Marketing skills for Claude Code
For high-performing marketers going AI-native (or trying to). Type a slash command, or describe the problem in your own words. The matching skill asks a couple of questions, reads your files only when you say yes, and does the work.
Who it's for
I built these for my own marketing work and I use them every week. They help you do the work faster and better, but you still do the work (that's kind of the point).
Marketers who want better work, faster
You know your audience and your product, and there's more on your plate than hours in the week. The skills take the slow parts like research and reporting, so your time goes to the thinking and the writing.
The curious, who want to use AI the smart way
You've seen what happens when AI writes everything (just open LinkedIn) and you want the other version, where AI is a tool and a helper and the thinking stays yours. Every skill is a plain text file, so you can open it and see how it works.
Not for you if you want AI to churn out your content or replace your marketers. That's not the way, and I wrote about why in Why is marketing so damn hard right now? and I love AI, I hate AI.
The shelf
One folder per skill, each with its own SKILL.md. The folder name is the slash command. Every skill is self-contained — take the whole library or copy one folder.
It's a live list, curated by me. Every few weeks I add new skills and rework or drop old ones. Last change: .
In action
Real output only. Everything below came out of a real run — nothing here is a mockup of what the library might do.
The hero at the top of this repository's README, before and after. Light mode moves only what is already in the image — the header's dot grid drifts one cell per loop, a highlight walks down each column's command list, the cursor blinks, the constellation breathes, current flows along the connectors. Nothing new was drawn. Drag the divider.
Shipped as an animated WebP with a static PNG fallback, or a GIF when it has to live in a README. The renderer checks the loop seam, the moving region and the file weight before it hands anything back. This one: 40 frames, a 3.2-second loop, 0.30 MB as a GIF; the wrap step measured 0.021 against a typical step of 0.022, and frame 0 differs from the original still by a 3×10 px patch.
Left is what a language model writes about this library when nobody stops it. Right is the same paragraph after the skill checked it against the live Wikipedia list of AI-writing tells. Highlighted words are the ones it flagged.
BEFORE · 89 WORDS
In today's fast-paced digital landscape, marketing-skills empowers marketers to unlock the full potential of their workflows. This comprehensive suite of powerful tools seamlessly integrates with Claude Code, enabling teams to effortlessly streamline everything from SEO audits to content strategy. Whether you're a solo marketer or part of a growing team, our robust library is designed to help you elevate your marketing game and drive meaningful results. It's not just a collection of skills — it's a game-changer.
AFTER · 62 WORDS
You type the problem the way you'd say it out loud: "why isn't this page ranking?" The matching skill asks a couple of questions, reads your files only when you say yes, and does the work. SEO audits, content strategy, board reports, ad budgets, a draft in your voice with the AI tells scrubbed out. Every skill here runs real marketing work every week.
Fourteen flags in four sentences. Dead-giveaway words (in today's X, empowers, comprehensive, seamlessly, streamline, robust, game-changer), difficulty-minimizers (effortlessly), empty amplifiers (powerful, meaningful results), and one structural tell: not just X, it's Y. The skill refetches the Wikipedia list on every run, so what counts as a tell stays current.
The orchestrator: crawls the site, detects the business type, fans out to specialists in parallel, and returns a health score with the findings ranked. Run on 2026-09-21 against skillcraft.cloud — the author's own marketplace, so the uncomfortable findings below are his to fix. Excerpt; the full report is 52 findings.
| Technical SEO | 22% | 58 |
| Content quality | 23% | 48 |
| On-page / search experience | 20% | 30 |
| Structured data | 10% | 55 |
| Performance (lab only, no CrUX key) | 10% | 74 |
| AI search readiness | 10% | 46 |
| Images | 5% | 82 |
- The catalogue is invisible to Google.CRITICAL
/marketplaceships 47 words and zero skill links in HTML; the grid is filled byfetch("/api/skills"), and robots.txt saysDisallow: /api/. Googlebot's renderer honours that rule, so the one page whose job is to list the skills lists none of them. - The skill pages are orphans.CRITICAL No page on the site links to any
/skill/URL in raw HTML; the sitemap is the only path, and it lists 13 of the 15 skills the API serves. Sixteen months of Search Console show two URLs that have ever had an impression:/and/whats-new. - URLs are UUIDs and titles never say "Claude Code".CRITICAL Every competing listing seen for "seo audit skill claude code" uses a keyword slug and the phrase verbatim. All five queries with impressions in 90 days are the brand name or a misspelling of it.
- The homepage has no structured data.CRITICAL Every other template carries JSON-LD — SoftwareApplication, BlogPosting, FAQPage, CollectionPage, BreadcrumbList. The front door has none: no WebSite, no Organization, no SearchAction.
- Soft 404s and no security headers.HIGH Unknown URLs return a
307to the homepage. CSP, X-Frame-Options, X-Content-Type-Options, Referrer-Policy and Permissions-Policy are all absent on a site with sign-in and user-submitted content.
Fired: technical · content · schema · sitemap · performance · visual · GEO · search-experience · backlinks (Common Crawl tier), plus live Search Console and GA4 through the workspace connectors. 52 findings: 7 critical, 11 high, 15 medium, 11 low, 8 info. Not run: PageSpeed and CrUX field data (no API key), paid backlink indexes. Where a check could not be made, the report says so instead of guessing.
How a skill runs
Five stages, the same every time. A number the skill cannot get is written as unknown, never invented.
Company context works the same way in every skill that needs it: scan the launch folder, list what was found, ask before reading a single file.
Take it home
Claude Code discovers skills one folder deep under ~/.claude/skills. This repo nests them by category, so the installer links each skill into place. Your edits land back in the repo, versioned — updating is a git pull.
git clone https://github.com/risukisu/marketing-skills D:\marketing-skills D:\marketing-skills\install.ps1
git clone https://github.com/risukisu/marketing-skills ~/marketing-skills ~/marketing-skills/install.sh
Just one skill
Every skill folder is self-contained. Copy it into ~/.claude/skills and it works on its own — no shared runtime, no registry.
Live data, optional
Connect Google Analytics 4 and Search Console and the data-driven skills stop asking you for numbers and start pulling them. Without them, every skill still runs; it asks instead.