
oscar gabriel
1.8K posts

oscar gabriel
@oscabriel
data analyst learning web dev / plant-based / movie enjoyer
pdx Katılım Aralık 2021
560 Takip Edilen448 Takipçiler
Sabitlenmiş Tweet

ok it's here!
introducing the offworld.sh cli
"one skill for your whole stack."
use the skill + the cli with your coding agent of choice to build a working memory of your favorite open source repos. ask questions, learn the codebase, and troubleshoot issues in your own project with real answers based on proper up-to-date source code.
don't waste your context window with docs links, bundled node_modules code, or heavy duty mcp servers.
just give your agent a map to your clone dir and let it rip.
English

@wesbos we need to get the web apps up to parity w/ the local harnesses if ever ppl are to start seeing that there is real potential in llms after all
English

goooot it. wow this is awesome. almost a year ago I stitched together cloudflare infra products with the ai-sdk to create a per-user DO based chat app (handling every step of the message lifecycle manually in the DOs). Refactoring that project to just be built on top of Think instead would make for a legendary diff haha
English

@oscabriel came before!
blog post: blog.cloudflare.com/project-think/
docs: developers.cloudflare.com/agents/api-ref…
more coming soon
English

I think you're right that github is less to blame for some of the new problems with open source, and that it instead lies more in all the ppl who irresponsibly point their agents at repos w/o regard to the humans on the other side of their (often passive) requests. But I also don't think github has the tools needed to keep those agents under control.
If something is to come after gh, it should be something that does that filtering on behalf of maintainers
English

Has been a while since I wrote about agentic engineering, so this time around some learnings of maintaining Pi as a junior maintainer to @badlogicgames :) lucumr.pocoo.org/2026/5/24/pi-o…
English

well yea that's just how LLMs work, right? to make an ai product that's actually "live" would require a fundamentally different underlying technology (and presumably another reappropriation of the term "ai")?
signüll@signulll
one of the most interesting things about ai products today is that almost none of them are *live*. there’s nothing running continuously, reacting to context as it changes.. maybe a scheduled digest here or a timer there, but that’s just pull dressed up as push. everything is fundamentally a vending machine where you walk up, ask, get an answer, & then leave. getting this right is obviously tricky & the business model behind must fit to justify the burn but this is where really interesting application layer problems live rn.
English


My personal best AI coding tip
You should be asking the agent a lot of questions that you already know the answer to. It's important to put all of the thoughts YOU have into its context BEFORE you try to have it DO something.
Just shooting off a request and expecting it to KNOW anything about it is still largely foolish.
English

there are jack sparrows everywhere for those with eyes to see
Gruz@damnGruz
Someone built a compass app that points to the nearest liquor store 😭 People on Reddit are on another level
English

@peduarte oh shit I didn't know it was free rn. def gonna try it out!
English

@nilofarzzzz that was the allure for me originally, did it right away and got like 98% of the way there I'm pretty sure. Still discover a straggler I forgot about like once or twice a year
English

@nbbaier I miss when things were happy to just be what they are
English












