Best Claude Code Skills in 2026: Seven Skill Repos Worth Installing

The best Claude Code skills right now come from seven public repositories: mattpocock/skills and addyosmani/agent-skills for engineering workflow, caveman for shorter and cheaper sessions, archify for diagrams, Cloudflare's security-audit-skill for security reviews, last30days-skill for research, and heydamir/unslop for cleaning up AI-looking UI. All seven are open source, and each installs with one or two commands.

A skill is a folder with a SKILL.md file: YAML frontmatter with a description, then Markdown instructions. Claude Code loads skills when a task matches, or when you type /skill-name (Claude Code skills docs). The format is the open Agent Skills standard, so most of the skills below also run in Codex and other clients.

The short list, with stars as shown on GitHub on 7 November 2026:

How we picked

We started from the skill repositories that repowise has indexed, then kept the ones that are maintained, documented and installable today. For each one we read the README and checked the repository on GitHub on 7 November 2026. We did not benchmark the skills against each other. They solve different problems, so the useful question is which one fits your work.

Every entry links to the GitHub repository and to its page on repowise, where you can read the structure of the code before you install it.

mattpocock/skills: engineering habits as skills

mattpocock/skills is a set of small, composable skills taken from Matt Pocock's own .agents directory, aimed at real engineering instead of one-shot generation. The README describes them as easy to adapt and usable with any model. MIT licensed.

• For: developers who want a working engineering loop (issue tracker, planning, review) they can edit.
• Install in Claude Code: claude plugin install mattpocock-skills@claude-plugins-official. For other agents: npx skills@latest add mattpocock/skills -a <agent>.
• Then: run /setup-matt-pocock-skills once per repository to pick your issue tracker.
• Links: GitHub · repowise page

Pick it if you want a small, hackable set of engineering skills instead of a framework that owns your process.

juliusbrussee/caveman: fewer tokens per session

caveman makes the agent answer in fewer words and shrinks logs, JSON and test output before the model reads them. The README reports 33.2% fewer input tokens across 54 Claude Code runs and credits the approach to its own measurements, so treat the numbers as the author's. It works with Claude Code, Codex, Gemini CLI, OpenCode and more than 30 other agents. Apache 2.0.

• For: anyone watching token spend or hitting context limits.
• Install: npx @caveman-ai/cli@latest (needs Node.js 22.13 or newer), or only the skill with npx skills add JuliusBrussee/caveman -g.
• Then: type /caveman in Claude Code, or $caveman in Codex.
• Links: GitHub · repowise page

Pick it if long sessions are your main cost and you are happy with terse answers.

addyosmani/agent-skills: a full development lifecycle

addyosmani/agent-skills packages 25 production-minded skills and nine slash commands that follow a project from spec to ship. Commands map to the lifecycle (define, plan, build, verify, review, ship), and skills activate by themselves when the task matches, for example API design or test-driven development. MIT licensed.

• For: teams that want a consistent definition of done across agents.
• Install: npx skills add addyosmani/agent-skills for all skills, or --skill code-review-and-quality for one. In Claude Code: /plugin marketplace add addyosmani/agent-skills, then /plugin install agent-skills@addy-agent-skills.
• Links: GitHub · repowise page

Pick it if you want opinionated review, testing and shipping gates in one install.

tt-a1i/archify: diagrams from an idea or a repo

archify turns a description, a plan or a codebase into an interactive diagram that opens as a self-contained HTML file. It runs in Claude Code, Codex, Cursor and OpenCode, and can answer a question as a one-page explainer with diagrams, tables and cited sources. Apache 2.0.

• For: engineers who explain systems to other people.
• Install: send your agent Install Archify for me by reading https://raw.githubusercontent.com/tt-a1i/archify/main/llms.txt. Other routes are listed in the README.
• Use: Use Archify to diagram a web request: Browser calls the API, the API checks Redis, and a cache miss queries PostgreSQL.
• Links: GitHub · repowise page

Pick it if you need a shareable diagram without opening a drawing tool. For a diagram generated from code structure, see generating architecture diagrams from a GitHub repo.

mvanhorn/last30days-skill: research across social and the web

last30days researches a topic across Reddit, X, YouTube, Hacker News, Polymarket and the web, scores results by engagement and writes a grounded summary. Reddit, HN, Polymarket and GitHub work with no setup, and a first-run wizard adds more sources. MIT licensed.

• For: founders and developers researching a tool, a competitor or a trend before deciding.
• Install in Claude Code: /plugin marketplace add mvanhorn/last30days-skill, then /plugin install last30days. For Codex, Cursor and other hosts: npx skills add mvanhorn/last30days-skill -g.
• Use: /last30days <topic>.
• Links: GitHub · repowise page

Pick it if you want current community opinion that a web search alone does not surface.

cloudflare/security-audit-skill: multi-phase security audits

cloudflare/security-audit-skill runs a multi-phase security audit of a codebase and reports machine-readable findings that are verified independently. It is published by Cloudflare. MIT licensed.

• For: teams that want a structured security pass before a release or a review.
• Install: npx skills add https://github.com/cloudflare/security-audit-skill --skill security-audit, with --global for a user-level install.
• Use: start the agent in the repository and ask security audit this codebase.
• Links: GitHub · repowise page

Pick it if you want an audit workflow whose findings you can check, not a one-line "is this secure" answer. Treat the output as input to a human review.

heydamir/unslop: fix AI-looking UI

unslop finds the tells of generated UI (purple gradients, glows, emoji labels, shouty copy, uneven spacing), sets a design direction from your product and fixes the screen pass by pass. It works in Next.js, React, Vue, Svelte, Tailwind or plain HTML and CSS, and can start from a live URL. CC-BY-4.0.

• For: anyone shipping a dashboard, app or landing page that was built quickly with an agent.
• Install for Claude Code:

git clone https://github.com/heydamir/unslop
mkdir -p ~/.claude/skills && cp -r unslop/skills/unslop ~/.claude/skills/

• Install for Codex: $skill-installer install https://github.com/heydamir/unslop/tree/main/skills/unslop.
• Use: /unslop this dashboard in Claude Code, $unslop in Codex.
• Links: GitHub · repowise page

Pick it if your UI works but looks like everyone else's.

Claude skills, agent skills and Codex skills: how they relate

Claude skills and Codex skills are the same format, called Agent Skills: a folder with a SKILL.md, loaded on demand. Agent Skills was developed by Anthropic and released as an open standard (agentskills.io). Only the install location and the way you invoke a skill differ per tool.

Sources: Claude Code skills and Codex skills, checked 7 November 2026. The npx skills add command from the open skills CLI installs one skill into many agents at once, which is why most repositories above document it first.

How to choose

Install two or three. Every installed skill adds its name and description to the agent's context, and overlapping skills can trigger on the same request, so a small set you know beats a large set you do not.

Before you install a skill

• Read the SKILL.md and any scripts it bundles. A skill is instructions plus code your agent can run.
• Check the license, the last commit and the open issues.
• Prefer project scope for a skill only one repository needs, so your team gets it through the repository.

Skills and codebase context

Skills tell an agent how to work: a review checklist, a diagram style, an audit procedure. They do not tell it what your own repository contains. That is the job of a codebase MCP server such as repowise, which answers questions about structure, history, ownership and risk from a precomputed index. The two combine well: a review skill with a risk lookup behind it. See giving AI coding agents real codebase context and the roundup of the best MCP servers.