Quick Answer
Most Claude Code users do not have an installation problem. They have a selection problem. The open skills ecosystem is large enough to be useful and messy enough to burn an afternoon if you evaluate every repository by hand. The official vercel-labs/skills project currently shows about 25k GitHub stars, and its README positions it as the CLI for an open agent skills ecosystem that supports Claude Code, Codex, Cursor, OpenCode, and many more agents. That matters because it shifts the bottleneck. The hard part is no longer “can I install a skill?” It is “how do I trust which skill to install next?” A discovery layer like Find Skills matters because it compresses browsing, trust checking, and first-use testing into one repeatable workflow instead of twenty random GitHub tabs.
If you already use Claude Code, the next thing slowing you down probably is not the terminal. It is the moment after the terminal. The moment when you know a skill probably exists, but you have no clean way to judge whether the repo is real, maintained, battle-tested, and worth adding to your workflow.
That is why “just install more skills” is weak advice.
The official vercel-labs/skills project is not small. On July 4, 2026, the repository page showed about 25k stars. A June 30, 2026 research snapshot recorded 24,418 stars exactly. The project README also says the CLI supports Claude Code, Codex, Cursor, OpenCode, and dozens more agents. In other words: the ecosystem is not the problem. The problem is what happens when a useful ecosystem gets bigger than your patience.
Definition
A discovery layer is the workflow that helps you find, test, and trust the right tool before you commit to it. In this case, it means moving from manual repo hunting to a skill-search process with visible install paths and trust signals.
The Real Bottleneck Is Not Installation
Most people describe this category the wrong way.
They say Claude Code skills are hard because setup feels technical. That is sometimes true for a first-time user, but it is not the bottleneck for anyone who has already installed a package, cloned a repo, or followed a README once in their life. The real drag is evaluation. You open one repo. Then another. Then a Discord thread. Then a gist. Then a video. Then a half-maintained project with a clever name and no proof behind it.
By the time you reach a decision, you have spent more energy choosing than using.
That is why the “too many repos” problem matters. It is not cosmetic. It changes whether the ecosystem feels usable at all.
| Workflow step | Manual repo hunting | Discovery-layer workflow |
|---|---|---|
| Find candidate | Search GitHub, X, Reddit, Discord, README links | Search one skills surface |
| Check trust | Scan stars, forks, commit recency, docs by hand | Use surfaced trust signals first |
| Test quickly | Read docs and infer next command | Use documented skills use path |
| Decide to keep | Hope the install was worth it | Upgrade from test to install intentionally |
Right there is the whole argument. The best tool is often not the one with the flashiest name. It is the one you can trust fast enough to actually use.
What The Official Project Already Proves
The official README for the skills CLI does a few important things that most social posts skip.
First, it frames the project plainly: the CLI for the open agent skills ecosystem. That matters because it is not pretending every useful skill lives in one monolithic product. It is acknowledging a wider ecosystem and giving you a way to navigate it.
Second, it documents both install and temporary-use paths. You can install a skill with:
npx skills add vercel-labs/agent-skills
And you can use a skill without installing it permanently:
npx skills use vercel-labs/agent-skills --skill web-design-guidelines --agent claude-code
That second path is more important than it looks. It turns evaluation into a reversible step. Instead of committing first and learning later, you can test first and keep only what earns its place.
Third, the README shows breadth. Claude Code is not a side note. It is named explicitly among the supported agents. That makes this a real Claude Code workflow story, not a generic “AI tools are everywhere” story.
Why Repo Breadth Creates Decision Fatigue
Open ecosystems are powerful for the same reason they are tiring.
They create choice.
Choice sounds good until the unit of choice is another repo, another README, another install method, another naming convention, another half-finished example, another “best practices” skill that may or may not match your actual work. The user experiences that as drag, not abundance.
That is the trap.
When people say they want more automation, what they often want is less uncertainty. They want the path from problem to action to feel shorter. They want fewer false starts. They want to stop installing three things to discover the fourth one was the right answer all along.
Key Takeaways
- The live `vercel-labs/skills` repo showed about 25k GitHub stars on July 4, 2026, which is a strong public trust signal.
- A June 30, 2026 source snapshot captured 24,418 stars exactly, useful when you need a dated, precise reference.
- The official README names Claude Code, Codex, Cursor, OpenCode, and 68 more supported agents.
- The documented workflow includes both `npx skills add` and `npx skills use`, which means you can test before you commit.
Find Skills Changes The Question
Without a discovery layer, the user keeps asking:
“Which repo should I read next?”
With a discovery layer, the better question becomes:
“What is the best-fit skill for this exact use case?”
That sounds subtle. It is not.
The first question is a browsing question. The second is a workflow question.
Browsing questions create chaos because the answer set is huge and unstructured. Workflow questions create momentum because they are tied to the job you actually need done. That is why a Find Skills style layer is valuable. It does not win by being another shiny repo. It wins by reducing the number of decisions between need and action.
You can see the difference in how quickly someone moves from “I heard there might be a skill for this” to “I tested the right one and installed it.” That is the difference between an ecosystem you admire and an ecosystem you actually use.
The Missing Trust Filter
Most discovery conversations stop too early. They say search is good. Search is not enough.
What matters is the trust filter around the search result.
For a Claude Code user, the filter usually looks like this:
- Is the source real?
- Is there visible adoption?
- Does the README explain the install path clearly?
- Can I test it before making it part of my setup?
- Does the result fit my actual use case instead of a generic demo?
That is what battle-tested really means in practice. Not hype. Not a loud launch thread. Not clever branding. A clear path from discovery to trial to decision.
The official project already gives you some of those signals. Public repository. Public README. Public install flow. Public temporary-use flow. The discovery layer matters because it lets those signals do useful work instead of forcing you to assemble them manually from ten browser tabs.
What This Means For Claude Code Users
If you are using Claude Code seriously, the question is not whether more skills exist. They do.
The useful question is whether your setup helps you find the right skill faster than your current habit of manual browsing.
If the answer is no, you are paying a decision tax every time a new use case shows up.
That tax is easy to ignore because it looks like “research.” But in practice it often means half an hour lost before any command gets run. It means momentum breaks. It means you postpone trying something useful because you do not trust the search process enough to start.
And once that happens a few times, the ecosystem starts to feel bigger than it is helpful.
The smartest move is not to memorize more repo names. It is to install a better way to choose.
Frequently Asked Questions
Is Find Skills replacing the whole skills ecosystem?
No. It is valuable because it helps you navigate the ecosystem, not because it replaces it. The official repository still acts as the public foundation, and the live GitHub page showed about 25k stars on July 4, 2026. The discovery layer makes that ecosystem easier to use without forcing you to browse everything manually.
Why not just read GitHub repos by hand?
You can, but the cost shows up in time and uncertainty. Manual repo hunting means every use case starts with browsing, trust-checking, and guesswork. The documented skills use flow matters because it creates a reversible test path before permanent install. That cuts down on blind installs and wasted evaluation time.
Does this help only beginners?
No. Beginners feel the confusion first, but even experienced Claude Code users pay the same selection tax when the ecosystem gets broad enough. The difference is that experienced users hide the pain better. The workflow benefit still exists: fewer false starts, faster testing, and cleaner decisions.
Why does the star count matter if it is only one metric?
Because trust starts with visible signals. A star count does not prove fit, but it helps establish whether a project has real public adoption. Here the current live page showed about 25k stars, and a June 30 source snapshot recorded 24,418 exactly. That kind of public traction is a better starting point than random repo browsing with no context.
Read Next
- Claude Code's God-Tier Slash-Command Stack for Day-Long Agent Loops
- Ponytail Makes Claude Code Write Less. That Is the Point.
If you want the exact setup flow, install-scope choices, and the decision checklist for testing a skill before keeping it, open the full resource: The Find Skills Install Playbook for Claude Code.