I have done this. 100% recommended!
Next challenge: share skills and config in your fleet 🤔
I'm sick of having to setup + auth MCP servers across multiple machines and harnesses
Giving @RhysSullivan's Executor a shot - connect apps once and then access everything over a single (or multiple) MCP connection
I've got mine self-hosted and only accessible on my tailscale tailnet
1
167
still collecting answers on this one. if agents keep getting Nuxt wrong for you, tell me where.
i'm exploring how we could improve agents writing Nuxt code
where do they struggle for you?
- not using h3 utils like readValidatedBody
- data fetching (useFetch vs $fetch)
- Nitro / server routes
- modules (Nuxt UI, Content, i18n, Image...)
- deployment
reply with the module or mistake you keep correcting
2
8
532
Maxi retweeted
Skills CLI v1.8 now ships with the programmatic API! Give it a try, and we would love to hear your feedback!
Do I know anyone working on the Vercel skills CLI? @vercel
I would love to chat about a programmatic install API. Frameworks like Nuxt, or people selling their skills, could build on top of it instead of spawning `npx skills add`.
github.com/vercel-labs/skill…
You can now try all 40+ Comark UI components in the browser before installing anything.
Pick an example or write your own Markdown, and watch it render right next to it.
comark-ui.onmax.me/play
ViteHub is becoming something serious, but it is still early.
Server primitives for Vite and Nuxt: database, blob, kv, queue, auth, agents... one API for every host.
I don't want to build it alone. Tell me what you need, or join as a maintainer 🙂
chat.vitehub.dev
Mmm, I don't agree with this one
if your laptop CPU is at the limit because of agents, you are running way too many agents. at that point yes, move to a cloud server. but a VPS, not vercel
I have a babysitter project which takes care of my PRs. he takes them to merge on 7 repos, 24/7. on a busy day is ~600 agent passes, 214 agent-hours. all of it runs on my vps (16 cores)
I asked my agent how much it would cost moving it:
- vercel sandbox: ~$350-820/month
- cloudflare containers: ~$220-520/month
and each agent uses ~0.3 CPU. most of the time he is just waiting the model, but you pay the memory all that time
vercel and cloudflare are great, but they solve a different problem: traffic when you have tons of users. a sandbox there makes sense if you need 1 isolated dev container for whatever reason. for your developer needs is way too expensive, and is not better DX either
At @vercel we got to test this model before release. Best way to summarize it: astonishingly brilliant.
One interesting takeaway: it's very clear that this new generation of models can take on unprecedented bodies of parallel work.
To actually seize this intelligence, you need cloud sandboxes. Full stop. The model can delegate and orchestrate far more than a single computer can handle. Likely far more than you can even imagine it can handle.
Your bottleneck is now ① your ambition and ② your parallel computing power.
Btw, I learned from @cramforce you don't need to overcomplicate it. Install Vercel CLI, setup the Plugin and tell the agent to use /𝚘𝚏𝚏𝚕𝚘𝚊𝚍-𝚝𝚘-𝚜𝚊𝚗𝚍𝚋𝚘𝚡 skill. It'll provision and de-allocate cloud sandboxes as needed. You don't need a "multi-agent remote harness ralph orchestrator" 😆
some things stick with you. @antfu7's post about Shiki Magic Move has been one of those for me.
antfu.me/posts/shiki-magic-m…
the other day i needed something similar for some Markdown i was writing. remembered that post, smiled, and built a component for Comark UI.
put code blocks inside ::magic-move and they become animated steps. matching tokens slide into place, new ones fade in, old ones fade out.
comark-ui.onmax.me/magic-mov…
i miss the deep focus sessions we had before agents. hours on one problem, no prompts, no 6 parallel threads.
today i got a bit of that back, and the soundtrack was semana santa processional marches.
a brass band slowly carrying a float down the street is somehow the perfect tempo for thinking.
open.spotify.com/playlist/25…
79
i'm exploring how we could improve agents writing Nuxt code
where do they struggle for you?
- not using h3 utils like readValidatedBody
- data fetching (useFetch vs $fetch)
- Nitro / server routes
- modules (Nuxt UI, Content, i18n, Image...)
- deployment
reply with the module or mistake you keep correcting
Comark UI now has quizzes in plain Markdown. A question is a task list, and each option says why it's right or wrong.
Pair it with @mattpocockuk's /teach skill and your agent writes lessons that quiz you back. Great way to find out what you didn't actually learn.
comark-ui.onmax.me/quiz
nuxt-formisch v0.1 is out, rebuilt on Formisch 1.
One Valibot schema in shared/ types the form and validates the request body. With Nuxt UI, <FAutoForm :schema="LoginSchema" /> renders the whole form.
I've used this in my projects for over a year. Never looked back.
nuxt-formisch.onmax.me
Comark renders Markdown to HTML 6-7x faster than remark, so I wrote a codemod to switch.
pnpm dlx remark-to-comark migrate
comark-ui.onmax.me/remark-to…
I wanted a Markdown editor where several people edit the same doc live.
So Comark UI has one now. Tiptap, Markdown in and Markdown out. Plug it into ViteHub Realtime and you get live cursors and who's editing.
comark-ui.onmax.me/play/edit…
Is just me or everything he does is inspiring?
As agents write more of our code, we'll read less of it, and the pull request becomes most of what we judge a change by.
So every PR I open now says who did what, shows the change with real evidence (before/after, both outputs, measured numbers), and ends in plain words. I wrote up how it works, what it doesn't fix, and a skill to do the same.
hugorcd.com/writing/what-my-…
2
9
517
Do I know anyone working on the Vercel skills CLI? @vercel
I would love to chat about a programmatic install API. Frameworks like Nuxt, or people selling their skills, could build on top of it instead of spawning `npx skills add`.
github.com/vercel-labs/skill…
Nuxt 5 is gonna be awesome
small, fast, and it does the job. we're still talking about parsers 🤏
xitter.twitterapi.xyz/_onmax/status/21172676…
tiny v0.1.0 is out: dependency-free replacements for the YAML, JSON5 and INI parsers you already use. 18 drop-ins for 11 libraries.
npx tinyswap
It rewrites your imports and swaps the dependencies. Or change one import by hand.
tiny.onmax.me
🧵
1
126
tiny v0.1.0 is out: dependency-free replacements for the YAML, JSON5 and INI parsers you already use. 18 drop-ins for 11 libraries.
npx tinyswap
It rewrites your imports and swaps the dependencies. Or change one import by hand.
tiny.onmax.me
🧵
The parsers are also TypeScript 7.1 content mappers, so importing a .yaml, .json5, .jsonc or .ini file gives you a typed object.
tinyaml also has a YAML linter, an ESLint plugin and a Vite Doctor extension.
1
32
Works for js-yaml, yaml, yamljs, @std/yaml, json5, strip-json-comments, jsonc-parser, comment-json, jju, ini and @one-ini/wasm.
Release notes: github.com/onmax/tiny/releas…
Repo: github.com/onmax/tiny
28
















