Marketing skills for Claude Code: four packs compared

Of the four packs of Claude marketing skills compared here, start with Corey Haines’ marketingskills if you want marketing skills for Claude Code and nothing else running on your machine. It is 53 Markdown skills that all read one file about your product. Take aaron-marketing-skills if you want a bigger system with commands and project memory, and you accept hooks on your tool calls. zubair-trabzada’s AI Marketing Suite suits agency work that ends in a scored audit or a PDF for a client. ericosiu’s ai-marketing-skills is a set of Python workflows you set up one folder at a time, with your own API keys.

Shared context, phased disciplines, subagents and scripts

In Corey’s pack all 53 skills point to .agents/product-marketing.md, which holds your product, audience and positioning. The product-marketing skill creates it and can draft the first version from your repository. Detail sits in 285 reference files that load only when a task needs them.

Seven disciplines of 16 skills, each in four phases, plus eight protocol skills make up the 120 in aaron-marketing-skills. Eight commands sit on top, and /aaron-marketing:auto routes a plain-language goal to the right one.

Everything in zubair-trabzada’s suite goes through /market. For /market audit <url>, Claude Code launches five subagents in parallel, each fetches pages with WebFetch, and the weighted score lands in MARKETING-AUDIT.md.

Code does the work in ericosiu’s repo, a folder of 36 workflows. cro_audit.py scores a page on eight dimensions in Python over the parsed HTML, and Growth Engine runs Mann-Whitney U tests.

Plugin, npx or curl to bash

Corey’s pack installs with npx skills add coreyhaines31/marketingskills or /plugin install marketing-skills, and the plugin route works only in an interactive CLI session. One trap is in the README itself. Run npx from inside an agent session and it may install into .agents/skills/, which Claude Code does not read, so add -a claude-code. If you ran 1.x, version 2.0 renamed 17 skills, and the upgrade leaves the old folders next to the new ones until you delete them.

aaron-marketing-skills installs as a plugin (/plugin install aaron-marketing@aaron). It adds no MCP server, and per the README every skill works with pasted data.

A curl ... install.sh | bash line puts zubair-trabzada’s skills into ~/.claude/skills and its subagents into ~/.claude/agents, so they load in every project you open. None of its 20 skill and agent files starts with YAML frontmatter, so none has a description field, the field Corey’s maintainer notes call “critical for skill discovery”, and you should plan on typing the /market commands.

ericosiu’s repo has no plugin manifest. The README tells you to copy skills into .claude/skills by hand, and warns that copying only SKILL.md leaves the bundled workflows incomplete.

What runs underneath

Hooks are where aaron-marketing-skills differs most. Its PreToolUse and PostToolUse hooks match Write, Edit, Bash and every MCP tool, and at session start it loads up to 9000 bytes of memory/hot-cache.md into context. Without python3, the hook denies tool calls that touch memory/. The rest needs only Bash and the Python standard library.

ericosiu’s setup is the heaviest: 30 separate requirements.txt files and a .env per category. 13 Python files read ANTHROPIC_API_KEY and call the Anthropic API directly, so those scripts need a key on top of your Claude Code login. 24 of the 36 SKILL.md files open with a preamble that runs a version check against the GitHub API. Remote telemetry is opt-in, and at this commit its endpoint is a stub.

Two of zubair-trabzada’s scripts, analyze_page.py and competitor_scanner.py, set ssl.CERT_NONE, which turns off TLS certificate checks. PDF reports need reportlab. The competitors skill calls python scripts/competitor_scanner.py, a relative path, while the installer copies the scripts to ~/.claude/skills/market/scripts.

Nothing in Corey’s plugin manifest declares a hook. Its 66 optional CLIs are single Node.js files that read API keys from environment variables, and the GA4 one has a --dry-run that shows the request instead of sending it.

What the Tokens column leaves out

For any of these Claude marketing skills, the part that stays loaded is the description. The body loads when the skill runs. The Tokens column in the table above is that up-front part, so it understates what a run costs. Corey’s ai-seo has a one-paragraph description and a SKILL.md body of 5,325 words. aaron’s session-start hook adds its hot cache when the file exists.

Questions

How do I install Corey Haines' marketing skills in Claude Code?

Run npx skills add coreyhaines31/marketingskills, or add the marketplace and run /plugin install marketing-skills. If you run npx from inside an agent session, pass -a claude-code, because otherwise the skills can land in .agents/skills, which Claude Code does not read.

Do Claude marketing skills need API keys?

The skills themselves are Markdown and work without keys. Corey's optional CLIs and ericosiu's scripts read keys from environment variables or .env files, and 13 of ericosiu's Python files read ANTHROPIC_API_KEY.

Sources

coreyhaines31/marketingskills at 96798cf
README.md · AGENTS.md · .claude-plugin/marketplace.json · skills/product-marketing/SKILL.md · skills/cro/SKILL.md · tools/clis/README.md · tools/clis/ga4.js · .claude-plugin/plugin.json · skills/ai-seo/SKILL.md

aaron-he-zhu/aaron-marketing-skills at 8a359cb
README.md · CONNECTORS.md · hooks/hooks.json · hooks/claude-hook.sh

zubair-trabzada/ai-marketing-claude at e5aa0ea
README.md · install.sh · market/SKILL.md · skills/market-audit/SKILL.md · skills/market-competitors/SKILL.md · agents/market-content.md · scripts/analyze_page.py · scripts/competitor_scanner.py · requirements.txt

ericosiu/ai-marketing-skills at d2237ec
README.md · growth-engine/SKILL.md · conversion-ops/SKILL.md · conversion-ops/cro_audit.py · content-ops/scripts/editorial-brain.py · content-ops/.env.example · telemetry/telemetry_log.py · telemetry/version_check.py