise, and per-source terms are tracked in the attribution ledger.

HotRepo's take

The repository appeared on HotRepo's daily all-language list on 2026-10-20 and 2026-10-21 with 58 and 83 new stars respectively. On 2026-11-01 it ranked 7th on the daily Python list with 74 stars gained that day. That ranking is the only recorded position for this project, and only three snapshots are available, so they show activity on those dates rather than a trend.

What stands out technically is the split between the catalog and the installer: the agent proposes a skill set in memory, validation and plan generation happen locally, and installation is a separate, preview-first step. That is a more cautious design than a one-shot install script, and the explicit statement that Core does not rank skills means the judgment call stays with the user.