AI Agent Marketing Skills: A Complete Operator's Guide

If you've watched AI agents go from "Claude can suggest copy" to "Claude actually launched a campaign," you've watched the Agent Skills ecosystem mature. Skills are the missing layer between a generic LLM and a domain-specific operator. They turn "smart but generic" into "actually useful for marketing work."

This guide covers everything an operator needs: the Agent Skills spec mechanics, the dominant repos (Corey Haines' OG coreyhaines31/marketingskills, Hyper's execution-layer hyperfx-ai/marketing-skills, plus OpenClaudia, BrianRWagner, zubair-trabzada, and aggregators), how to install and combine them, and the operator-grade workflow patterns that produce real ROI.

What is an Agent Skill

The Agent Skills spec was published as part of the Claude Code rollout and adopted across the AI tooling ecosystem through 2025-2026. Skills differ from plugins or extensions in three important ways:

1. Skills are content, not code. A skill is a Markdown file with structured instructions. The agent reads it at task time. There's no runtime, no separate process, no API surface.
2. Skills are agent-agnostic. The same skill works across Claude Code, Cursor, Codex, Gemini CLI, and OpenClaw because the spec is implementation-neutral.
3. Skills compose. A complex workflow can pull from 5-10 skills simultaneously without conflict; the agent picks which to load based on the task.

How skills are structured

A typical skill folder:

my-skill/
├── SKILL.md              # Required. YAML frontmatter + Markdown instructions.
├── references/           # Optional. Long-form material loaded on demand.
├── scripts/              # Optional. Executable helpers.
└── assets/               # Optional. Templates, configs, fixtures.

The SKILL.md file:

---
name: my-skill
description: One-paragraph description that helps the agent decide when to invoke this skill.
---

# My Skill: instructions

[Markdown body with procedural knowledge, decision frameworks, examples,
and references to assets/, scripts/, references/.]

The name field must match the folder name. The description is what the agent uses for skill selection; write it clearly.

Installation via npx skills

Most marketing skills repos ship via the npx skills CLI:

# Install all skills from a repo
npx skills add coreyhaines31/marketingskills
npx skills add hyperfx-ai/marketing-skills

# Install specific skills only
npx skills add coreyhaines31/marketingskills --skill page-cro copywriting

# Update installed skills
npx skills update

# List installed skills
npx skills list

The CLI installs to ~/.agents/skills/<skill-name>/ and creates symlinks in Claude Code's .claude/skills/, Cursor's .cursor/skills/, and Codex's skill directory automatically when those agents are detected.

For monorepo setups, you can also install skills as Git submodules:

git submodule add https://github.com/hyperfx-ai/marketing-skills.git skills/external/hyper-marketing

The marketing skills ecosystem in 2026

The dominant repos as of May 2026:

For the ranked-by-fit comparison, see /blog/best-marketing-skills-github-repos-2026.

Procedural skills vs MCP execution

The single most important distinction in this ecosystem:

Procedural-knowledge skills

Tell Claude HOW to do a task. Methodology, decision frameworks, examples. Claude reads the skill, applies the methodology, produces output. The skill itself doesn't connect to external services.

Example: coreyhaines31/marketingskills/seo-audit teaches Claude how to audit a website for SEO. Claude reads the skill, then uses its native tools (WebFetch, browsing) to actually do the audit and produce findings.

MCP execution skills

Tell Claude how to do a task AND give Claude a connected execution surface to actually do it against external systems. Skills are paired with MCP (Model Context Protocol) servers that expose specific tools.

Example: hyperfx-ai/marketing-skills/google-ads teaches Claude how to plan and launch Google Ads campaigns. Paired with the Hyper Cloud MCP, Claude can also actually launch the campaign by calling the Google Ads API through the MCP layer.

Why both matter

For research, audit, copy, and methodology tasks, procedural-only skills are sufficient. For execution tasks (launch a campaign, send an email sequence, update a budget), MCP execution is required. Most operator-grade stacks combine both:

• Procedural skills from Corey Haines or OpenClaudia for top-of-funnel and methodology
• MCP execution skills from Hyper for ad-account and platform operations

Combining repos in practice

A real operator stack as of mid-2026:

# Procedural marketing department
npx skills add coreyhaines31/marketingskills

# Execution layer for paid ads, social, analytics
npx skills add hyperfx-ai/marketing-skills

# Optional: open-source backup
npx skills add OpenClaudia/openclaudia-skills

# Optional: framework methodologies
npx skills add BrianRWagner/ai-marketing-claude-code-skills

This gives Claude (or Cursor, Codex) access to roughly 100 marketing skills plus the production MCP execution layer through Hyper. Skill names don't typically collide because each repo focuses on different operations: Corey on CRO/SEO/content, Hyper on ad-ops execution, BrianRWagner on frameworks. When two repos do offer similar skills, the agent picks based on the task description and context.

Operator-grade workflows

What this stack actually produces in daily use:

Workflow 1: Weekly paid media review

Operator says: "Review our Meta and Google performance this week, flag underperformers, draft fixes."

Agent reads:

• hyperfx-ai/marketing-skills/meta-ads for the weekly review methodology
• hyperfx-ai/marketing-skills/google-ads for Google-specific patterns
• hyperfx-ai/marketing-skills/analytics-insights for cross-platform reconciliation
• Hyper Cloud MCP pulls live account data

Output: a structured weekly review document with actual ROAS data, identified underperformers, and recommended actions. Operator approves; Hyper MCP executes.

Workflow 2: Competitor research

Operator says: "Research our top 5 competitors' Meta ad creative and produce a battle card."

Agent reads:

• coreyhaines31/marketingskills/competitor-alternatives for positioning frameworks
• hyperfx-ai/marketing-skills/competitor-intel for execution methodology
• hyperfx-ai/marketing-skills/meta-ads-library for live ad library research

Output: structured battle card with creative samples, targeting analysis, and positioning recommendations.

Workflow 3: SEO content production

Operator says: "Audit our top 20 blog posts, identify thin-content posts, draft refresh outlines."

Agent reads:

• coreyhaines31/marketingskills/seo-audit (98.4K installs; battle-tested methodology)
• coreyhaines31/marketingskills/programmatic-seo
• coreyhaines31/marketingskills/copywriting

Output: 20-post audit with action items per post.

Workflow 4: Launch a campaign end-to-end

Operator says: "Launch a TikTok campaign for our spring product line, 5K USD/week budget, run it for 4 weeks."

Agent reads:

• hyperfx-ai/marketing-skills/tiktok-ads for launch playbook
• hyperfx-ai/marketing-skills/ad-creative-generation for creative
• hyperfx-ai/marketing-skills/analytics-insights for tracking setup
• Hyper Cloud MCP authenticates against TikTok Ads Manager and executes

Output: live TikTok campaign with audiences, ad sets, creative, and tracking.

This last workflow is the one that procedural-only repos cannot deliver. Skills tell Claude how; MCP execution lets Claude do.

Writing your own marketing skills

If you have a unique marketing methodology, packaging it as a skill is the fastest distribution path. The minimal viable skill:

1. Create a folder: my-marketing-skill/
2. Add SKILL.md with YAML frontmatter:
   Copy

   ---
   name: my-marketing-skill
   description: One-paragraph description.
   ---
   

3. Write the procedural body in Markdown. Include decision frameworks, examples, and clear when-to-invoke triggers.
4. Test with Claude Code locally; iterate based on actual agent invocation patterns.
5. Publish to GitHub.
6. Submit to skills.sh and VoltAgent/awesome-agent-skills for discovery.

For full skill-authoring conventions, see Anthropic's official Skills documentation and the patterns in coreyhaines31/marketingskills and hyperfx-ai/marketing-skills.

How Hyper fits in this ecosystem

The argument for adding hyperfx-ai/marketing-skills plus the Hyper Cloud MCP to whatever skills stack you've already assembled: the procedural-only repos give Claude the methodology to do marketing work; Hyper gives Claude the execution surface to actually do it against your ad accounts, social platforms, analytics tools, and 80+ other integrations.

For operators running 5K USD/month or more in paid spend, the math is straightforward. Procedural skills are free; methodology helps but doesn't ship campaigns. Hyper at 49 USD/month with a free 30-day trial is the cheapest production execution layer available; pairing it with Corey Haines' procedural depth gives you the most complete agent marketing department available in 2026. Real customer outcomes at /blog/ai-marketing-case-study.

For the ranked comparison of all dominant marketing skills repos, see /blog/best-marketing-skills-github-repos-2026.

Frequently asked questions