Quick answer: the free agency stack is coreyhaines31/marketingskills for strategy and copy, AgriciDaniel/claude-seo for SEO and GEO audits, AgriciDaniel/claude-ads for paid media audits, and Anthropic’s Marketing and brand-voice plugins. What makes it an agency setup is structure: one project folder per client with its own brand skill, and a private git repo so the whole team runs the same skill versions. For the individual marketer’s shortlist, see Claude Code skills for marketers.
Most agencies trying Claude Code hit the same wall. One strategist builds a clever prompt, it lives in their head or a Notion page, and nobody else on the team gets the same output twice.
Skills fix that. A skill is a folder with a SKILL.md file that tells Claude how to run a specific job, and it loads the same way for every person who has it installed. For an agency, that means your audit method, your report format and each client’s brand rules can become files your whole team runs.
This guide covers the skills worth installing for agency work, free and paid, with links, prices checked on 1 November 2026 and the exact install commands from each README. It also covers the parts most lists skip: client confidentiality, git libraries for teams, and how to price AI-assisted services without overselling them.
What are Claude Code skills, and why do agencies care?
Direct answer: Claude Code skills are reusable instruction folders (a SKILL.md plus optional scripts and reference files) that Claude loads when a task matches. They live in ~/.claude/skills (personal), .claude/skills inside a project (shared with anyone who opens that project), or inside plugins. For agencies, they turn one person’s method into a repeatable team process.
If you are new to the format, start with what Claude Code skills are and then come back.
The three places a skill can live
| Location | Who gets it | Agency use |
|---|---|---|
~/.claude/skills/ | Only you, in every project | Your personal writing and research helpers |
.claude/skills/ in a project folder | Anyone who opens that folder | Per-client brand rules, client-specific report formats |
| A plugin (from a marketplace) | Anyone who installs the plugin | Agency-wide audit and reporting skills |
Why this matters more for agencies than in-house teams
An in-house marketer has one brand. An agency has ten, or forty, each with its own voice, banned claims, competitors and reporting cadence. The context switching is where quality slips. A junior account manager writing for a fintech client at 4pm, after writing for a skincare brand at 2pm, is the moment a wrong tone or a non-compliant claim gets through.
Project-level skills attack exactly that problem. When each client has its own folder with its own .claude/skills/brand-voice/SKILL.md, opening the folder loads that client’s rules. Nobody has to remember to paste the brand guide.
There is a second reason. Agencies sell process. If your audit method lives in a skill file under version control, it is an asset you can improve, review and train people on. If it lives in one senior person’s head, it walks out when they do.
Takeaway: skills give an agency consistent output across people and clients, which is the thing clients pay a retainer for.
The best Claude Code skills for marketing agencies in 2026
Direct answer: For most agencies, the free core stack is coreyhaines31/marketingskills for strategy and copy, AgriciDaniel/claude-seo for SEO and GEO audits, AgriciDaniel/claude-ads for paid media audits, and Anthropic’s official marketing and brand-voice plugins. Paid packs are optional add-ons, not requirements.
Star counts and last-push dates below came from the GitHub API on 1 November 2026. Stars measure attention, not quality, so treat them as a signal that a repo is maintained and widely tried, nothing more.
Free skills from GitHub
1. Marketing Skills by Corey Haines (github.com/coreyhaines31/marketingskills)
About 57,800 stars, last pushed 31 October 2026. Free. A large set of marketing skills covering CRO, copywriting, SEO, email sequences, pricing and launch work. Its standout feature for agencies is a shared product marketing context file (.agents/product-marketing.md in v2, with the older .claude/product-marketing-context.md still read as a fallback). Every skill reads that file first, which is a ready-made pattern for per-client context.
Install (from the README):
npx skills add coreyhaines31/marketingskillsOr inside an interactive Claude Code session:
/plugin marketplace add coreyhaines31/marketingskills
/plugin install marketing-skills2. Claude SEO by AgriciDaniel (github.com/AgriciDaniel/claude-seo)
About 25,400 stars, pushed 1 November 2026. Free. Commands for technical SEO, schema, sitemaps, local SEO, backlinks, and a /seo geo command for AI Overviews and generative engine optimization. The README lists a /seo google command that covers Search Console, PageSpeed, CrUX, GA4 and PDF reports, which is the piece agencies want for client deliverables.
/plugin marketplace add AgriciDaniel/claude-seo
/plugin install claude-seo@agricidaniel-claude-seo3. Claude Ads by AgriciDaniel (github.com/AgriciDaniel/claude-ads)
About 10,600 stars, pushed 17 October 2026. Free. The README says it is “made for agencies, consultants, and in-house marketing teams” and that it reads your data and reports back without changing accounts. /ads audit produces passed, failed and unchecked items with evidence, and /ads report renders a saved run as Markdown, HTML or PDF.
/plugin marketplace add agricidaniel/claude-ads
/plugin install claude-ads@ai-marketing-hub-claude-ads4. GEO SEO Claude by Zubair Trabzada (github.com/zubair-trabzada/geo-seo-claude) About 11,400 stars, pushed 1 November 2026. Free. A GEO and SEO audit toolkit focused on AI search visibility and citability, with PDF report generation. Install with the script from the README (macOS and Linux):
curl -fsSL https://raw.githubusercontent.com/zubair-trabzada/geo-seo-claude/main/install.sh | bashRead any install script before piping it to bash. That is basic hygiene, and it matters more when the machine holds client data.
5. Claude Skills by Alireza Rezvani (github.com/alirezarezvani/claude-skills) About 29,100 stars, pushed 31 October 2026. Free. A broad library split into plugins. The marketing plugin is listed in the README as 43 marketing skills.
/plugin marketplace add alirezarezvani/claude-skills
/plugin install marketing-skills@claude-code-skills6. Anthropic Knowledge Work Plugins (github.com/anthropics/knowledge-work-plugins)
About 34,300 stars, pushed 1 November 2026. Free and official. The marketing plugin drafts content, plans campaigns, enforces brand voice and reports on performance, with connectors listed for Slack, Canva, Figma, HubSpot, Ahrefs, SimilarWeb and Klaviyo. The marketplace also includes a partner-built brand-voice plugin.
claude plugin marketplace add anthropics/knowledge-work-plugins
claude plugin install marketing@knowledge-work-plugins
claude plugin install brand-voice@knowledge-work-pluginsThe plugin names marketing and brand-voice come from the repo’s marketplace.json, and the command pattern from its README.
7. SE Ranking SEO Skills (github.com/seranking/seo-skills) About 190 stars, last pushed 25 June 2026. The skills are free, but they are built for the SE Ranking MCP, so you need an SE Ranking account for live data (see seranking.com/claude-seo-skills.html). Useful if your agency already pays for SE Ranking.
/plugin marketplace add seranking/seo-skills
/plugin install seo-skills@seranking8. Awesome Claude Skills by Composio (github.com/ComposioHQ/awesome-claude-skills) About 78,500 stars, pushed 18 September 2026. A curated directory, not a skill pack. Use it to find niche skills (a specific CRM, a reporting tool) rather than installing it whole.
Directories worth browsing
| Directory | What it is good for |
|---|---|
| claudemarketplaces.com | Searching plugins and skills by name, including the marketing packs above |
| claudskills.com | Browsing an open SKILL.md registry by use case, such as SEO research |
| skillsmp, kissmyskills, cultofclaude | More aggregators; quality varies, so always click through to the source repo |
A directory listing is not a review. Before installing anything, open the source repo, read the SKILL.md, and check the last push date.
Paid skill packs
Prices below were read from each live page’s price metadata on 1 November 2026. They can change, so check before you buy.
| Product | Where | Price | What the seller says it includes |
|---|---|---|---|
| Master Marketing Skills Bundle (Inflectual) | Gumroad | $167 one-time | 167 Claude skills across email, SEO, paid ads, strategy, research, copy, content and social (seller’s claim) |
| The Ultimate Claude SEO & AEO Skills Pack (Samuel Uebersax) | Contra | $39 (shown against $79) | An SEO and AEO skill system described as practitioner-built (seller’s claim) |
| SuperSkills | superskills-one.vercel.app | $50 one-time | 139 skills across 15 categories, including marketing, CRO and SEO (seller’s claim) |
None of these pages showed countdown timers or “spots left” counters when checked. Some other packs in this space do use fake urgency, and they are left out of this guide on purpose. The skill counts above are the sellers’ own figures and have not been independently verified.
Free vs paid Claude Code skills: which should an agency buy?
Direct answer: Start free. The open-source repos above cover audits, copy, ads and brand voice, and you can read every line before trusting it with client work. Buy a paid pack only when it saves your team build time on a specific gap, and read it as carefully as you would a free one.
| Factor | Free GitHub skills | Paid packs |
|---|---|---|
| Cost | $0 | $39 to $167 one-time (packs listed here) |
| Can you read the source before using it? | Yes, always | Usually only after purchase |
| Update visibility | Public commit history and push dates | Depends on the seller |
| Install | npx skills add, /plugin or script from README | Usually a zip you copy into a skills folder |
| Licensing for client work | Check the repo license file | Check the seller’s terms |
| Agency fit | Strong for audits and reports | Can fill category gaps fast |
| Risk | Abandoned repos, unreviewed scripts | Paying for prompts you could write yourself |
Here’s the thing: the value of a skill to an agency is mostly in the parts you customize. Your report template, your client’s banned claims, your audit scoring. No pack ships those. So the real question is which base saves you the most editing time, and you can only answer that by opening the files.
One practical rule: whatever you buy or install, fork it into your own team library (covered below) before rolling it out. That way an upstream change never silently alters a client deliverable.
Using client audits as lead magnets (SEO, GEO and ads)
Direct answer: Agencies can run an SEO, GEO or paid media audit skill against a prospect’s public site, review the findings by hand, and send a short branded summary as a free audit. The skill does the first pass. A human must check every finding before it reaches a prospect, because a wrong claim in a sales audit costs you the deal.
What each audit type can actually see
| Audit | Skill to start with | Needs prospect access? | What it checks |
|---|---|---|---|
| Technical and on-page SEO | claude-seo (/seo technical, /seo schema) | No, public pages | Crawlability, schema, sitemaps, images |
| GEO / AI search | claude-seo (/seo geo) or geo-seo-claude | No | Citability, AI crawler access, llms.txt |
| Paid media | claude-ads (/ads audit) | Yes, exports or account data | Tracking, structure, wasted spend signals |
| Landing pages | marketingskills CRO skills | No | Message, forms, trust signals |
Notice the split. SEO and GEO audits work from public data, so they suit cold outreach. Ads audits need account data, so they belong after a first call, once the prospect has shared an export.
A sensible audit-as-lead-magnet workflow
- Run the audit skill on the prospect’s homepage and two key pages.
- Read the raw output. Delete anything you cannot verify on the live site.
- Pick the three findings with the clearest fix. Prospects act on three, not thirty.
- Write a one-page summary in your voice, with a screenshot per finding.
- Offer the full audit on a call, where you can explain context.
The claude-seo README itself shows an illustrative comparison of agency fees against a free tool. Take that as a warning rather than a threat: if your audit is only the raw skill output, a prospect can run it themselves. Your value is the judgment layer, the prioritization and the plan.
For deeper SEO tactics see Claude Code skills for SEO, and for paid media see Claude Code skills for PPC.
White-label PDF reports with Claude Code skills
Direct answer: Several free skills already export PDFs. claude-ads renders saved audit runs as Markdown, HTML or PDF, claude-seo lists PDF reports under its Google APIs command, and geo-seo-claude generates PDF reports. To white-label, you wrap that output in your own report skill with your logo, colors and section order.
How to build a white-label report skill
The simplest pattern is a small project-level skill that sits on top of an audit skill:
.claude/skills/agency-report/
SKILL.md # section order, tone, what to include and exclude
template.html # your branded layout with placeholders
logo.png
style.cssIn SKILL.md, tell Claude to take the audit output, map it into your template’s sections (summary, top three priorities, full findings, next steps), and render HTML that you print to PDF. Keep the brand assets in the folder so every team member produces the same look.
What to put in every client report
| Section | Why it is there |
|---|---|
| One-paragraph summary | Clients read this and little else |
| Top three priorities with owners | Turns findings into work |
| Evidence per finding | Screenshot, URL or metric, so it can be checked |
| What changed since last report | Shows progress on a retainer |
| What the report did not cover | Sets honest scope |
But here’s the catch: an automated report is only as good as its data source. If a skill pulls from an API you have not connected, it may fill gaps with estimates. Make your report skill label anything estimated, and never let a number reach a client that you cannot trace to a source.
If you want to build your own from scratch, how to build a Claude Code skill walks through the file structure.
Per-client brand files with project-level skills
Direct answer: Give each client their own project folder with a .claude/skills/ directory holding a brand voice skill, an approved and banned claims list, and audience notes. Opening that folder loads that client’s rules automatically, so writers stop pasting brand guides into prompts.
A client folder structure that scales
clients/
acme-fintech/
.claude/skills/
brand-voice/SKILL.md # tone, words to use and avoid
claims/SKILL.md # approved claims, banned claims, legal notes
.agents/product-marketing.md # read by marketingskills
briefs/
reports/
bloom-skincare/
.claude/skills/...This works neatly with the marketingskills context file, since those skills read .agents/product-marketing.md first. Write that file once per client and the copy, CRO and email skills all pick it up.
What goes in a client brand skill
- Voice in five to ten rules with examples of right and wrong
- Words the client never uses (their banned list, not yours)
- Approved claims with the source for each
- Claims that need legal sign-off
- Audience segments and the one objection each segment raises
- Competitors you must not name in copy
Anthropic’s brand-voice plugin can help draft these from existing material, which is a quicker start than a blank page. See Claude Code skills for brand voice for more detail.
Takeaway: the client folder is the unit of work. Everything a writer needs for that client should load when they open it.
Writing proposals and pitches with skills
Direct answer: A proposal skill stores your agency’s proposal structure, pricing rules, case-study library and scope language, so every proposal follows the same format. Pair it with a prospect research step and an audit summary, and a strategist reviews a strong draft rather than starting from nothing.
Pieces of a proposal skill
| File | Contents |
|---|---|
SKILL.md | Section order, tone, rules (for example, never promise rankings) |
services.md | Your service menu with scope and deliverables |
pricing-rules.md | How you price, minimums, what is included |
proof.md | Only real, approved case studies and permissions to cite them |
terms.md | Standard scope, exclusions and change-request language |
The proof.md file is the one to guard. A proposal skill will happily write “we grew traffic 300%” if nothing stops it. Tell it in plain words to use only the case studies in that file and to leave a visible gap where proof is missing.
Anthropic’s sales plugin in the same knowledge-work marketplace covers prospect research and outreach drafts, which can feed the research section. For sales-side workflows, see Claude Code skills for sales teams.
Building a team skill library in git
Direct answer: Keep your agency’s skills in a private git repository, structured as a plugin marketplace, so every team member installs the same versions with /plugin marketplace add your-org/your-repo. Changes go through pull requests, so a senior person reviews any edit to an audit method or report template before it reaches clients.
Why git, and not a shared drive
- Review: a pull request shows exactly what changed in a skill.
- Rollback: if a change breaks a report, you revert one commit.
- Pinning: forked third-party skills stay at the version you tested.
- Onboarding: a new hire installs one marketplace and has the agency method.
A simple layout
agency-skills/
.claude-plugin/marketplace.json
plugins/
audits/ # forked claude-seo and claude-ads, plus your scoring
reports/ # white-label report skill
proposals/ # proposal skill
writing/ # house style, humanizer rulesKeep client-specific skills out of this repo. They belong in each client’s own folder or repo, with access limited to the people on that account. That separation is also your first privacy control.
The official plugin docs explain marketplace files; our guide on how to install Claude Code skills covers the install side for non-technical staff.
Client data privacy and confidentiality
Direct answer: Treat Claude Code like any other tool that touches client data. Check your Anthropic plan’s data terms, keep client material in separate folders, never commit credentials or exports to shared repos, read third-party skills and scripts before running them, and get client approval in your contract for AI-assisted work.
A practical confidentiality checklist
| Risk | Control |
|---|---|
| Client data used in ways the contract forbids | Review Anthropic’s commercial terms for your plan and match them to client contracts |
| One client’s data visible to another’s team | Separate folders or repos per client, with access by account |
| API keys and exports in git | .gitignore for .env, exports and report data; use a secrets manager |
| Third-party skill runs an unknown script | Read SKILL.md and any scripts before install; fork and pin |
| MCP server sends data to a vendor | Read the vendor’s data policy before connecting a client account |
| AI-written claims reach a regulated client’s ads | Human and, where needed, legal review before publishing |
A note on MCP servers: vendor servers (SE Ranking, Ahrefs, HubSpot and others listed in Anthropic’s marketing plugin) connect Claude to live accounts. That is powerful, and it means client data flows through another vendor. Use read-only access where the vendor offers it, and connect a client’s account only with their permission.
Claude-ads states in its README that it reads data and reports back without changing accounts. That read-only design is the right default for agency tooling. Prefer skills that behave this way for anything touching client ad accounts.
This section is general guidance, not legal advice. For GDPR, HIPAA or similar obligations, talk to your counsel.
How to price productized AI services honestly
Direct answer: Price on the outcome and the expert review, not on the AI. Clients pay for judgment, accountability and a result they can act on. Be open that you use AI tools, do not bill old hourly rates for work that now takes minutes, and never promise rankings, citations or ROAS figures you cannot control.
Honest pricing models for skill-powered services
| Model | Example | Works when |
|---|---|---|
| Fixed-fee audit | SEO and GEO audit with a prioritized plan | Scope is clear and repeatable |
| Productized retainer | Monthly report plus three implemented fixes | Ongoing work with defined deliverables |
| Free audit as lead magnet | Three-finding summary | You have a clear paid next step |
| Implementation sprint | Fix the top ten issues in two weeks | Client has dev access to give you |
This guide does not give price points, because there is no reliable public benchmark for AI-assisted agency services and any number here would be a guess. Base yours on your costs, your review time and what your market already pays for the outcome.
Lines to keep out of your sales copy
- “Guaranteed page one” or “guaranteed AI citations”
- “Fully automated” when a person reviews the work (say a person reviews it; that is the selling point)
- Time-saving figures you have not measured on your own team
- Any case study you lack permission to share
What to say instead: describe the method, the review step and the deliverable. “Every audit is run with our documented method and checked by a senior strategist before you see it” is true, specific and easier to defend than a big number.
For solo operators pricing similar services, see Claude Code skills for freelancers.
Common mistakes agencies make with Claude Code skills
Direct answer: The usual mistakes are installing dozens of skills nobody maintains, sending raw skill output to clients, mixing client data in one folder, trusting star counts as quality, and selling automation as if no human reviews it.
- Installing everything. Fifty overlapping skills confuse which one triggers. Pick a core stack and remove the rest.
- Shipping raw output. Audit skills produce first drafts. Clients deserve reviewed findings.
- One folder for all clients. That leaks context and breaks confidentiality.
- Never updating forks. Pin versions, but review upstream changes every month or so.
- No owner. Each skill in the team library needs one person responsible for it.
- Overclaiming in proposals. A proposal skill without a proof file will invent proof.
Your 30-day rollout plan
Direct answer: Week one, install the free core stack in a test project. Week two, build one client brand folder. Week three, build your white-label report skill. Week four, move it all into a private git marketplace and train the team.
| Week | Task | Done when |
|---|---|---|
| 1 | Install marketingskills, claude-seo, claude-ads, official marketing plugin | Each runs on a test site |
| 2 | Build one client folder with brand voice and claims skills | A writer produces on-brand copy without pasting a guide |
| 3 | Build the white-label report skill | One real report goes out after review |
| 4 | Private git marketplace, pull-request reviews, team training | A new team member installs it in one command |
Non-technical team members can follow Claude Code skills for non-developers. For the wider marketing toolkit, see Claude Code skills for marketers.
FAQ
What are the best Claude Code skills for marketing agencies?
A strong free core is coreyhaines31/marketingskills, AgriciDaniel/claude-seo, AgriciDaniel/claude-ads and Anthropic’s official marketing and brand-voice plugins. Add paid packs only for specific gaps.
Are Claude Code skills free?
Most are. The GitHub repos in this guide are free. Some paid packs exist, such as the $167 Inflectual bundle, the $39 Contra SEO and AEO pack and the $50 SuperSkills bundle.
How do I install a Claude Code skill?
Use npx skills add owner/repo where the README supports it, or /plugin marketplace add owner/repo then /plugin install name@marketplace in an interactive session. Always copy the exact command from the README.
Can Claude Code create white-label client reports?
Yes, with setup. claude-ads exports Markdown, HTML or PDF, and claude-seo and geo-seo-claude include PDF reporting. You add your branding through your own report skill.
Is it safe to use Claude Code with client data?
It can be, with controls: check your plan’s data terms, separate client folders, keep secrets out of git, review third-party scripts and get client consent in your contracts.
What is the difference between a skill and a plugin?
A skill is one instruction folder. A plugin is a package that can hold several skills plus commands, agents and MCP connections, installed from a marketplace.
Where should client-specific skills live?
In that client’s project folder under .claude/skills/, so they load only when someone opens that client’s work.
Can I share skills across my agency team?
Yes. Put them in a private git repo set up as a plugin marketplace, and each person installs it once.
Do I need to code to use Claude Code skills?
No. You need to run a few terminal commands and edit Markdown files. Writing a skill is mostly writing clear instructions.
Can Claude Code run a GEO audit?
Yes. claude-seo has a /seo geo command and geo-seo-claude is built for AI search visibility audits. Results still need human checking.
Are paid Claude skill packs worth it?
Sometimes. They can save build time, but you usually cannot read them before buying, and much of their value depends on how much you customize afterward. Start with free options.
Should agencies tell clients they use AI?
Being open is the safer and more honest choice, and some contracts require it. Frame it around your review process and results.
How should I price an AI-assisted SEO audit?
Price the outcome and your expert review time, based on your own costs and market. There is no reliable public benchmark, so be wary of anyone quoting one as fact.
Can a proposal skill write case studies?
It should only use case studies you supply in a proof file. Without one it may invent results, which you must never send.
Do Claude Code skills work with Google Ads and Meta accounts?
claude-ads audits Google, Meta, LinkedIn, TikTok and Microsoft Ads from exports or connected data, and its README says it does not change accounts on its own.
Sources and methodology
All GitHub star counts and last-push dates were pulled from the GitHub API on 1 November 2026. Install commands were copied from each repository’s README or marketplace file on the same date. Paid prices were read from each live product page’s price metadata on 1 November 2026. Seller claims about skill counts are labelled as claims and were not independently verified. No product here was tested on live client accounts for this article, so no performance results are stated.
- coreyhaines31/marketingskills README
- AgriciDaniel/claude-seo README
- AgriciDaniel/claude-ads README
- zubair-trabzada/geo-seo-claude README
- alirezarezvani/claude-skills README
- anthropics/knowledge-work-plugins README and marketplace.json
- seranking/seo-skills README and seranking.com/claude-seo-skills.html
- Gumroad, Contra and SuperSkills product pages
The short version
Skills let an agency turn its method into files: an audit everyone runs the same way, a report that always looks like yours, and client folders that load the right voice on their own. Start with the free stack, build one client folder and one report skill, then move them into git. Keep a human on every client-facing output, and price the judgment, not the tool.
Next step: install the free core stack on a test site this week, then read our guide to building your first skill to make the report template yours.
