rampstackco/claude-skills: A 103-Skill Claude Catalog for the Web Lifecycle

The Problem: Skill Sprawl Across a Web Project

Most teams that use Claude for web work end up with a folder of prompts that grew organically. One person has a landing page prompt, another has a technical SEO checklist, a third has something for launch comms. They overlap, they contradict each other, and nobody knows which one is current. The README frames this library as the answer: not a curated list of other people's skills, but a single opinionated library where every skill follows the same structure and conventions. The claim is that consistency is what makes the skills compose. That is a real problem, and the framing is honest about the scope. The catalog targets people who touch the whole lifecycle of a brand and a website, which in practice means small teams, agencies, and solo operators who move between a brief, a build, and a launch in the same week. It is explicitly stack-agnostic: Next.js, WordPress, Shopify, Webflow, plain HTML. If your work is one narrow slice of that lifecycle, the catalog is larger than you need. The README itself provides focused subsets for that case, which is a more useful admission than a pitch would be.

How a Skill Loads and What It Contains

Each skill is a folder with a SKILL.md file holding instructions plus YAML metadata, and optional reference files such as templates, checklists, decision matrices and worked examples. Claude loads a skill automatically when a user request matches the skill's description. The README states that skills load on demand, contributing roughly their name and description until Claude needs them, which matters because 103 skills would otherwise flood the context window. The library reports 490 reference files across the catalog. Composition is the second mechanism: skills reference each other by name. The README gives two examples, creative-brief pointing to brand-voice, and incident-response pointing to monitoring-and-alerting. Each skill also carries a When NOT to use section that names the sibling skill for adjacent work. That is the part worth paying attention to. A library that tells you where a skill stops is more useful than one that claims universal fit, and it is the mechanism that keeps a 103-skill catalog from becoming a pile. The featured table lists six entry points, one per audience track: creative-direction, experiment-design, feature-launch-playbook, pillar-content-architecture and landing-page-copy among them.

Installing the Catalog in Claude Code

The install path is the Claude Code plugin marketplace, not a pip install, despite the repository's primary language being Python. Add the marketplace, then install a plugin:

/plugin marketplace add rampstackco/claude-skills /plugin install rampstack-skills@rampstack

The full catalog is 103 skills. The README also documents three subsets, rampstack-starter@rampstack, rampstack-seo@rampstack and rampstack-pm@rampstack, installed the same way. There is a second marketplace, rampstackco/plugins, described as a lighter listing that carries only the curated subsets and not the full catalog. If you want the starter set and nothing else, that is the shorter path:

/plugin marketplace add rampstackco/plugins /plugin install rampstack-starter@rampstack

The README does not print a per-skill configuration file or environment variable, so there is nothing to tune at install time. Skills are portable across Claude.ai, Claude Code and the Anthropic API according to the documentation, but the install instructions given here are Claude Code specific.

The Ahrefs Dependency Inside the SEO Suite

The library describes itself as stack-agnostic with one named-tool exception: the SEO audit suite, which assumes the Ahrefs MCP. That is a meaningful boundary. The rest of the catalog is built on principles and durable specs (W3C, WHATWG, Schema.org, MDN, NN/g, WCAG), which age slowly and do not require a subscription. The audit suite is different. It depends on an external data source and a paid product, so the audit skills are only as usable as your Ahrefs access. If you install rampstack-seo@rampstack without that MCP configured, the foundation skills should still work, but the audit half has nothing to call. The README lists a Recommended MCPs section but the supplied excerpt does not show its contents, so the exact setup steps for wiring the Ahrefs MCP are not verifiable from this material. Treat that as the first thing to check before committing to the SEO subset.

Where a Broad Stack-Agnostic Catalog Fails

Stack-agnostic is a trade-off, not a free win. A skill that works on Next.js, WordPress, Shopify, Webflow and plain HTML cannot contain framework-specific guidance, because that guidance would be wrong on four of the five. So the value sits at the level of process, vocabulary and output format: how to write a four-axis creative brief, how to structure a hub-and-cluster content plan, how to sequence a feature launch. If what you actually need is a Next.js image optimization recipe or a Shopify Liquid pattern, this catalog will not supply it, and a framework-specific skill set would serve you better. The second failure mode is breadth without triage. 103 skills with 490 reference files is a lot of surface area to keep coherent, and the release history shows the catalog has already been revised twice in quick succession: v1.0.0 shipped 59 skills on 2026-04-29, v1.1.0 followed the same day with a quality and community standards pass, and v1.2.0 in May 2026 added catalog growth, audience tracks and a visual refresh. That pace is normal for a young project but it means skill names and boundaries can move. Pin a version if you build internal workflows on top of specific skills.

Against a Hand-Rolled Prompt Library

The obvious alternative is not another skills collection. It is the prompt folder your team already has. The difference is structural. A hand-rolled prompt is usually a single block of text with no metadata, no declared scope and no stated failure conditions. A SKILL.md carries YAML metadata that lets Claude decide when to load it, plus reference files that a prompt cannot hold, plus the When NOT to use section that routes you elsewhere. That last part is the real gap. A prompt library has no mechanism for telling you it is the wrong tool; it just answers. The cost of the switch is the install flow and the Claude Code dependency. If your team works in Claude.ai or the API rather than Claude Code, the plugin marketplace commands in this README do not apply, though the README states skills are portable across all three surfaces. Verify that portability on your surface before planning around it.

Maintenance, Licence and What to Verify

The repository is MIT licensed, which permits commercial use and modification; that is a factual note, not legal advice. The last push recorded is 2026-09-07, and the three releases span April to May 2026, so the project is active but young. Because skills are plain markdown folders, forking and editing them locally is straightforward, and a fork insulates you from upstream renames. The maintenance cost is on you either way: a 103-skill catalog with 490 reference files is a lot of text to keep aligned with your own brand voice and output formats, and drift between the library's conventions and yours will show up as inconsistent deliverables. Start with the focused subsets rather than the full catalog, and read skills/creative-direction/SKILL.md plus one SEO audit skill before you install anything. Those two files tell you whether the library's structure matches how your team already works.

Editorial conclusion

Adopt it if you already work inside Claude Code and want a shared vocabulary for briefs, audits and launch checklists across several projects; skip it if your work is narrow enough that one or two skills would cover it, or if you cannot accept the Ahrefs MCP dependency in the SEO audit suite. Before installing the full catalog, open skills/creative-direction/SKILL.md and one SEO audit skill, and confirm the output format matches what your team already produces.

Official sources

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