Awesome Claude Skills: The Best Agent Skills by Category

Awesome Claude Skills is a curated directory of the best Agent Skills for Claude Code, grouped by the job they do. Agent Skills are folders of instructions, scripts, and resources that Claude loads on demand, and the strongest maintained collections in July 2026 are Anthropic's own anthropics/skills (Apache 2.0), Jesse Vincent's obra/superpowers-skills (MIT), and OpenAI's Codex-focused openai/skills catalog. Below is a best pick per category with a one-line description, an install path, and a source link for each, followed by the fastest way to browse them all.

How Claude Skills Install

A Claude skill is a folder containing a SKILL.md file with instructions the agent loads when a task matches, so installing one usually means placing that folder where Claude Code looks. There are three common paths, and they follow the open Agent Skills standard Anthropic published in 2025.

# 1. Install a whole collection from its marketplace
claude
/plugin marketplace add anthropics/skills
/plugin install document-skills@anthropic-agent-skills

# 2. Or drop a single skill folder into your project
mkdir -p .claude/skills
# copy the skill's folder (with SKILL.md) into .claude/skills/

Because the format is standardized, a skill written for one agent generally works in another. That portability is why the collections below, even the Codex-focused ones, are worth mining for Claude Code.

The Best Claude Skills by Category

These are the strongest maintained skills per category as of July 2026, with the lead pick listed first in each group. Skill names are the exact folder names in their source repositories.

Code review

Best pick: requesting-code-review from superpowers, which drives an agent to request and act on structured review before finishing a branch. Pair it with receiving-code-review for the reverse flow. From OpenAI's catalog, gh-address-comments resolves review comments on a GitHub pull request automatically. Source: obra/superpowers-skills. See our best code review skills roundup for more.

Testing

Best pick: test-driven-development from superpowers, which enforces a red-green-refactor loop instead of letting the agent bolt tests on afterward. Companions include testing-anti-patterns and condition-based-waiting (superpowers), Anthropic's webapp-testing for browser-driven checks, and OpenAI's playwright and playwright-interactive skills. Deep dive: best testing skills for AI agents.

Debugging

Best pick: systematic-debugging from superpowers, which forces a hypothesis-driven loop instead of random edits. Strong companions are root-cause-tracing, verification-before-completion, and defense-in-depth, all in the same repo. Source: obra/superpowers-skills.

Documentation

Best pick: doc-coauthoring from anthropics/skills, which helps Claude draft and revise documents interactively. The same repo ships the document workhorses docx, pdf, pptx, and xlsx for creating and editing Office and PDF files. Source: anthropics/skills.

Security

Best pick: security-threat-model from OpenAI's catalog, which walks an agent through threat modeling a codebase. Companions are security-best-practices and security-ownership-map. For breadth, the community Anthropic-Cybersecurity-Skills project packages hundreds of cybersecurity skills mapped to frameworks like MITRE ATT&CK and NIST CSF. See our best security skills guide.

Frontend and design

Best pick: frontend-design from anthropics/skills, which pushes Claude toward production-grade, distinctive UI instead of generic output. The same repo adds web-artifacts-builder, canvas-design, theme-factory, brand-guidelines, and algorithmic-art. OpenAI's figma-implement-design turns Figma into code. Source: anthropics/skills.

Planning and collaboration

Best pick: brainstorming from superpowers, which refines a rough idea through questions before any code is written. It pairs with writing-plans, using-git-worktrees, dispatching-parallel-agents, and subagent-driven-development. These lean on Claude Code's multi-agent features covered in our agent teams guide.

Deployment and DevOps

Best pick: vercel-deploy from OpenAI's catalog, which ships a project to Vercel end to end. Siblings cover other targets: netlify-deploy, cloudflare-deploy, and render-deploy, plus gh-fix-ci for repairing failing GitHub Actions runs. Source: openai/skills.

Data and notebooks

Best pick: xlsx from anthropics/skills, which reads, builds, and edits spreadsheets with formulas intact. From OpenAI's catalog, jupyter-notebook runs and edits notebooks, while transcribe and speech handle audio in and out. For chart work, Claude Code now ships a built-in /dataviz skill as of the July 2026 release. Source: anthropics/skills.

Skill building and MCP

Best pick: mcp-builder from anthropics/skills, which scaffolds a new Model Context Protocol server. Pair it with skill-creator for authoring your own skills, and browse ready-to-use servers in our MCP directory. OpenAI's cli-creator covers command-line tooling, and migrate-to-codex ports an existing setup. Source: anthropics/skills. If you are new to authoring, start with our first skill walkthrough.

Where These Skills Come From

The four collections worth tracking differ in maintainer, license, and install path, summarized below. Star counts and licenses were checked against each repository in July 2026.

One caveat: the openai/skills repo is marked deprecated in favor of OpenAI's Plugins repository, and it targets Codex rather than Claude Code. Because skills follow the open standard, you can still adapt its SKILL.md folders for Claude by copying them into .claude/skills/, but expect some Codex-specific assumptions.

Browse and Submit Skills

The fastest way to work through these collections is a searchable directory rather than nine separate GitHub repos. The Remote OpenClaw skills directory indexes skills across sources with filters for category, install command, and license, and it links out to each skill's original repository. For prebuilt agent tools that pair with these skills, see the tools hub.

Built something useful? You can list your own skill for other operators at /submit. For a wider tour of where skills live, our best places to find Claude Code skills guide maps the ecosystem.

Limitations and Tradeoffs

A skill is just instructions the agent reads, so a poorly written one can waste context or push Claude toward the wrong pattern, and a malicious one can ship harmful steps. Install skills only from sources you trust, read the SKILL.md before enabling it, and use per-skill disallowed-tools where the format supports it. More skills also means more context overhead, so do not enable dozens at once.

This list favors actively maintained repositories, but maintenance can lapse. The openai/skills catalog is already deprecated, and any awesome-list snapshot lags the source repos. Treat category picks as a starting point and verify each skill against its current README before relying on it in production.

Related Guides

• Best Places to Find Claude Code Skills in 2026
• 15 Best OpenClaw Skills You Should Install Right Now
• Best Code Review Skills for AI Agents
• Best Testing Skills for AI Agents