Tons of Skills: A 442-Plugin Marketplace for Claude Code, With a Model-Agnostic Pitch

What This Marketplace Actually Contains

The repository is a catalog and distribution hub for agent skills. The README's scale table gives the most reliable numbers: 442 catalog plugins, 3,067 distinct marketplace-visible skills, and 347 agent definitions. Those numbers come from specific commands, so they can be reproduced. The description in the repository metadata says 471 plugins and 3,069 skills, but the README's own table, which is generated by a script, reports 442 and 3,067. That discrepancy is worth noting: the project itself refuses to give a single unlabeled count, because different cohorts produce different numbers. The plugin categories number 19, which is a manageable taxonomy for a catalog this size. The intended user is someone who wants to extend Claude Code with prebuilt skills, agents, and plugin bundles without writing them from scratch.

The Model-Agnostic Claim and Its Limits

The README opens with a strong statement: the platform is 'model-agnostic' and the canonical layer is 'harness-free by construction.' But it immediately qualifies that Claude Code is the only 'verified-native harness.' Other harnesses are described as 'engineering candidates' until their native-path integration is verified. This is an honest limitation. The project does not claim that its skills work verbatim in other tools. It says that source research alone is not presented as public support. For an engineer evaluating this for a non-Claude tool, that means the skills may or may not transfer. The design choice to keep the canonical layer harness-free is logical, but the verification gap is real. If you are using Cursor or a custom agent, you should treat every skill as unverified until you test it yourself.

Installation: The Frozen Slug and the ccpi CLI

There are two installation paths. Inside Claude Code, a single slash command adds the whole marketplace: '/plugin marketplace add jeremylongshore/claude-code-plugins'. The README notes that this slug is a 'FROZEN compatibility contract,' hardcoded in the CLI and many downstream READMEs. That is a maintenance red flag: the project is aware that renaming the repo would break things, so the slug is load-bearing. The second path is the npm CLI: 'pnpm add -g @intentsolutionsio/ccpi' followed by 'ccpi install devops-automation-pack'. The CLI is a package manager for plugins. The README gives no further commands, so you will need to run 'ccpi --help' after installing to see the full command surface. The npm namespace is '@intentsolutionsio', which is separate from the GitHub repo owner. That split means package versions do not match the marketplace display version, and the README explicitly says they are 'intentionally not expected to equal' each other.

Scale and the Version-Surface Checker

The project has a mechanism called the 'version-surface checker' that governs display surfaces without rewriting package semver. This is a response to a real problem: when a marketplace aggregates hundreds of packages, the display version can drift from the actual package versions. The checker keeps the README badge and website in sync with something, but not with package versions. The README says the release badge is the marketplace's display version, and npm packages retain their own versions. This is a sensible separation, but it creates confusion for users who expect the marketplace version to tell them anything about the plugins they install. The npm stats table shows 396 published packages and daily download counts, but those are downloads, not adoption rates. The top packages, like '@intentsolutionsio/openrouter-pack' and '@intentsolutionsio/groq-pack', are integration packs for AI providers, which suggests the catalog leans toward connecting Claude Code to external services.

A Genuine Limitation: Curation and Trust

The biggest limitation is curation. With 442 plugins and 3,067 skills, the catalog is too large for any single maintainer to vet deeply. The README highlights a 'Killer Skill of the Week' and gives grades, but that is a spotlight, not a review process. The scale table itself is a defensive move: the project knows that unlabeled counts lead to contradictory claims. The 'no-ai-slop' skill, which is the spotlighted pick, explicitly refuses to score drafts or guess if AI wrote them, because 'AI detectors guess.' That philosophy is a good guide for evaluating this marketplace: you cannot trust a single score or count. You must inspect the specific plugin's source, its recent commits, and its npm package version. The README does not mention a security review process for third-party skills. That is a gap. A skill is code that runs inside your agent, so a malicious or buggy skill could have real consequences.

The Alternative: Building Skills Yourself or Using a Smaller Store

The obvious alternative is to write your own skills and agents for Claude Code, using the official Claude Code documentation. That approach gives you full control over the code and avoids dependency on a third-party marketplace. It costs time, but for a team with specific workflows, it may be cheaper than auditing 442 plugins to find the one that works. Another alternative is a smaller, curated marketplace like the one from Anthropic or a vendor-specific store. The difference is in the trade-off: Tons of Skills offers breadth and a unified install command, while a smaller store offers trust and consistency. If you only need five skills, you do not need a marketplace with 3,000. The ccpi CLI is convenient, but convenience is not the same as reliability.

Maintenance and License

The repository is MIT-licensed, which means you can fork it, modify it, and redistribute it with minimal legal friction. That is a positive. The maintenance picture is mixed. The last push was July 23, 2026, and there are recent releases for packages like '@intentsolutionsio/klaviyo-pack' and '@intentsolutionsio/intercom-pack', so the project is active. But the README reveals a heavy reliance on scripts: a daily cron updates npm stats, a script generates the table of contents, and another renders the spotlight. That automation is good for consistency, but it also means the README is partly generated and may not reflect manual curation. The version-surface checker is a sign that the project is aware of its own complexity. The upgrade cost for a user is low: you re-run the install command to get new plugins. The upgrade cost for a maintainer is high, because the frozen slug and the checker constrain any restructuring.

Editorial conclusion

Adopt Tons of Skills if you are a Claude Code user who wants a single command to pull in hundreds of plugins and skills, and if you accept that the marketplace's model-agnostic claim is not yet backed by native verification for any harness other than Claude Code. Do not adopt it if you need a stable, vetted catalog: the README's own scale table admits that counts vary by cohort, and the marketplace relies on a single maintainer plus community submissions. Before using it, verify that the specific plugin you need has a recent release and an active maintainer, and check the npm package version separately from the marketplace display version, since they are intentionally decoupled. The project is MIT-licensed, so license risk is low, but the real cost is curation: you are trading a curated plugin store for a large, uneven corpus.

Official sources

1. Official documentation
2. Official README
3. Project repository
4. Release notes