The install is very opaque. It's not clear where these skills are installed, how to upgrade them or remove them.
Here's the `skills` package on NPM: https://www.npmjs.com/package/skills - it's MIT licensed but I can't find it on Github.
`skills` looks to be a wrapper around `add-skill`: https://github.com/vercel-labs/add-skill
From the docs, `add-skill` auto detects from 16 different potential paths to copy skills to in a repository (.claude/, .codex/, .Gemini/, etc).
`add-skill` also let's you install skills globally (~/). From the code, `skills` looks like it doesn't support global installs but under the hood it passes all args to add-skill, so you should be able to install skills globally or install multiple skills (even if the wrapper doesn't expect it).
Aside: although lots of agents have adopted SKILLS.md conventions, they're currently all using their own paths. There doesn't seem to be a consensus yet, like there is with AGENTS.md. There are even 3 generic paths: .agent/skills/, .agents/skills/ and just skills/
The best thing I can come up with right now for multi-agent installation is symlinks.¹ This tool doesn’t seem to even try to solve the updating or versioning requirements.
¹ https://github.com/hodgesmr/agent-fecfile?tab=readme-ov-file...
Although it's not clear how to upgrade them (I doubt there is any version management built-in), the installer does specify where it will be installed (And lets you choose global/project level.
Skills certainly got a wildwest pre VCS copy-paste feeling going on.
At risk of piggybacking off someone elses post, my team just published noriskillsets.dev. It's maybe not as polished as this beautiful vercel site, but I think it solves a bigger problem: every skill in noriskillsets.dev is something that we've personally battle tested and evaluated as best in class for that particular task.
I love vercel, but I think the 'collect a bunch of random skills' approach just isn't it. You need versioning, linking between skills, an easy install client...basically a full package manager, which this is not.
If any of that is interesting, check out noriskillsets.dev. Would love to get feedback!
As someone who has found skills useful, seeing skills like this[0] raises the same question about (a subset of) skills as did MCP: why not just have the agent run ‘tool --help’?
The point is for the agent to have an index of available skills so that it can decide autonomously if it needs to load one.
https://zzbbyy.substack.com/p/what-are-skills
In my experience it doesn’t work too well with codex, but I expect llm providers to train them on that use case and improve the situation soon.
Not all flags will be relevant, you can trim down and contextualise the instructions for your specific project.
I've been using Nix to manage my skills instead. It's been great. Especially because I can now declaratively manage all the cli tools and mcps my skills depend on.
What is this? How does it work? How are skills ranked? Seems a little bit fishy to me that you can only tell it's from Vercel if you click the top left corner, and the top two skills come from vercel... despite there definitely being much more used skills in the overall AI coding ecosystem.
The UI looks nice, otherwise. I had thought about building something like this - maybe this just increases my confidence that this is needed, just not affiliated with a company.
skills are ranked by anonymous telemetry from running `npx skills add <owner/repo>`
Vercel's skills are popularly installed because we initially launched `npx skills` with the launch of our `react-best-practices`
But have been developing the tool in tandem!
> How are skills ranked?
By npm weekly installs (??). Famously good signal for quality.
Edit: Not even npm, their own tools download count...
Honest question: has anyone found skills that fundamentally changed their workflow vs. ones that are just ‘nice to have’? Curious what the actual power-user stack looks like.
Anyways, great work on this btw, the agent-agnostic approach is the right call
Basically skills are /commands that can have attached scripts, that's about it.
If your "skill" doesn't come with scripts/executables, it's just a fancy slash command.
I've had success with code quality analysis skills that use PEP723 Python scripts with tree-sitter to actually analyse the code structure as an AST. The script output and switches are specifically optimised for LLM use, no extra fluff - just condensed content with the exact data needed.
I would double - do skills reliably work for you? I mean are they reliably injected when there is a need, as opposed to being actively called for (which in my opinion defeats the purpose of skills - because I can always ask the llm to read a document and then do something with the new knowledge).
I have a feeling that codex still does not do it reliably - so I still have normal README files which it loads quite intelligently and it works better than the discovery via skills.
> power-user stack
Try installing the Claude Superpowers skills - you can install them one by one from here, but it's easier to install the superpowers plugin. Try using it for a couple of sessions and see how it works for you.
For a full test, try starting with the brainstorming one which then guides you from brainstorming though planning, development etc.
I've been using it for a few days and I would say it's enhanced my workflows at least.
One simple but useful flow is to ask cc to review a session and find miss-matches between initial skills / agent.md and current session, and propose an edit. I then skim over it and add it. It feels like it helps, but I don't have quantitative data yet.
Define "fundamental", but I added skills to run complicated evaluation procedures for my ML research. This way I can open 5 CC instances, let them run and iterate on research without intervention. After that, I can do the final review.
My experience with them is limited, but I’m having issues with the LLMs ignoring the skills content. I guess it makes sense, it’s like any other piece of context.
But it’s put a damper in my dream of constraining them with well crafted skills, and producing high quality output.
Yeah, the context window is a blunt instrument, everything competes for attention. I get better luck with shorter, more opinionated skills that front-load the key constraints vs. comprehensive docs that get diluted. Also explicitly invoking them (use the X skill) seems to help vs hoping they get picked up automatically
Yes, unfortunately the most reliable way is to inject them into the user prompt at a fresh session. My guess is that biasing towards checking for the tools availability too much affects performance, which might explain why it is quite rarer that I see it just choose to use a skill without previous prompting.
Yeah, I'm still trying to figure out why some skills are used every day, while others are constantly ignored. I suspect partially overlapping skill areas might confuse it.
I've added a UserPromptSubmit hook that does basic regex matches on requests, and tries to interject a tool suggestion, but it's still not foolproof.
So this is a page made by Vercel which conveniently puts the Vercel skill as top of the list.
Nice.
The page should properly advertize that this is a corporate built/sponsored site so people can understand that it's totally biased.
Nice work! I don't think Vercel is the first to do this, but it's a good idea and I'm glad to see more players in this space.
A small UI suggestion: it would be helpful if hovering on a row showed the skill description, along with a button to copy the install command.
For anyone interested, there are two other sites already doing something similar:
- claudemarketplaces.com - A comprehensive directory with 1900+ marketplaces, shows descriptions directly in the list view with copy-to-install commands
- skillsmp.com - Has 77K+ skills indexed from GitHub. Cool developer-style UI, but honestly the UX could use work—the search is hidden behind cryptic command-style buttons and it's not obvious how to actually search
Also worth checking out the Claude Code Mastery guide (thedecipherist.github.io/claude-code-mastery) for a deeper dive into skills, hooks, MCP, and CLAUDE.md.
If you want to share skills (and other tools) with teams at scale, consider sx - https://github.com/sleuth-io/sx
[yup, my project :)]
Also this one is trending on HN.
Nice, thanks for sharing these, I have a folder where I try to save all resources that will be beneficial. So much to discover and keep track of.
We had already done it before Vercel did.
openskills.space
This is a really good implementation, but I don’t lean too heavily into skills especially not other people‘s. If I’m doing design who’s to say I want instructions in there in the first place like “pick an extreme“ (instructions in the design skill featured on the homepage)
The leaderboard is ranked by the weekly download count by their "npx skills" command. This is Vercel new "standard" skills installer so obvious their skills are at the top.
Assuming that's a response to me... that explanation would have been nice, and it's misleading without it.
Hi! This is all explained in our docs: https://skills.sh/docs/faq
Why do none of these “npm for Skills” document any way to do basic package management things like updates, version-pinning, or even uninstalls?
;) because you can ask Claude to do all of those things, but which package manager? there's a dozen ways to install packages, will Claude pick the right one you wanted? Your skill will guarantee consistency
I don’t understand this reply at all. If I use this skills.sh tool to install the Foo Skill for one or more agent harness, how do I update Foo later?
> but which package manager?
The one being linked to.
I misunderstood your question, thought you were confused about a skill to enforce NPM usage by Claude to use NPM strictly as opposed to another package manager. Last time I post on HN past midnight.
I wish I knew why my skills are never called…including my custom sub agents.
Maybe it’s my own ignorance, but Claude loves to ignore its CLAIDE.MD which says it’s mandatory to leverage sub agents to delegate tasks and use skills for accomplish specific workflows.
Every time I call Claude out it tells me it knows and chose to ignore it, even going as far as saying it’s not my decision.
Any tips?
Create a hook that would ask Claude Code to evaluate all skills in the project and decide which are applicable to the current task at hand. It is easy and works very well.
Forget Claude.md
I can’t thank you enough for course correcting me here, this is exactly what I should have been doing. I did some reading and I clearly wasn’t using any of the features correctly or efficiently.
My Claude.md was nearly 1,900 lines. It’s down to 150 lines now with Skills fully built out for the agents, and a hook to steer the ship. It’s all working perfectly now.
Thank you again!
You are welcome. I have myself picked up this trick reading through Reditt and later polished that myself. No official guide is suggesting that
Now that it’s set up it feels obvious, but I agree nowhere was this clear to me that this is how skills, hooks, and subagents were intended to be configured.
I was so much more productive today and used far less tokens. Thank you again!
Thanks for the tip.
Please make a rest API!