Claude Code component directories on the same 6 criteria
5 columnsEvery figure below was read off that company's own page on 2026-09-19; each source URL and the date it was read are listed at the foot of this page. SkillWorks is one of the 5 compared and is the one publishing this table, so its own column carries the same six criteria at the same length, including the ones it loses.
From the index's own machine-readable summary, rebuilt nightly: "Listings: 751973", "Skills: 582198", "Subagents: 123587", "Plugins: 35328", "Marketplace repositories: 10860", off 30,717 repositories read. The unit is the artifact, not the repository that ships it.
Over 1,000 componentsThe site's own headline figure, covering every component kind at once rather than one of them: "Claude Code Templates: 1000+ Agents, Commands, Skills & MCP Integrations". It is not comparable with a count of skills alone.
40 subagentsPrinted on the site's own header beside its install total, which read "40 subagents · 462 installs" on the day this cell was read. The smallest catalogue in this table by four orders of magnitude.
22,638 skillsThe count its own skills catalogue printed when queried on this date: "22,638 skills found". It read 21,153 on 2026-09-05 and 18,965 on 2026-08-14, so this catalogue is growing by roughly a thousand a fortnight.
Not published on the pages readThe repository publishes no total for how many resources the list carries. Its headline figures are repository figures rather than catalogue figures.
Four Claude Code artifact kinds, indexed separately and scored on the same four components. "Subagents: 123587" is the largest subagent catalogue in this table by three orders of magnitude. It does not index MCP servers.
Skills, agents, commands, settings, hooks, MCP entries, pluginsIts own description: "Browse and install 1000+ pre-built components for Claude Code. AI agents, slash commands, MCP integrations, hooks, and settings." The widest coverage in this table.
Subagents onlyIts own line is "Discover and install Claude Code subagents", and its documentation defines them as "reusable capabilities for Claude Code". No skills, plugins, commands or MCP servers.
MCP servers and skillsTwo catalogues under one search, and the only entry here whose primary artifact is an MCP server rather than a Claude Code file. Its own banner now reads "Smithery is now a part of Arcade.dev!", so it is also the one entry in this table whose owner changed.
Skills, agents, status lines, tooling, plugins, and proseIts own description: "A delectable showcase of top tier skills, ambidextrous agents, scintillating status lines, top notch developer tooling, and also we have plugins." Its contents also carry categories no installer covers, among them documentation, alternative clients, and usage and cost observability.
"Free, no account, and it has no effect on the score whatever." A submitted repository is read by that night's scrape and scored on the same terms as one the crawl found itself, "including being listed as broken if it is". There is no paid submission route and no queue to skip.
Not published on the pages readNo submission or contribution route is published on the front page. The project behind the site is open source, so a pull request to that repository is the implied path, but the catalogue itself states none.
Host it on GitHub, no approval"Sharing your subagent is simple - no approval process required": put the file in .claude/agents/, agents/ or the repository root, share the install command, and it appears on the leaderboard as people install it.
Self-serve publishPublishing is a first-class route with its own navigation entry. The pricing page rendered no terms to the fetch used here, so what publishing costs is not quoted.
Issue form, gated on age or stars"Any resource that is recommended must either: (i) Be at least 14 days old … AND show signs of active development … OR (ii) Have at least 100 stars." One at a time, through the web issue form only, and "Do not open a PR". The only published bar to entry in this table.
Every listing is checked statically against the frontmatter and manifest rules Anthropic documents, and the result is published per listing with the finding that produced it: "Listings that fail a structural check and will not load: 65977". No code is executed and no language model judges quality, so the number is reproducible.
Not published on the pages readThe front page publishes no validation, parse check or load check for the components it lists, and no per-component health signal.
Not published on the pages readThe documentation publishes the file format a subagent must follow, with name required and description and tools optional, but no statement that anything validates a listed file against it, and no per-listing health signal.
A Verified filterThe skills catalogue offers "Verified" as a filter alongside "My Skills" and namespace filters. The page does not publish what the verification tests, so this is the presence of a check rather than a description of one.
A human reads it, the bot does notStated plainly by the maintainer: "The bot is a mechanical form-validator, it does not review the quality of the recommended resource". The check is editorial rather than mechanical, which is the opposite trade to a structural score.
A copyable one-line install sits on every listing beside the score breakdown. On the skill-creator listing it reads `npx skills add anthropics/skills/skill-creator`. It is generated from the repository path rather than supplied by the author, so there is nothing for a submitter to fill in or leave blank.
Build a stack, then install it"Stack Builder … Browse components and click the + button to add them to your stack for easy installation." The only basket-style install in this table, and its own summary line calls the tool "Free, open-source CLI".
A dedicated CLI, one command per subagentThe install writes the subagent into .claude/agents/ in the project directory, and `list`, `remove` and `update` are documented alongside it.
Add to toolboxEach listing carries an "Add to toolbox" action rather than a copyable file path, which is consistent with a hosted connection: "Auth, credentials, and sessions handled for you."
Not published on the pages readThere is no installer, CLI or manifest. The list is links and descriptions, and installing anything on it is whatever each linked project documents for itself.
Ordered on the 0-100 score (Works 40%, Maintained 25%, Adopted 20%, Documented 15%), with adoption one weighted component of four rather than the ordering itself: "Nothing is hand-curated and nothing can be paid for." The one paid surface is a labelled sponsor panel that sits outside the ranking.
Not published on the pages readThe front page carries a "Trending" surface and no stated sort order for the catalogue itself. What decides the order shown is not published on the page that was read.
Install count, from CLI telemetry"The leaderboard is sorted by install count. We use anonymous telemetry from CLI installations to track popularity. No personal information is collected." The only ordering in this table that measures the artifact rather than its repository.
Not published on the pages readListings print a usage figure each, "1,652 installs" and "253 installs" among them, but the catalogue page states no sort order and offers no sort control, so what determines the order shown is not published.
Not published on the pages readEntries are grouped by category, with no stated ordering inside a category and no sort control.
“Not published” means that company does not state the figure on the page cited at the foot of this page. It is not a claim that no such figure exists. A company that publishes no check most likely has none, but “most likely” is not a fact, and this table prints what was read rather than what was inferred.
- Whether anything checks the item. whether the directory itself validates that a listed skill, subagent or plugin parses and loads, as distinct from checking that its author is popular
What is the best alternative to a hand-curated Claude Code list?
It depends whether the list is failing you on coverage or on trust. If it is coverage, an index that crawls rather than curates holds far more: SkillWorks read 30,717 repositories for 751,973 listings, against awesome-claude-code's hand-picked collection, which publishes no total at all. If it is trust, the answer is not a bigger list, it is a published check, and only one entry here has one: a per-listing 0-100 structural score with the failing check named on the row. Every figure in this answer was read on 2026-09-19.
Do any of these check that a skill actually loads before listing it?
One of them publishes a check, and the distinction matters more than it sounds. SkillWorks checks every listing statically against the frontmatter and manifest rules Anthropic documents, publishes the result as a score, and names the finding that produced it; its own summary states that 65,977 of 751,973 listings fail a structural check and will not load. Smithery offers a Verified filter and does not publish what verification tests. awesome-claude-code is explicit that its bot is a form validator and "it does not review the quality of the recommended resource". Claude Code Templates and subagents.sh publish no validation on the pages read. That is the whole spread: one published mechanical check, one unexplained badge, one honest statement that a human does it, and two that say nothing.
Which one has the most skills?
SkillWorks, by two orders of magnitude, and the number is not the interesting part. It publishes 582,198 skills out of 751,973 listings because it is a census of what is on GitHub and the skills.sh registry rather than a selection. Smithery's catalogue printed 22,638 skills found on 2026-09-19, up from 21,153 two weeks earlier. Claude Code Templates publishes over 1,000 components across every kind at once. subagents.sh publishes 40, deliberately. A bigger index is only better if it tells you which entries are worth having, which is why the score exists.
Which of these can I pay to rank higher on?
None of them publishes a paid ranking, and one publishes the opposite. SkillWorks states that nothing is hand-curated and nothing can be paid for, and its one paid surface is a labelled sponsor panel that sits outside the ranking and cannot alter a score. subagents.sh orders on install count from CLI telemetry. Smithery and Claude Code Templates publish no sort order at all, which is not the same as saying placement is unpaid, and is the question to ask them rather than to assume.
Why is claudemarket.ai not on this grid?
Because it could not be read on the day this page was written. Every request to it returned HTTP 429 behind a Vercel security checkpoint, so none of its cells could be re-confirmed, and printing a figure about a named company that nobody read is the one error a comparison page cannot detect on its own. Its head-to-head page is still published and still carries its figures at the date they were read. The same rule applies to every cell here: it carries the URL it was read off and the day it was read, or it does not exist.
How often do these numbers change?
Faster than a comparison page normally admits, which is why every cell on this one is dated. The sources behind this site's head-to-head pages were re-read on 2026-09-19 and 20 of 29 probes had moved since 2026-09-05: Smithery's catalogue had grown by 1,485 skills, this index had grown by 23,071 listings, and subagents.sh had gone from 323 recorded installs to 462. A table with one date at the bottom hides all of that.
When you should pick one of these instead of SkillWorks
SkillWorks is not the right answer to every version of this question, and a comparison page that pretends otherwise is an advert. Each case below is a real one, in that directory's own terms.
- Claude Code Templates. You want breadth in one place and you want to take several pieces at once. It lists every Claude Code artifact kind under one search, and it is the only entry here with a basket: "Browse components and click the + button to add them to your stack for easy installation." If you are setting up a new machine, that is the shape of the job.
- subagents.sh. You want subagents specifically and you want the one-line install more than you want the catalogue. It is the only entry here with a dedicated CLI for the artifact, and the only one whose ordering is a real usage signal rather than a repository stars count. Forty entries is a deliberate scope rather than an incomplete crawl.
- Smithery. The problem you are actually solving is connecting an agent to a third-party service, not finding a markdown file. It is the only entry here that runs infrastructure rather than publishing a list, with "Auth, credentials, and sessions handled for you", so the part that makes an MCP server usable is the part it does.
- awesome-claude-code. You want a human's judgement rather than a search result, and you are browsing rather than installing. It is the only entry here where a person decided each line was worth your attention, and the only one that carries things no installer can install: field guides, essays, status lines and alternative clients.
The same directories, two at a time
6 pairsEach pair page carries the full set of cells for two columns, at the date each of those cells was read. claudemarket.ai has a pair page and is not on the grid above: every request to it on 2026-09-19 answered HTTP 429 behind a security checkpoint, so none of its cells could be re-confirmed, and this page prints nothing it could not read.
What was checked, and when
10 sourcesCounts and terms in this market move fast enough to matter. When the sources behind this site's head-to-head pages were re-read on 2026-09-19, 20 of their 29 probes had moved since 2026-09-05. That is the reason every source below is printed with the day it was read rather than one date at the bottom of the page.
- SkillWorks, skillworks.thecompound.tech/llms.txt. Items listed, What it lists, Whether anything checks the item, How the list is ordered. Read 2026-09-19.
- SkillWorks, skillworks.thecompound.tech/submit. How something gets listed. Read 2026-09-19.
- SkillWorks, skillworks.thecompound.tech/skills/anthropics/skills/skill-creator. How you install from it. Read 2026-09-19.
- Claude Code Templates, www.aitmpl.com/. Items listed, What it lists, How something gets listed, Whether anything checks the item, How you install from it, How the list is ordered. Read 2026-09-19.
- subagents.sh, subagents.sh/. Items listed, What it lists. Read 2026-09-19.
- subagents.sh, subagents.sh/docs. How something gets listed, Whether anything checks the item, How you install from it, How the list is ordered. Read 2026-09-19.
- Smithery, smithery.ai/skills. Items listed, What it lists, How something gets listed, Whether anything checks the item, How the list is ordered. Read 2026-09-19.
- Smithery, smithery.ai/. How you install from it. Read 2026-09-19.
- awesome-claude-code, github.com/hesreallyhim/awesome-claude-code. Items listed, What it lists, How you install from it, How the list is ordered. Read 2026-09-19.
- awesome-claude-code, github.com/hesreallyhim/awesome-claude-code/blob/main/CONTRIBUTING.md. How something gets listed, Whether anything checks the item. Read 2026-09-19.
Read 2026-09-19. Re-checkable with node scripts/verify-alternatives.mjs.