What is Agentic Awesome Skills?
Agentic Awesome Skills is a library of 2,884 SKILL.md playbooks for Claude Code, Codex, Cursor, Gemini CLI, Antigravity, Kiro, Copilot and other agents, maintained by sickn33. It was created in January 2026 as antigravity-awesome-skills; the old name now redirects. Only a minority is original: the machine-readable index marks 451 skills as the contributor's own work, while most are imported from vendor and community repositories such as obra/superpowers, anthropics/skills, BagelHole/DevOps-Security-Agent-Skills and addyosmani/agent-skills.
It ships a direct installer (npx agentic-awesome-skills), 13 focused plugins for Claude Code and Codex, and AAS Core, a local MCP server with nine read-only tools that let Codex or Claude search and read the catalogue, record the skills the agent chooses in an aas-stack.json manifest, and preview an install plan. Core does not rank or recommend skills, and its apply and recovery steps are experimental.
The repository's own code is MIT and its original prose CC BY 4.0; imported skills keep their upstream licences. It has about 48,900 stars and 7,100 forks, and release v20.3.0 came out on 31 October 2026.
Is Agentic Awesome Skills worth it?
Agentic Awesome Skills is mostly an aggregation of other people's skills, and its value depends on treating it that way. Of the 2,884 entries in skills_index.json, 451 are marked as the contributor's own work. Another 1,486 are labelled only "community", and 1,166 of those name no upstream repository. 1,836 entries carry no licence field at all. It is not a blind scrape, since many imports do name their source and licence, but it is not a set you can redistribute without checking each skill.
In our reading of a sample, the gaps sit in the older imports. systematic-debugging is a file-for-file copy of the skill in Jesse Vincent's obra/superpowers, which is MIT, down to a creation log that quotes Vincent's own CLAUDE.md; the folder has no licence notice or author line, the index calls it "community", and the only credit is one line in the README. Anthropic's frontend-design and mcp-builder keep their Apache-2.0 LICENSE.txt but are also tagged "community". Newer imports are cleaner: the 163 skills from BagelHole/DevOps-Security-Agent-Skills and the 82 from elementalsouls/Claude-BugHunter name their repository and MIT licence in the frontmatter, although the folders we read bundle no LICENSE file, which the contributing guide now requires. The attribution ledger records the licence as "Compatible" in 52 rows. To its credit, release 20.0.0 removed the four -official document skills (docx, pdf, pptx, xlsx) after a restrictive-licence review.
Quality control is heavier than the size suggests. Pull requests must state change, validation and source evidence; scripts and executables need a maintainer path; CI runs CodeQL, dependency review and a Tessl semantic review with a pass threshold of 80. NVIDIA's SkillSpector also scans changed skills, but as an advisory job set to continue on error, so a finding does not block a merge. Version 20.0.0 reported a catalogue-wide review of 2,368 bundles, 2,042 of them corrected and 4 removed. Risk labels are declared by authors: 1,544 skills are marked critical, meaning they change state, delete files or push, and 151 offensive.
The install path is the weak point. The README pins agentic-awesome-skills@20.3.0, but npm's latest is 19.15.0 from 22 October; the notes for 20.1.0, 20.2.0 and 20.3.0 all say npm publication failed or was unavailable. Without a selector, the direct installer copies the complete catalogue into Claude, Cursor, Gemini and most other targets; only Antigravity now refuses, because the full set can exhaust its context. AAS Core is the careful route: a local, read-only MCP server over stdio that returns skill text as untrusted content and installs nothing. Its context cost is not published, and its built-in brief asks the agent to cover ten capability areas with a manifest limit of 128 skills, which pulls against a lean setup.
Telemetry is not a concern. The privacy policy says plugins collect none and the MCP server has none; the hosted catalogue uses Vercel Web Analytics with query strings removed. Next to Superpowers or Addy Osmani's agent-skills, where one author writes and maintains a small set, this is a search index over other people's work with a review layer on top, and for a skill you find here the better copy to install is usually the upstream one.
Where it earns its place
A Claude Code user on a solo TypeScript project
Find a skill for a task with no instructions yet, such as Drizzle migration conflicts
Search aaskills.tech or skills_index.json, read the candidate's SKILL.md and bundled files, run npx agentic-awesome-skills audit --skills <id>, then install one or two IDs with --dry-run first. Free; the risk is a copy that has drifted from its upstream.
A platform engineer standardising skills for a 20-person team on Codex and Claude
Give every repository a reviewed, reproducible skill set
Configure AAS Core, let the agent compose an aas-stack.json, validate it and keep the plan in the repository. Pin a version that exists on npm (19.15.0 at the time of writing) or install from the GitHub tag; Core apply is experimental, so installation still goes through the direct installer.
An open-source programme office lead at a company that ships agent tooling to customers
Decide whether skills from this catalogue can be bundled into a product
Treat the repository's MIT licence as covering only its own code. For each skill, follow source_repo and license_source; where they are missing, trace the upstream by hand or drop the skill. Skills taken from Superpowers need Jesse Vincent's MIT notice restored.
A security lead at a 200-developer company
Keep offensive-security and state-changing skills off general developer machines
Allow only installs filtered with --risk safe,none, since 151 skills are labelled offensive and 1,544 critical. The labels are self-declared, so run SkillSpector or snyk-agent-scan on the exact set before rollout.
Before you commit
- Setup
- npx agentic-awesome-skills --claude --skills <ids> --dry-run (the 20.3.0 installer needs Git 2.25+), or aas mcp configure with a preview and an --approve digest to add Core to Codex or Claude. The README's @20.3.0 pin is not yet on npm, whose latest is 19.15.0.
- Integrations
- Direct install targets for Claude Code, Codex, Cursor, Gemini CLI, Antigravity, Kiro, Copilot (through gh skill), OpenCode, AdaL and custom paths; 13 plugins for Claude Code and Codex; AAS Core MCP for Codex and Claude with nine read-only tools.
- Data and privacy
- Plugins and the MCP server collect no telemetry, and the MCP runs locally over stdio; aaskills.tech uses Vercel Web Analytics. Selection evidence files store search queries verbatim. Imported skills keep upstream licences, and 1,836 index entries declare none.
- Cost at scale
- Free. The cost is agent context: installed skills load into the agent, the default install on most hosts is the full catalogue, and the docs warn that the full set can exhaust Antigravity's context.
- Limits
- Releases 20.1.0 to 20.3.0 are on GitHub but not npm; Core apply and recovery are experimental; risk labels are author-declared; SkillSpector is non-blocking; the 20.0.0 review did not exercise external services or paid workflows.
What’s good
- Machine-readable index with source, licence and risk fields
- Read-only local MCP that installs nothing
- Catalogue-wide review in 20.0.0 with published batch findings
- No telemetry in plugins or the MCP server
What’s not
- 1,836 skills declare no licence
- Some copied skills lack the upstream licence notice
- Documented 20.3.0 package is not on npm
- Installer copies the full catalogue by default on most hosts
- SkillSpector scan is advisory only
Best for
Developers and platform teams who want to search a large pool of agent skills, read the candidates and then install a handful, ideally from their upstream source.
Agentic Awesome Skills alternatives
Agentic Awesome Skills: common questions
- Is Agentic Awesome Skills worth it?
- A catalogue worth searching and not worth installing whole: the review process is real, but the licence bookkeeping for its imported skills has not caught up. We rate it 3.4 out of 5.
- How much does Agentic Awesome Skills cost?
- Free and open source (MIT for code, CC BY 4.0 for original prose; imported skills keep upstream licences). You pay for your agent's model usage.
- What are the downsides of Agentic Awesome Skills?
- 1,836 skills declare no licence; some copied skills lack the upstream licence notice; documented 20.3.0 package is not on npm; installer copies the full catalogue by default on most hosts; skillSpector scan is advisory only.
- Who is Agentic Awesome Skills best for?
- Developers and platform teams who want to search a large pool of agent skills, read the candidates and then install a handful, ideally from their upstream source.
- How much does Agentic Awesome Skills cost for a team or in production?
- Free. The cost is agent context: installed skills load into the agent, the default install on most hosts is the full catalogue, and the docs warn that the full set can exhaust Antigravity's context.
- Is Agentic Awesome Skills safe for company data?
- Plugins and the MCP server collect no telemetry, and the MCP runs locally over stdio; aaskills.tech uses Vercel Web Analytics. Selection evidence files store search queries verbatim. Imported skills keep upstream licences, and 1,836 index entries declare none.
- What are the best Agentic Awesome Skills alternatives?
- The closest alternatives we have reviewed are Superpowers, Anthropic Skills, Agent Skills (Addy Osmani) and wshobson/agents.
Answers checked Oct 31, 2026.