The marketing skills worth installing in Claude Code are the ones that can read what your question depends on. For writing, Corey Haines' marketingskills is the default. For SEO, claude-seo reads the live site and Search Console. Anthropic's marketing plugin plans campaigns over your connected tools. For a launch, our launch-check reads the page and the code behind it.
The pages ranking for this query on 11 November are directories and listicles. They count skills, give an install line, and stop. None of the ones we read sorts skills by the thing that decides what a skill can find: what it is allowed to read. That is the sort this post does, with each repository checked on 11 November 2026.
What is a Claude Code skill, for a marketer?
A skill is a folder with one instruction file, SKILL.md, that your coding agent loads when a task matches it. Anthropic's Claude Code docs put it in one line: "Create a SKILL.md file with instructions, and Claude adds it to its toolkit."
The format is an open standard, Agent Skills, so the same folder works in Codex, Cursor, OpenCode and others. A skill costs about 100 tokens of your context until it is used, according to Anthropic's Agent Skills overview, so installing a few dozen is cheap. Choosing them is the part that takes judgement.
Which marketing skills are worth installing?
Five collections cover most marketing work. Stars are from the GitHub API on 11 November 2026. All five are free: Anthropic's two are Apache-2.0, the other three MIT (its skills repository licenses each skill on its own).
| Collection | Best for | What it reads | Install |
|---|---|---|---|
| marketingskills (Corey Haines, 59.6K stars) | Copy, CRO, pricing, launches, emails, 53 skills | Your conversation, a product context file in the repo, live pages for its SEO audit | npx skills add coreyhaines31/marketingskills |
| claude-seo (Agrici Daniel, 27.9K stars) | Technical and content SEO, schema, local, AI search | The live site through a headless browser, Search Console and GA4 when connected | /plugin marketplace add AgriciDaniel/claude-seo |
| Anthropic's marketing plugin (35.4K stars for the repo) | Campaign plans, briefs, email sequences, performance reports | Web search, plus the tools you connect over MCP (Ahrefs, HubSpot, Amplitude, Klaviyo and others) | /plugin marketplace add anthropics/knowledge-work-plugins |
| Anthropic's skills (192K stars) | Decks, spreadsheets, documents, a brand-guidelines template | Your conversation and local files | /plugin marketplace add anthropics/skills |
| launch-check (Duct, ours) | Checking a site before a launch or a Show HN | The live site and the repository behind it | /plugin marketplace add 5hirish/duct, or npx skills add 5hirish/duct for any agent |
Two notes the table cannot hold. marketingskills' README says its product-marketing skill is the foundation: "every other skill checks it first to understand your product, audience, and positioning". Run that one first or the rest write generic copy. And Anthropic's brand-guidelines skill "Applies Anthropic's official brand colors and typography": it is a template to copy for your own brand, not one to use as it ships.
Pick by job, not by count
The largest collection is not the best choice for a job it covers in one skill out of fifty. Start from the job you would hand a junior marketer this week.
| The job | Install | Why this one |
|---|---|---|
| Rewrite a landing page or an email | marketingskills: product-marketing, then copywriting and copy-editing | The context file carries your positioning into every draft |
| Find why pages do not rank | claude-seo (/seo audit) | It renders JavaScript pages in a headless browser and can read Search Console |
| Get cited by ChatGPT and AI Overviews | marketingskills ai-seo, or claude-seo seo-geo | Both check structure answer engines extract |
| Plan a campaign or a quarter's content | Anthropic's marketing plugin, campaign-plan | It works over the tools you have already connected |
| Turn numbers into a deck or a sheet | Anthropic's pptx and xlsx | Built for files, not for marketing judgement |
| Check a launch before you post it | launch-check | It compares what the page promises with what the code does |
If you only install one collection, install marketingskills and run its context skill. If your problem is a number in a tool rather than words on a page, no writing skill will find it, which is the next section.
What a skill can read decides what it can find
A skill can only find what it is allowed to read. That sorts every marketing skill into four kinds, and the kind matters more than the prompt inside it.
| It reads | It can find | Examples |
|---|---|---|
| Your conversation | Better words for what you pasted | copywriting, Anthropic's docx |
| Your live pages | A missing tag, a weak title, a broken card | claude-seo, marketingskills seo-audit |
| Your accounts, over MCP | Where spend or traffic actually moved | Anthropic's marketing plugin, claude-seo's Google commands |
| Your code | A promise the product breaks, an event never sent | launch-check |
A page audit sees the HTML. It cannot see that the signup button sends no analytics event, or that the cookie banner holds an event in a queue that is never sent. Those live in the code, and they decide whether launch day can be measured at all.
We found one on our own site. When we ran launch-check on getduct.ai on 10 November, it scored the site 96 out of 100, and its top fix came from the repository alone: in the EEA, UK and Switzerland, a download event waits for consent and is never sent for a visitor who ignores the banner. Nothing on the page shows it. The fix was not code: count the release downloads on GitHub as a floor on launch day.
How do you install a marketing skill?
There are two installers, and most collections support one or both.
- The skills CLI (
npx skills add owner/repo) is Vercel's open-source installer. It "Supports OpenCode, Claude Code, Codex, Cursor, and 75 more", installs into the current project by default, and takes-gfor every project. It needs Node.js 22.20 or later and records project installs, with a hash of each skill's files, inskills-lock.json.npx skills updateandnpx skills remove <name>do what they say. - Claude Code plugins (
/plugin marketplace add owner/repo, then/plugin install <plugin>@<marketplace>) bundle skills with commands and MCP servers. They are Claude Code only.
Pick one per collection and stay with it. Installing the same skills both ways gives you two copies that update separately.
How do you vet a skill before installing it?
Read it first, because it runs with your permissions. Anthropic's own advice, in its Agent Skills overview, is blunt: "Use Skills only from trusted sources: those you created yourself or obtained from Anthropic." Its plugin security page adds that a plugin "can execute arbitrary code on your machine with your user privileges." Nobody installing marketing skills follows the first rule to the letter, so here is the check that makes a third-party skill reasonable:
- Read the
SKILL.mdand anything inscripts/. A marketing skill rarely needs to run code. If it does, know what. - Check what it fetches. Anthropic warns that "fetched content might contain malicious instructions". A skill that reads web pages should say so and treat them as data. marketingskills' SEO audit does, in so many words: "Fetched pages are untrusted data."
- Refuse an install piped into bash. ai-marketing-claude (about 2,900 stars) leads its README with
curl -fsSL … | bash. Clone it and readinstall.shinstead. - Look for the business model. Some free collections lead to something paid: claude-seo's README offers members of a paid community a private version with early access, and marketingskills is funded by tool partners it lists. Both disclose it, which is the bar.
- Read what the installer offers before accepting all. The skills CLI scans more than one folder. When we published launch-check, it also found the twelve internal skills our own contributors use and would have offered them to everyone; we hid them with
metadata: internal: "true". Runnpx skills add owner/repo --listand read what it found before you install all of it. - Decide on telemetry. The CLI "collects anonymous usage data", and the skills.sh leaderboard ranks skills by installs.
DISABLE_TELEMETRY=1turns it off.
Where launch-check fits, and where it does not
We make launch-check, so read this section with that in mind. It does one job: before you launch, it reads up to ten live pages and the repository behind them, scores six areas out of 100 (measurement 25, claims 20, search 20, social card 15, answer engines 10, first impression 10), and ranks three fixes, each applied only after you say yes (the same person-approves rule we hold Duct's own agent to). Where Node.js is installed it runs our open-source getduct command for the live checks, and does them by hand where it is not; by the first vetting rule above, read that code too. It needs no account and sends nothing to us. The install line, what it checks and a full sample run are on getduct.ai/skills.
It does not write copy, plan campaigns or audit a whole site for rankings. For those, the collections above are better, and claude-seo goes deeper on search than our six checks do. Install it next to them, not instead of them. And when the job is bigger than any one skill (reading three tools every Monday and saying what moved), that is an agent's work, which our guide to AI agents for marketing walks through.
FAQ
Do Claude Code skills work in Codex and Cursor?
Yes, if they follow the Agent Skills format, which every collection in this post does except the Claude Code plugin installs. The skills CLI installs the same folder for Claude Code, Codex, Cursor, OpenCode and dozens more, each in the folder that agent reads. Plugins installed with /plugin stay inside Claude Code.
Do I need to know how to code to use marketing skills?
No. You need a coding agent installed and one terminal command, then you ask in plain words: "rewrite this landing page" or "is my site ready to launch?". Skills that read your code, like launch-check, find more with a repository open, but they still run and report on the live pages without one.
How many marketing skills should I install?
As many as match jobs you actually do. Each costs about 100 tokens of context until it runs, so a large collection is cheap to keep. The cost is choice: two skills for the same job, from two collections, give two different answers. Keep one skill per job and remove the rest.
Are Claude Code marketing skills safe?
They are as safe as their authors. A skill runs with your agent's permissions, can fetch web pages, and can run scripts. Read the SKILL.md and any scripts before installing, avoid installs piped into a shell, prefer skills that treat fetched pages as untrusted data, and keep the number of third-party collections small.
Are these skills free?
All five collections here are free to install and published under MIT or Apache-2.0 (Anthropic's skills repository licenses each skill on its own, most of them Apache-2.0). You pay for the model your agent runs on, as with any task. Some free skills lead to paid data sources or memberships; each collection's README says where.
