# RampStack Claude Skills: 103 opinionated skills with two lock files and a counted README > A single curated library of Claude Skills spanning the whole website lifecycle, sold on uniform structure and cross-references rather than breadth of coverage alone. The interesting mechanics are the two plugin marketplaces, the auto-counted README markers, and one skill that quietly assumes a paid tool. **rampstackco/claude-skills** — Stack-agnostic Claude Skills covering the full website lifecycle: brand, design, content, SEO, dev, ops, growth, and research. Build, ship, audit, optimize. - Repository: https://github.com/rampstackco/claude-skills - Website: https://rampstack.co - Stars: 930 · Forks: 127 - Language: Python - License: MIT - Published: 2026-09-10 · Updated: 2026-09-10 · Language: en - Canonical page: https://hysenlabs.com/projects/rampstackco-claude-skills ## One library, not a link roundup, and that is the pitch The first claim the readme makes is a negative one: this is not a curated list of other people's skills. It is a single library where every skill follows the same structure and conventions, so they compose across a real project lifecycle. The numbers behind that claim are specific. There are 103 skills across 16 categories, and every one has a complete skill file plus at least one reference file, with 490 reference files in total across templates, checklists, decision matrices and worked examples. The categories are wide enough to describe a whole company rather than a marketing team: brand strategy and identity, design systems, content production, the SEO suite, product management, growth tooling for interactive web tools, paid media discipline, frontend development and accessibility, performance and QA, launch and incident operations, UX research, and a meta-skill that teaches you to write your own. Uniformity is the differentiator. Every skill uses the same section order, the same tone and the same authoring conventions, which the readme summarises as predictable in and predictable out. That is a real advantage when the same agent touches six different kinds of work in a week, and a real cost when you want one skill that does something different from the rest. ## Two marketplaces, because 103 skills is a choice The install section is the most useful operational detail in the readme, and it offers two different repositories. ``` /plugin marketplace add rampstackco/claude-skills # full catalog (103 skills) /plugin install rampstack-skills@rampstack # focused subsets /plugin install rampstack-starter@rampstack /plugin install rampstack-seo@rampstack /plugin install rampstack-pm@rampstack ``` So the first marketplace carries the entire catalog as one plugin, and the same marketplace also offers three curated subsets for people who do not want all of it. There is a second repository that lists only those curated subsets, with the same plugin names installable from it. The readme frames it as a lighter alternative for anyone who would rather not have the full catalogue listed at all. That split is a thoughtful answer to a problem most skill collections ignore. A marketplace listing is not free: each installed skill contributes roughly its name and description to the model's context before the model decides it needs anything. A hundred and three descriptions is a real cost paid on every conversation, whether or not the skills are used. So the choice is between the whole library and a curated subset, and the subset route does not mean giving up the ability to add more later. ## Skills load on description match, and portability is the reason to care Two mechanisms are worth understanding before you install anything. The first is how a skill gets used. A skill is a folder containing a skill file with instructions plus YAML metadata, and optional reference files of templates, checklists and worked examples. The model loads a skill automatically when a user request matches that skill's description. There is no command to invoke; the trigger is the description matching, which means the descriptions are the API and they have to be written precisely. The second is portability. The same skill works in the hosted assistant, in the command line coding tool, and through the API, and a skill you write once is usable in all three. That is the difference between a skill and a prompt: prompts get copied between tools by hand, skills move because the format is the same. The cost follows from the same design. Because loading is automatic and description-driven, a badly described skill is invisible and an over-broad description hijacks unrelated requests. A library that fixes one structure for 103 skills fixes the description problem too, which is another argument for the uniformity claim. And because everything loads on demand, installing the full catalogue is a budget decision rather than a capability decision. ## One skill assumes a paid tool, and the readme says so There is exactly one place where a named tool appears, and the readme flags it rather than burying it. The SEO audit suite is described as powered by a specific third-party search platform's MCP interface, and it is called out as the only exception to the stack-agnostic claim. Everything else works on whatever stack you have, including the generator-heavy frameworks, the hosted site builders and plain HTML. So the honest summary of the library's stack independence is: the method layer is portable, and the measurement layer is a vendor integration. That is a defensible split, because an audit tool that produces numbers is different in kind from a skill that teaches a framework of thinking, and pretending otherwise would be worse than saying so. The practical consequence is that you can adopt 102 skills with no account anywhere, and the audit suite is the part that needs a decision and possibly a budget. The readme also makes a durability argument for the rest of the catalogue: principles over tools, stable concepts over trending techniques, and references to durable specifications and interface guidelines rather than to content that ages with every algorithm update. That is a claim about maintenance burden, and it is the one that matters most for a library of written instructions. ## Six entry points, one per audience, and they reference each other The featured table is the best way to understand the library, because each of the six entries is described by what it decides rather than by what it produces. The creative direction skill is a four-axis brief covering tone, aesthetic, audience and sensory ambition, and its stated job is to give every downstream skill a coherent direction. Everything else inherits its coherence from there. The experiment design skill runs from hypothesis to decision, covering sample size, duration, segment analysis and the failure modes that produce wrong shipping calls. The feature launch playbook covers positioning, internal alignment, customer communications, enablement, rollout and monitoring. The pillar content architecture skill is about hub-and-cluster topical authority, including pillar selection, cluster planning, internal linking and refresh discipline. The landing page copy skill covers the hero-to-call-to-action flow. The funnel flow architecture skill covers cross-tool conversion flows matched to audience and funnel stage. The composition claim is the interesting part, and it is checkable from the readme alone. A creative brief skill points at a brand voice skill. An incident response skill points at monitoring and alerting. Each skill carries a section explaining when not to use it and which sibling covers the adjacent work. A library that tells you which skill not to use is making a different claim from one that lists capabilities, and it is the more useful claim. ## The README counts itself, which is either discipline or a warning There is a detail in this repository that rewards attention, because it is unusual. The skill count appears in at least four places in the readme: in the introduction paragraph, in the table of contents, in the summary of what you get, and around the featured skills table. Each of those spots is wrapped in marker comments that look like they exist to be rewritten by a script. There is a scripts directory in the repository, and the skill and workflow directories both have lock files alongside them. So the numbers in the readme are generated rather than typed, which means the failure mode where a catalogue grows and the introduction keeps claiming the old count should not happen silently. The reason this is worth mentioning is the gap between the release history and the current count. The first release, tagged v1.0.0, is described as an initial 59-skill catalogue, and it shipped on the same day as v1.1.0. The newest tag, v1.2.0, came nine days later and is described as catalogue growth among other things. The readme now claims 103 skills. So the catalogue went from 59 to 103 in under two weeks of tagging, and the last tag is five months old while the last commit to the branch is more recent. That is not a criticism so much as a measurement problem: the version you install and the count you read are tracked in different places, and only one of them is generated. ## Lock files, a starter kit, and a trust section The repository root tells you what kind of project this is. There are two lock files, one for skills and one for workflows, which means the catalogue and the automation around it are versioned separately. There is an authoring guide for skills and a starter kit document, so the project is set up to be extended rather than only consumed. There is a mapping document, a changelog, a security policy, a code of conduct, a contributing guide, a documentation directory, a tools directory, a built distribution directory committed alongside the sources, and a card image used for link previews. The authoring documentation matters more than it might appear. A library with 103 entries needs a defined structure or the uniformity claim decays within a year, and the uniform section order, tone and conventions the readme advertises are exactly what an authoring guide enforces. The trust and security section, which the table of contents lists alongside the contributing guide, is the one to read before installing. A skill is text that a model follows, so a catalogue of 103 of them is a meaningful amount of instruction to accept from a third party, and the fact that the project ships a security policy at all is a point in its favour. Everything is MIT licensed, and the readme notes the library is listed in a community awesome list, which is where most people will encounter it. ## Conclusion RampStack Claude Skills fits a team that wants one consistent method across brand, design, content and growth work rather than a folder of unrelated prompts, and the cross-references between skills are the part that makes a library better than a list. Skip it if you want tool-specific automation, since the explicit design goal is principles over tools. Before you install, pick the lighter marketplace if 103 skills is more than you want loaded, and read the Ahrefs MCP assumption before you rely on the SEO audit suite. ## FAQ ### How do I install claude skills Add the marketplace with /plugin marketplace add, then install a plugin from it: the full catalog as rampstack-skills, or a curated subset such as rampstack-starter, rampstack-seo or rampstack-pm. A second, lighter marketplace carries only the subsets. ### What do Claude skills actually do? Each is a folder with a SKILL.md holding instructions and YAML metadata plus optional reference files. Claude loads a skill automatically when a user request matches its description, and a skill written once works in the hosted assistant, the coding tool and the API. ### How many skills are in the RampStack claude-skills catalog? The readme claims 103 skills across 16 categories with 490 reference files, and those figures are inserted by a script between marker comments. The first release, v1.0.0, shipped as a 59-skill catalogue. ### Do the RampStack claude-skills work with any web stack? They are stack-agnostic by design, working across the major frameworks, site builders and plain HTML. The single exception is the SEO audit suite, which assumes an Ahrefs MCP interface. ### Can I write my own skills with this library? Yes. The repository ships an authoring guide and a starter kit, and the catalogue includes a meta-skill that teaches you to write your own, plus separate lock files for skills and workflows. ## Sources - [License: MIT](https://github.com/rampstackco/claude-skills/blob/main/LICENSE) - [Project website](https://rampstack.co) - [rampstackco/claude-skills on GitHub](https://github.com/rampstackco/claude-skills) - [README](https://github.com/rampstackco/claude-skills/blob/main/README.md) - [Releases](https://github.com/rampstackco/claude-skills/releases) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/rampstackco-claude-skills