marketing-skills: a versioned AI agent skill library for SEO and marketing workflows

A skill library built for repeatable AI agent workflows, not one-off prompts

A one-off prompt written for a task disappears when the task is done. Recreating the same workflow for a similar task means rewriting the instructions from scratch, with no guarantee the method is the same. Marketing Skills addresses this by packaging marketing and SEO workflows as Agent Skills: version-controlled, reusable instructions an AI agent loads when a matching task comes in.

What separates a one-off prompt from an Agent Skill is persistence. A skill's methods, decision criteria, input requirements, and verification steps are written into a file that the agent reads when a matching task appears, rather than being composed fresh each time. The library supports Cursor, Claude Code, and OpenClaw, and the README describes it as having no lock-in between those tools.

What lives inside a SKILL.md file: scope, methods, inputs, and verification

Each Agent Skill is a directory with a SKILL.md file. That file carries 2 distinct layers: discovery metadata that tells the agent which tasks match this skill, and procedural instructions that govern what to do once it is loaded.

A skill's procedural content addresses 6 distinct concerns. The first 2 are scope: which task this skill handles, and where its responsibility ends relative to adjacent skills. The next 2 are process: how to proceed step by step, and what inputs must be gathered before starting (including what to do when they are missing). The last 2 are quality: what the output should contain, and how to handle claims that require current external evidence. The missing-information handling is the one that matters most in practice, because a skill that detects what it cannot answer will ask for the information rather than inventing it. Without that check, an agent fills gaps from its training data rather than from the project's actual facts.

Installing specific skills vs. loading the entire library

The npx CLI handles installation. To install 3 skills relevant to a keyword and content task, the command is:

npx skills add kostja94/marketing-skills \
  --skill seo-strategy keyword-research content-strategy

To browse all available entries before deciding:

npx skills add kostja94/marketing-skills --list

To install everything at once:

npx skills add kostja94/marketing-skills

The README recommends loading 2 to 3 relevant skills per task rather than the full library. Installing the complete set is possible, but it puts more context into a single task than most tasks need. The CLI installs skills only; cloning the repository is necessary when the project also needs the templates/ directory or the full documentation set under docs/.

170+ entries across 9 categories, but the count includes compatibility shims

The repository description says "160+ open-source skills" while the README body says "170+ installable entries across 9 top-level categories." The README explains the discrepancy: the total includes both active marketing skills and compatibility entries for page and component names that have moved to Pagina and Bricks. A compatibility entry routes the old skill name to the new project without duplicating the underlying guidance. So the number of active marketing skills is smaller than 170+, and the real count is not stated.

The 9 categories are SEO, Content, Paid Ads, Channels, Platforms, Strategies, Analytics, Pages, and Components. The Pages and Components categories are largely compatibility entries: complete page construction routes to Pagina and interface component construction routes to Bricks, with only brand visual, favicon, and URL slug remaining under Marketing Skills. Anyone building a count of active skills for their own use should consult docs/skills-reference.md rather than relying on the header number.

Contextus as the recommended context layer, and what happens without it

Marketing Skills reads task-relevant context but does not create a context schema of its own. The README recommends Contextus as the maintained source for durable product truth, installed via:

npx skills add kostja94/contextus --skill contextus

Contextus provides a root contextus.md as the standard entry point for organization, product, audience, brand, website, technical, decision, and change context. Without it, a project can still use Marketing Skills by providing relevant documents or facts in the task description, and the README says the skills remain usable in that case.

An important design boundary: agent research stays provisional. When an agent finds information during a task, that finding does not automatically become part of the project's confirmed facts. Someone has to confirm it and write it into the project context explicitly, keeping agent-generated content separate from verified product truth. This is the design choice that separates a structured workflow system from agents that silently treat research findings as established facts.

Where the ecosystem boundary runs: when Pagina and Bricks take over

The README shows a 3-project ecosystem. Contextus holds durable product truth. Marketing Skills handles SEO, content, channels, paid ads, strategy, and analytics. When a task requires building a complete page, Marketing Skills routes to Pagina; when it requires building an interface component, it routes to Bricks. That routing is what the compatibility entries in the Pages and Components categories do.

A complete marketing task may cross 2 or more repository boundaries. An article-plus-landing-page project uses Marketing Skills for the content strategy and keyword work, then adds Pagina for the page structure, with each project doing only the work it owns. A project that stays within content strategy and SEO has no need for Pagina or Bricks and can use Marketing Skills on its own. The docs/usage.md file has the full prompt-to-skill mapping and docs/skills-reference.md has ownership information for each entry.

What the skills do not do: no platform access, no independent research

The README is direct about the limits of what a skill can provide. Skills do not replace product facts, professional judgment, platform access, or human approval. An SEO strategy skill can produce a structured plan from the provided context, but it cannot log into Google Search Console, run a crawl, or verify whether a sitemap is indexed. Those steps require the agent to have the appropriate tool access and platform credentials separately.

The dependence on provided context is the other constraint. A skill that asks for missing information is behaving correctly, but if the missing information is not available in the task or in contextus.md, the agent cannot produce accurate output. Skills that require current external evidence, such as competitor pricing or live search volume data, name that requirement explicitly. Without those external inputs, the output will be based on whatever training knowledge the agent has, not on verified current facts.

For teams that already have a prompt library for marketing tasks, Marketing Skills adds version control and structured routing but not a new capability the team lacked before.

Editorial conclusion

Marketing Skills is the right choice for AI agent workflows where the work needs repeatable, versioned methods rather than ad-hoc instructions. The 170+ entries include compatibility shims alongside active skills, so checking docs/skills-reference.md before loading a skill confirms whether it is still active or now delegated to Pagina or Bricks. Projects without Contextus can still use the library, but the agent will ask for missing context during the task rather than finding it in a standard file. Loading 2 to 3 skills per task is the approach the README recommends, not the complete library at once.

Frequently asked questions

Official sources

1. kostja94/marketing-skills on GitHub
2. License: MIT
3. Project website
4. README
5. Releases