ClaudSkills vs Awesome Claude Code: An Honest Head-to-Head

Published 5 June 2026 · 12 min read · By the ClaudSkills team

TL;DR — the one-paragraph verdict

Awesome Claude Code is a curated list. 226 hand-picked entries in a single CSV, vetted by one maintainer, distributed under a non-commercial no-derivatives Creative Commons license. It’s the right destination if you want a fast 5-minute skim of resources that someone with editorial taste has personally vouched for. It is currently in the middle of a stated reorganization — the README’s table of contents reads “I. TODO” as of June 2026, with a public note from the maintainer that “the old ways have come and gone.”

ClaudSkills is a comprehensive catalog. 122,000+ Claude Code skills mined daily from 26 community sources, scored with a multi-signal quality blend, distributed under a permissive CC BY 4.0 license with a public API, a CSV / NDJSON / Parquet / Atom dataset on GitHub and Hugging Face, per-skill HTML pages, author profile pages, a desktop installer, and a daily Skill-of-the-Day. It’s the right destination if you want machine-readable data, search and filter across the whole population, or a way to install something with one click.

The two aren’t really substitutes. You can use both, and many practitioners do.

By the numbers (verified 2026-06-05)

If you want to verify any of these yourself: the Awesome Claude Code numbers come from the live repo at github.com/hesreallyhim/awesome-claude-code and THE_RESOURCES_TABLE.csv at its root. The ClaudSkills numbers come from /stats on this site, with the dataset itself at github.com/claudskills/catalog-public.

Curation: editorial vs automated

This is the defining structural difference between the two catalogs, and it shapes everything downstream.

Awesome Claude Code: one person, one taste

Awesome Claude Code is the work of one maintainer, hesreallyhim, in the classic GitHub “awesome list” tradition. Resources are added by community pull requests; the maintainer reviews each one and merges if it clears the editorial bar. The bar isn’t explicitly documented, but the implied standard, going by what’s in the CSV, is “something the maintainer would personally recommend to a friend.” That kind of human-judgement filter has real value — you get a small, high-signal set that’s been vetted by a person with taste — and it’s what 45,000+ people are starring the repo for.

The cost is that the model doesn’t scale, and the maintainer has been open about this. The current state of the README (a transitional “I. TODO” with a public note saying “the old ways have come and gone”) is a maintainer admitting in public that the single-CSV, single-taste approach has hit a ceiling. We don’t know yet what shape the next phase will take; the maintainer hasn’t published a roadmap. But the current friction is real.

ClaudSkills: automated admission, content-derived quality

ClaudSkills works the opposite way. A daily miner crawls 26 sources — GitHub code search, GitLab, Codeberg, Hugging Face, gists, awesome-lists, GitHub Topics, Reddit, Hacker News, Bluesky, Mastodon, dev.to, YouTube, marketplaces, RSS feeds, and direct submissions — pulls every SKILL.md it can find, and applies a deterministic admission rule (currently structural quality score ≥ 65, raised from 50 in May 2026). Anything above the bar is admitted; anything below it isn’t. There is no individual taste in the loop. The methodology is documented, the scoring code is in the open dataset, and the threshold can be re-tuned and the catalog re-scored from public data at any time.

The cost of this approach is that the bar can’t do everything a human curator can. A skill that’s exquisitely well-written but doesn’t hit the structural signals (anti-trigger discipline, metadata depth, code-block density) will pass — but a skill that’s mediocre on every axis but happens to be by a famous author won’t get any extra credit. That’s the trade. The signal we’re willing to commit to is content-derived. Anything popularity-based (stars, install counts) was explicitly removed from the score after a design review in May 2026, because popularity tracks promotion, not quality.

Neither model is “correct.” They’re different bets on what users want from a catalog. Awesome Claude Code is betting that 200 vetted entries are more useful than 122,000+ Claude Code skills ranked by quality. ClaudSkills is betting the other way.

Coverage and scope

Both catalogs cover roughly the same kinds of things — skills, hooks, slash commands, agent orchestrators, plugins — but the depth varies wildly.

What Awesome Claude Code captures well: high-profile, well-marketed, frequently-shared resources. The kind of thing that ends up in a Twitter thread, an HN post, or a newsletter roundup. If someone is loud enough on the timeline to get a PR opened against an awesome list, they end up in Awesome Claude Code. The CSV has 226 such entries as of June 2026.

What ClaudSkills captures well: everything else. The miner is built specifically to find the long tail — skills that live in personal dotfiles repos with three stars, gists nobody links to, Mastodon posts, Telegram channels, dev.to articles, marketplace extensions whose authors never wrote a tweet about them. There are many useful skills in the ClaudSkills catalog that you would never find on Twitter, simply because their authors aren’t on Twitter.

This shows up most starkly in two places. Author distribution: Awesome Claude Code is dominated by a small number of prolific contributors (look at the “Author Name” column). ClaudSkills has 2,355 admitted authors with at least two skills each. Geographic and linguistic distribution: the ClaudSkills miner pulls from non-English sources too — gists with Polish, Spanish, Japanese, Chinese descriptions clear the bar if their SKILL.md is structurally sound. Awesome Claude Code skews English by virtue of where its PRs come from.

If you’re looking for a specific niche — “a skill that does X for Y database in Z language” — the chance of it existing somewhere in ClaudSkills is much higher than the chance of it being one of the 226 awesome entries. Conversely, if you ask “what are the most-shared Claude Code resources on the timeline this month?”, Awesome Claude Code is a tighter signal.

License: CC BY 4.0 vs CC BY-NC-ND 4.0

This one matters more than it looks, especially for anyone planning to build on top of either catalog.

Awesome Claude Code: CC BY-NC-ND 4.0

Awesome Claude Code is licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International. Three constraints:

• BY — attribution required. Fine; nobody objects to that.
• NC — non-commercial only. You cannot use the data in a product or service you sell or monetize, including indirectly through ads or SaaS.
• ND — no derivatives. You cannot publish a corrected, filtered, transformed, or annotated version of the data. You can’t fork the CSV and remove rows you disagree with and republish the fork — that’s a derivative.

The practical effect is that Awesome Claude Code is for individual human reading and personal use, not for downstream products. A startup can’t train an LLM on it, a SaaS can’t embed it, a competing catalog can’t reuse its categorizations. That’s the maintainer’s choice, and there are coherent reasons to make it — it prevents large companies from absorbing the curation work without giving anything back. But it’s a strong constraint, and it’s often missed by people who assume “Creative Commons” means “use it for anything.”

ClaudSkills: CC BY 4.0

ClaudSkills’ open dataset at github.com/claudskills/catalog-public is CC BY 4.0. One constraint:

• BY — attribution required (just credit ClaudSkills with a link). That’s it.

Commercial use is fine. Derivatives are fine. Training models on it is fine. Republishing a filtered or annotated version is fine. The reason we picked the permissive license is that we want the data to be reused — the value of a catalog grows with the number of downstream products that build on it. The constraint is just that the credit chain stays attached.

If you’re a developer planning to build something on top of community Claude Code data — whether it’s an internal tool, a startup, an LLM-evaluation harness, or a research paper — this licensing difference is the first thing to know.

Discovery and install flow

How do you actually find and install something?

Awesome Claude Code

The discovery surface is currently in transition. Until recently, you scrolled the README’s table of contents grouped by category and followed links to upstream repos. The TOC was removed in the current README phase, with a note that a new organization is being prepared. In the meantime, the realistic way to browse is to open THE_RESOURCES_TABLE.csv in your editor of choice and grep or filter.

Install flow: each row in the CSV points at the upstream repo. You click through, read the upstream repo’s install instructions, and follow them. The instructions vary by repo — some are git clone into ~/.claude/skills/, some are npm install, some are manual file copies. Awesome Claude Code itself does not standardize the install path; that’s the upstream author’s decision.

ClaudSkills

The discovery surface is a search-and-filter web app at claudskills.com. You can browse by category, tag, author, use case, or free-text search. Each skill has a dedicated page at /skills/<slug>/ with the description, body preview, install command, related skills, attribution, license, and links back to the upstream source. The Skill of the Day archive surfaces one editorially-eligible skill per day with a stable URL that’s safe to share on social.

Install flow has two paths. The free path: copy the install command from the skill page, paste into your terminal, done. The Pro path: install the desktop Electron app, sign in with your subscriber email, click “Install” on any skill card. The desktop app handles the file fetch, the directory placement under ~/.claude/skills/<slug>/, and the version tracking. Either way the skill ends up at the same place on disk — the Pro app is a convenience, not a gate on the content itself.

Quality signals: implicit endorsement vs explicit score

The most user-visible difference between the two catalogs is how they answer the question “is this skill any good?”.

Awesome Claude Code: inclusion as endorsement

Awesome Claude Code’s quality signal is binary: a skill is in the list or it isn’t. There is no ranking, no rating, no star count column. If the maintainer accepted the PR, the skill is endorsed; if they didn’t (or nobody opened one), it isn’t. The endorsement is human and editorial; it carries the weight that 45,000+ stars represent. That’s a real signal.

The trade is that you can’t tell how good a listed skill is relative to others on the list, or what dimension it’s strong on. The CSV does have categorization (skills, hooks, slash commands, agents, applications), but within a category, everything is flat.

ClaudSkills: multi-signal quality score (Pro)

ClaudSkills surfaces a quality score on every admitted skill, but only for Pro subscribers ($9/month, $79/year, or $149 one-time lifetime). The score is computed server-side from two content-derived signals:

• Structural heuristic (80%) — description depth, vocabulary diversity, body length and structure, anti-trigger discipline (the “When NOT to use” section that good skills have), pricing/quota disclosure, custom frontmatter richness.
• Metadata depth (20%) — how many meaningful frontmatter fields beyond name and description the skill declares (capped at 10 distinct fields to prevent gaming).

The score is rescaled into a 50–100 range so that “50” means “barely passed admission” and “100” means “best in catalog.” The current distribution is roughly 0.5% at 50–59, 53% at 60–69, 34% at 70–79, 12% at 80–89, and 0.4% at 90+. The reasoning behind the formula is documented at the skill quality rubric.

The Pro paywall on the score is a deliberate revenue model choice: end-users pay for the pre-computed evaluation layer; skill authors never pay anything, ever, under any circumstance. We explicitly reject the “sponsored placement” or “pay-to-rank” model because the moment ranking can be bought, the score collapses into a paid-promotion proxy.

Data access for agents and LLMs

If you’re building an agent, an LLM retrieval pipeline, an internal tool, or a research paper — not just browsing — the data-access story matters more than the browse UI.

Awesome Claude Code

The data is a single CSV at the root of the repo, fetched via raw.githubusercontent.com/hesreallyhim/awesome-claude-code/main/THE_RESOURCES_TABLE.csv. It has 226 rows and 20 columns including ID, display name, category, primary link, author, date added, last modified, license, and description. There is no API, no JSON endpoint, no NDJSON, no Parquet. You parse the CSV yourself.

The license (CC BY-NC-ND) means you cannot embed the CSV in a commercial product, derive a transformed dataset from it, or republish it in your own format. Practical implication: a startup that wants to use the data needs to either ask the maintainer for a different license arrangement, or treat the data as “read it ourselves, build our own structured representation, attribute the inspiration without copying the data.” That’s a real legal exposure if not handled carefully.

ClaudSkills

Multiple equivalent surfaces, refreshed daily, all under CC BY 4.0:

• JSON at claudskills.com/data/skills.json (42 MB, public)
• Public API at claudskills.com/api/v1/ — 8 endpoints, OpenAPI 3.1 spec at /api/v1/openapi.json, no auth required for read access
• GitHub catalog-public repo at github.com/claudskills/catalog-public — 6 artifacts (JSON, NDJSON, CSV, Parquet, feed.atom, README), refreshed daily at 03:00 UTC, CITATION.cff included
• Hugging Face dataset mirror, refreshed in the same daily run
• llms.txt at claudskills.com/llms.txt — ~146 KB structured summary specifically for LLM crawlers, with the top 500 indexed skills
• Sitemap XML at claudskills.com/sitemap.xml — sitemap index + ~22k indexed skill URLs

The dataset is designed for downstream reuse. If you’re training a model, retrieving as RAG context, or building a recommendation engine, you can pull the Parquet file once a day and have a complete snapshot. The Atom feed is suitable for “subscribe to new skills” consumers. The OpenAPI spec lets you generate client libraries automatically.

When to choose each

Choose Awesome Claude Code if…

• You want a fast, taste-vetted skim of ~200 resources without filter or rank fatigue.
• You trust one maintainer’s editorial judgement and want a small high-signal set.
• You’re writing a personal “here’s what I use” reference and want to pull from a list that’s already been pre-curated by someone with taste.
• You’re browsing for personal/individual use and don’t need commercial-license clearance.
• You’re happy to follow per-repo install instructions for each resource.

Choose ClaudSkills if…

• You’re searching for a skill for a specific task and need filter / search / category / tag / author navigation.
• You need to install many skills and want a one-click flow (Pro), or want to copy-paste install commands from the page (free).
• You’re building something that consumes the catalog as data — an LLM, agent, retrieval pipeline, recommendation engine, internal tool.
• You need a commercial-friendly license (CC BY 4.0) on the underlying dataset.
• You want quality scoring across the full population to short-list 122,000+ Claude Code skills down to the few that matter for your use case.
• You’re publishing your own skill and want it discoverable by the long-tail audience that doesn’t follow Twitter.
• You want per-author profile pages, per-category pages, daily Skill-of-the-Day, a public API, and machine-readable bulk dataset.

Notice that none of these “choose Awesome Claude Code if” bullets are wrong for ClaudSkills, and vice versa. The two catalogs solve overlapping but distinct problems. For most working developers the honest answer is “use both, for different things.”

Can they coexist? Yes, here’s how

A practical workflow that uses both:

1. Start at Awesome Claude Code when you want a Saturday-morning skim of what’s good in the ecosystem. 226 entries is a quantity a human can actually read top to bottom; you’ll come away with a mental model of what’s out there.
2. Come to ClaudSkills when you have a specific need that didn’t turn up in the awesome list, or when you want to install something cleanly without working through each upstream repo’s instructions. Search, filter, install, done.
3. Use ClaudSkills’ dataset when you’re building anything — agent, tool, paper, internal product — that needs the data programmatically with a commercial-friendly license.
4. Submit to both when you publish a new skill. There’s no conflict: ClaudSkills will mine your repo automatically if it has a SKILL.md, and you can open a PR to Awesome Claude Code if you think it clears their editorial bar. Different distribution channels for the same artifact.

Awesome Claude Code is the Coffee Table Book of the Claude Code ecosystem — you read it and feel like you’ve seen the highlights. ClaudSkills is the Library Catalog — you go to it when you need to find a specific thing.

If you’re the maintainer of Awesome Claude Code reading this and you spot anything we’ve gotten wrong, email [email protected]. We’ll fix the page within 24 hours.

Found a bug or want a topic covered? Email [email protected] or open an issue via GitHub.