What 69,000 Claude Code Skills reveal about how developers use Claude

Published 23 May 2026 · 8 min read · Based on the live ClaudSkills catalog at /stats/

The ClaudSkills catalog crossed 69,000 community-contributed SKILL.md files this week. That's a useful round-number milestone, but the more interesting question is: what does the distribution actually look like? Where are people building, what are they building, and which categories quietly outperform the headline ones? Below is a clean read of the data — no spin, no marketing arithmetic.

1. The headline distribution

69,016 skills, ten categories, 118 sub-topics, 5,053 distinct authors. The category split is wider than the round-number summary suggests:

More than half the catalog ends up in General. That's not a quality complaint — many genuinely cross-cutting skills (template renderers, file processors, generic agents) don't fit a single vertical and are correctly classified as general-purpose. But it does mean the headline "69k skills" figure underrepresents how concentrated the catalog is. The vertical-specific buckets — Sales, Growth, Advertising — together account for less than 5% of the volume.

Engineering at ~20% is consistent with what you'd expect from an AI-coding-agent ecosystem. The interesting outlier is Security at 5.5%: a thousand skills here have a measurable economic ROI per use, which usually predicts above-average authoring discipline. We'll see that play out in section 3.

2. Top tags — what the catalog is actually for

Categories tell you the rough vertical. Tags tell you the actual intent. We use a flat tagging vocabulary (lang:python, type:integration, ai:claude, etc.) and the distribution is striking:

The top three intents — integration, review, audit — account for 17,969 skills together, more than a quarter of the catalog. That's revealing. Developers aren't (mostly) using Claude as a general-purpose coding assistant when they sit down to write a skill. They're using it as a structured glue layer and a structured checker. Both of those are workflows where a deterministic, repeatable Claude invocation is a better fit than ad-hoc prompts in a chat window.

The language breakdown also tells you something. Python (1,462) leads TypeScript (944) by ~55%, and TypeScript leads JavaScript (774) by ~22%. That ordering matches what you'd expect from AI-tooling ecosystems but inverts the broader open-source repo distribution. Make of that what you will — our take is that AI-first tooling slants more toward Python because the training data, the prompt-engineering ecosystem, and the popular ML libraries all do too.

3. Quality rate by category — the counterintuitive finding

Volume is the obvious metric. Per-skill quality is the metric that actually matters when you're trying to discover something useful. ClaudSkills computes a multi-signal Quality Score for every admitted skill — anti-trigger discipline, structural depth, frontmatter completeness, length-vs-signal ratio, the works. The top ~12% of the catalog gets daily_eligible: true on the public payload, which is the same pool the mobile app's daily push and the social drafter both pull from.

Here's the surprise. The eligibility rate per category looks nothing like the volume distribution:

Sales tops the chart with a 53.2% eligibility rate. Engineering — the category with 11× more skills — runs at 13.8%. Security beats Engineering and Engineering beats General. The volume ranking and the quality ranking are nearly orthogonal.

Two structural reasons. First, sales skills tend to be written by full-time SDR/AE/RevOps teams who are paid to produce structured playbooks — the same authoring discipline they bring to call scripts and email cadences shows up directly in the Quality Score. Second, Sales is small enough (1,253 skills) that one team of disciplined authors moves the rate visibly. Engineering has 13,558 skills with much wider variance: the top decile is exceptional, but the median is dragged down by quick-and-dirty per-project utilities that nobody intended as reusable.

The practical implication is that "search by category" and "filter by daily_eligible" produce dramatically different shortlists. If you're looking for a polished, broadly-reusable skill in a vertical you don't know well, lead with the daily_eligible filter — it surfaces ~660 polished sales skills in a way that browsing the 1,253-skill category never would.

4. Author dynamics — a deep long tail with a few power users

5,053 distinct authors have contributed at least one skill. Of those:

• 2,404 authors have published ≥2 skills
• 557 authors have published ≥10 skills
• 104 authors have published ≥50 skills
• 28 authors have published ≥100 skills

That distribution is classic power-law. The top-10 authors collectively account for about 11% of the catalog. The single largest contributor — Jeremy Longshore at Intent Solutions — has 3,304 admitted skills, more than all 28 sub-100-author authors below them combined. The next tier down (Mahipal, Diego Souza, Pranav Nagrecha) is in the 600-700 range.

For the catalog's purposes, both ends of the distribution are healthy:

• The head is a small number of focused authors building large, internally-consistent skill libraries (often 3-clause MIT and openly reusable). They're where the daily_eligible pool draws disproportionately from.
• The tail is everyone else — 2,649 authors with exactly one skill in the catalog. Most of those are one-off contributions, often a single brilliant skill for a niche workflow, which would be invisible without an open registry.

The point of an open catalog is that both tiers get the same visibility — admission is content-derived, not author-derived. A first-time contributor whose single skill clears the admission threshold appears next to a top-10 author's hundredth skill on the same category page, sorted by the same Quality Score.

5. What the daily_eligible bucket reveals

The 7,285 daily_eligible skills are the catalog's "polished" tier. Browsing it makes the medium-and-above quality watermark concrete in a way browsing all 69k can't. Three patterns stand out:

1. Authoring discipline transfers across verticals. The 47 authors with ≥10 daily_eligible skills include people who publish primarily in sales, primarily in engineering, and primarily in security — but their per-skill scores are nearly indistinguishable. Authoring craft compounds; vertical doesn't matter much once you've internalized the structure.
2. Integration and review tags overrepresent in the eligible bucket. If type:integration is 12.8% of the whole catalog, it's ~20% of the daily_eligible bucket. Tasks where the success criterion is unambiguous — "did this connector ship?" — produce skills that themselves get scored higher.
3. Most languages over-index. Skills tagged lang:python or lang:typescript hit daily_eligible at a higher rate than the catalog average — possibly because language-specific skills are written by people who already maintain code in that language and bring the same care to the SKILL.md.

6. What's missing — the gaps an outside contributor could fill

The most useful read of any catalog is what isn't in it. A few observations:

• Mobile-specific skills are scarce. 642 skills tag tool:docker; only a handful tag iOS, Android, or React Native equivalents. There's an obvious gap for skill authors who work in mobile-app codebases.
• Database-specific tooling is underrepresented. Less than 1% of the catalog tags itself with a specific RDBMS or storage backend, while ~12% of repos on GitHub use one. Almost certainly an artifact of where the mining sources draw from rather than a real authoring gap.
• Non-English skills are very rare. The catalog accepts SKILL.md in any language and the mining queries do try i18n keyword variants, but the admitted count of non-English skills is small. If you maintain a Claude Code workflow in Japanese, Chinese, or Spanish, your SKILL.md would be one of the only ones in its language tier.
• Skills with explicit failure-mode sections score highest. The 2026-05 score axis "anti-trigger discipline" looks at What this skill does NOT do sections, "When to skip" sections, and out-of-scope statements. Skills that include these consistently land in daily_eligible. If you're authoring a new skill, a 50-word "When NOT to use this" section is the highest-leverage edit you can make.

FAQ

How was this catalog assembled?

ClaudSkills mines 26 public sources twice daily. Every admitted skill clears a content-quality threshold. No skill author pays anything to be listed — admission is rule-based and entirely content-derived.

What is the daily_eligible flag?

daily_eligible = true is set on the public payload for any skill whose Pro Quality Score is at least 80 on a 0-100 scale. That's the top ~12% of the catalog. The Skill of the Day picker, the mobile app's daily push, and the social drafter all pull from this same pool so they stay in lockstep.

Is the catalog comprehensive?

No catalog of a fast-growing ecosystem ever is. ClaudSkills covers public sources only — anything behind authentication, private GitHub repos, internal corporate skill libraries, and skills that exist solely as Loom-walkthrough screenshots are out of scope. The 26 mining sources cover the bulk of the public surface; the remaining gap is the unknown unknown.

Why does Sales have the highest quality rate?

Sales skills tend to be written by full-time SDR/AE/RevOps teams paid to produce structured playbooks — the discipline shows up in the Quality Score. Plus the category is small enough that one team of disciplined authors visibly moves the rate.

Can I submit my own skill?

Yes — paste your SKILL.md URL into the Submit form on the homepage. The miner picks up submissions on its next nightly run (01:00 local, typically finishes by ~05:00). If your skill passes the admission threshold it appears in the catalog within ~24 hours. There is no fee, no acceleration tier, no paid placement.

Related reading

• How to write a Claude Code skill (long-form tutorial)
• SKILL.md frontmatter reference
• Claude Code Skills vs MCP servers — when to use which
• Claude Code skill catalogs compared
• Live catalog stats · Skill of the day archive