AGENT SKILLS
Agent skills for search and ads,read in full and run once each
A collection, not a catalogue. Six open-source libraries hold a few hundred marketing skills for Claude Code, Cursor and Codex; we read the ones that matter for SEO, AEO, paid ads, CRO and content, ran each with a real request, and kept the ones worth installing, with what each needs and where we disagree.
npx skills add coreyhaines31/marketingskills --skill seo-audit- Skills read and run
- 25
- Libraries
- 6
- Run with no key
- 23
- Toolkits
- 6
- Licences
- MIT, Apache
- Last read
- 11 Sept 2026
Browse by channel
Five shelves. Each one is a workflow, not a list: the category page says what the skills do together and what they cannot replace.
- SEO 5Audits, technical checks, product-page scoring, programmatic pages and structured data, from two libraries.
- AEO and GEO 5Five skills for the surface where the answer is written for the reader: AI Overviews, AI Mode, ChatGPT search, Perplexity and Claude.
- Paid ads 5Five skills that cover a campaign from the decision to run it to the decision to scale it, and the two toolkits that let an agent read the account.
- CRO 5Five skills from a fast page review to a research-first programme, with two on experiment statistics because that is where most CRO goes wrong.
- Content 5Five skills across the life of a page: the draft, the edit, the pre-publish gate, the anti-slop ruleset, and the quality score.
- Toolkits 6MCP servers and plugins that connect an agent to accounts you own: GA4, Google Ads, Search Console, a browser, and two operating systems for marketing work.
Sources
Six libraries, none of them ours
Stars and push dates as read on 11 September 2026 from the GitHub API. We selected 25 skills from the 255 in these repositories and left the rest where they are.
coreyhaines31/marketingskills
Marketing skills for AI agents: CRO, copywriting, SEO, analytics, growth engineering.
AgriciDaniel/claude-seo
A Claude Code plugin: 25 SEO sub-skills, 18 sub-agents, 56 Python scripts, optional DataForSEO, Google API and Firecrawl extensions.
rampstackco/claude-skills
Stack-agnostic skills across the website lifecycle: brand, design, content, SEO, dev, ops, growth, research.
wondelai/skills
Frameworks from bestselling business, marketing and UX books as skills, plus guided multi-skill journeys.
inhouseseo/superseo-skills
Eleven SEO skills: page audits, link building, article writing, E-E-A-T audits, with a generation-time anti-AI-slop ruleset.
leopard627/fire-your-seo-agency
One skill that diagnoses and implements SEO, AEO, GEO, LLMO and Naver optimisation; instructions written in Korean, references mirrored in English.
Read and run
All 25 skills, with what each needs
Every item page carries the same eight blocks: the record, the install command, a needs badge, a four-row safety review, the triggers, what is inside, one real run cropped, and where we would differ.
- programmatic-seo SEOPlan pages at scale from templates and data without building a thin-content farm: twelve playbooks, a data-defensibility ladder, and a pre-launch checklist.
- schema SEOAdd, fix or validate JSON-LD structured data: ten common types with their required properties, an @graph pattern, and the two validators to run before deploying.
- seo-audit SEOA full-site SEO audit that crawls, delegates to specialists and returns a 0 to 100 health score.
- seo-ecommerce SEOProduct-page SEO scoring, Product schema validation, and, with a DataForSEO key, Google Shopping and Amazon marketplace intelligence.
- seo-technical SEOA nine-category technical audit: crawlability, indexability, security, URLs, mobile, Core Web Vitals, structured data, JavaScript rendering and IndexNow.
- ai-seo AEOThe longest AEO skill on the shelf: a three-pillar method (structure, authority, presence), an AI visibility audit, machine-readable files for agents, and monitoring that treats one run as an anecdote.
- fire-your-seo-agency AEOFive lanes, SEO, AEO, GEO, LLMO and Naver, diagnosed with curl and fixed in your codebase, by a skill whose instructions are written in Korean.
- seo-aeo-geo AEOA five-layer AEO and GEO framework, extractable structure, citation worthiness, schema depth, AI-readable access, real-world entity signals, with a rule that a stated gap beats an estimated number.
- seo-content-brief AEOCompetitive content briefs with per-section word counts, scored competitors, keyword placement rules, and a non-negotiable information-gain paragraph, in a new-page mode and an improve-existing-page mode.
- seo-geo AEOA scored GEO analysis of one URL: citability, structure, multimodal content, authority, and technical access, with the most accurate AI-crawler table we found on any shelf.
- ad-creative Paid adsAd copy and creative at volume, grounded in your winning ads, reviews and ad comments: four modes, platform character limits, an S-to-F format taxonomy, and a review page for approvals.
- ads-performance-analytics Paid adsHow to read a paid-media dashboard without fooling yourself: platform reporting quirks, seven attribution models, multi-platform reconciliation, the ROAS-versus-LTV trap, statistical noise, incrementality, and twelve interpretation failures.
- ads Paid adsPaid campaign strategy across Google, Meta, LinkedIn, TikTok and X, with the account decisions (kill, keep, scale, budget) routed to fourteen playbooks and guarded by an audit rulebook.
- analytics Paid adsSet up or audit measurement: a tracking-plan framework, object-action event naming, GA4 and Tag Manager implementation, UTM conventions, and a debugging checklist.
- attribution Paid adsWhich channel caused the sale: seven models and how each one lies, three measurement paradigms, reconciliation of platform, analytics and CRM numbers, and a first-party build track for stitching bookings to their source.
- ab-testing CRODesign an A/B test that can be trusted, then run testing as a programme: hypothesis framework, sample-size table, metrics selection, allocation, the peeking problem, ICE prioritisation and a winners’ playbook.
- conversion-optimization CROA resumable seven-phase journey for one leaking flow: find the leak with numbers, learn the reason from customers, fix message, offer, proof and friction in that order, and prove each fix with a pre-committed test.
- cro-methodology CROConversion Rate Experts’ method as a skill: discover why visitors do not convert, inventory the proof you already have, map every objection to a placed counter-objection, and test bold changes with pre-committed rigour.
- cro CROA page review in seven dimensions ordered by impact: value proposition, headline, CTA hierarchy, scannability, trust, objections, friction. Output in four bands: quick wins, high-impact changes, test ideas, copy alternatives.
- experiment-design CROTwelve considerations for an experiment that answers the question you asked: four-part hypotheses, MDE and power, duration and novelty, what not to test, segments, interactions, ratio metrics, network effects, peeking, pre-commitment, the three result buckets.
- copy-editing ContentSeven sequential editing passes over copy you already have: clarity, voice, so-what, prove-it, specificity, emotion, zero-risk, each looping back to check the earlier sweeps, plus an expert-panel scoring gate for high-stakes pages.
- copywriting ContentConversion copy for any marketing page: five principles, six writing-style rules, an above-the-fold structure with headline formulas, CTA guidelines, and page-type guidance, delivered with annotations and alternatives.
- editorial-qa ContentPre-publish QA that catches problems instead of ticking boxes: brief adherence, voice consistency, fact accuracy as a halt condition, an AI-content audit with eleven tells and six hallucination patterns, SEO and AEO compliance, link and schema validation, and sampling discipline at scale.
- seo-content ContentContent quality and E-E-A-T scoring against Google’s own who-how-why test, with an AI citation readiness score and a deterministic draft cleanup that strips invisible Unicode watermarks and AI-typical phrasing.
- write-content ContentWrite a complete SEO article in four phases (research, content-type decision, knowledge extraction, write) under an explicit anti-AI-slop ruleset: banned words, banned phrases, banned structural patterns, and a 30% rule for details no generic model could produce.
The question people ask
Are Claude skills useful?
Yes, for the part of the work that is reading, structuring and checking. In our 25 runs, the audits found real defects on our own site that we had not logged: a navigation that exists only after JavaScript runs, hero images at a megabyte each, a broken link in the mobile menu, a product page whose copy describes a different product from its photos. None of that needed a key.
No, for the part that is measurement. No skill can see rankings, field Core Web Vitals, ad-account data or AI citations without a connection, and the better ones say so in their own text. The four RampStack skills carry an explicit rule that a stated gap is a complete answer and a fabricated number never is; the six claude-seo skills tell the agent not to guess site content when a fetch fails. In the runs, they did as told.
And the runs found the skills’ own defects: a sample-size table that disagrees with its reference file by 2.3x, a crawler list that conflates OpenAI’s training bot with its search bot, a schema skill that still lists rich results Google retired. Each item page names what we found.
Questions
Questions people ask before installing
Do these work on the free version of Claude?
Skills are a Claude Code and Claude.ai feature; Claude Code needs a subscription or API access, and Claude.ai custom skills depend on your plan. The skills themselves are free, MIT or Apache-2.0 licensed files.
Do they work in Cursor or Codex?
The ones installed with npx skills add land in the folder each agent reads. The claude-seo plugin and its sub-skills are Claude Code only because they call plugin scripts.
What is the difference between a skill and an MCP server?
A skill is instructions the agent reads when a task matches; an MCP server is a program that gives the agent tools, such as reading your GA4 property. Skills tell it how; toolkits let it reach.
How do I install one?
For most: npx skills add owner/repo --skill name. For the plugins: the two /plugin commands on the item page. Our install page walks through both and the manual copy.
Are they safe?
We read every SKILL.md and every script in the folder and wrote what we found into a four-row table on each page: network, shell, secrets, deletes. None of the 25 deletes user files; five call plugin scripts that fetch the web; one edits your codebase if you let it.