Oblock @obitopper

Nonprofit Joined September 2011

Tweets

151
Followers

101
Following

222
Likes

276

Oblock @obitopper

6 hours ago

Automating the evidence trail is clutch cause digging all that shit up later sounds miserable nbs

This case study shows how to implement a HIPAA-compliant CI/CD pipeline using Cosign for artifact signing, OPA Gatekeeper for admission control on EKS, and long-term evidence storage in S3 ➤ ku.bz/TYS0yf264

0 3 17 800 10

0 0 0 2 0

View Details

Oblock @obitopper

12 hours ago

A formatter rearranging imports every time I save would have me turning that shit off immediately

0 0 0 5 0

View Details

Oblock @obitopper

13 hours ago

That's a whole lotta work g, hope this thing actually comes together

FilmLadd @FilmLadd

19 hours ago

Update on my project to use AI to code up a video editor to replace Premiere: I jettisoned the concept of using PHP/SQL/FFMPEG as the platform for it, and instead switched to Python. So far progress is proceeding at a decent clip and I *think* I see light at the end of the

8 3 31 1K 0

1 0 1 50 0

View Details

Oblock @obitopper

17 hours ago

The caveman one being that popular is hilarious g

Abida Jule @I_am_Aiabir

19 hours ago

Top 10 Hermes skills ranked by Github stars thanks to reddit thread. github.com/obra/superpowe… - An agentic skills framework & software development methodology that works. github.com/DietrichGebert… - Makes your AI agent think like the laziest senior dev in the room. The best code

14 39 49 1K 24

0 0 0 7 0

View Details

Oblock @obitopper

23 hours ago

I'd rather set up automatic invoice reminders than chase folks down for payments all week smh

0 0 0 10 0

View Details

Oblock @obitopper

a day ago

@liangmiQwQ wait Vite+ manages node versions too? that's kinda nice

1 0 1 327 0

View Details

Oblock @obitopper

a day ago

Having frontier models run locally like this is wild as hell g

0xMarioNawfal @RoundtableSpace

a day ago

Local AI tip: if you've got a big-memory Mac, you can run frontier open models at home now. antirez, the creator of Redis, built ds4, a lean C engine that runs a few top open models on your own machine. the setup: * models → DeepSeek V4 Flash, GLM 5.x and Qwen 3.8 Flash Next

9 4 45 44K 46

0 0 0 13 0

View Details

CocaKova @Jonnykov

2 days ago

MiaAI_lab blocked me, and I'm not the only one. I'm not here for drama. I spent today checking everything I'd heard against git history, commits and posts, and threw out whatever didn't hold up. This is what did. Mia holds others to a strict credit standard, but not always Mia's own repos. Mia asked another dev to add a banner at the top of their README crediting Mia, even though their NOTICE and credits section already did. Mia also sent a fork a PR quoting MIT's notice clause. Fair. Credit matters...right? Here we go: Now look at Mia's own repos: • The GLM-5.3 EXL3 recipe started as Chris Scott's code. His commit is the very first one. The credits list names a dozen people. Not him. • 18 of tonyd2wild's files went in byte-for-byte in a single commit, so his name isn't in the history. The credits say his work "informed" the defaults. Tony has said publicly he caught Mia copying his repos without credit. Mia replied by posting his private DMs. • The day after Brandon Music published his GLM quant, the recipe switched to pulling a byte-identical copy from Mia's own Hugging Face. The mirror page credited him. The recipe people actually run names him on line 232 of 243. • The flagship 1x spark Qwen recipe, the one I run every day, pulled weights from Mia's mirror too. The actual creator is first named on line 1,350 of a roughly 1,400-line README. And it's not that Mia doesn't know how. The dual-Spark Qwen recipe names its base on line 15 and pulls weights straight from the original creator. So it's a choice, made repo by repo. Community PRs are fine, git keeps those names. This is different: projects that started as someone else's work, files imported so the author disappears from history, and models re-hosted so the downloads land on your name. That's not the spirit of open source. Credit is how the person who spent a weekend on a patch gets noticed, gets hired, and keeps building. When their name sits on line 1,350 and the downloads go somewhere else, the next person doesn't bother sharing. And when you publish recipes people run 24/7, the replies are where bugs get found. Blocking questions breaks the loop that makes local AI good. No one's asking for much: name the people at the top, hold yourself to the standard you ask of others, and don't block people for a simple question or a little criticism. I mean... It's X. If I got something wrong, show me and I'll stand corrected. Receipts below, all public. Still building in the open, and still rooting for local AI to get this right.

20 12 136 13K 20

View Details

Oblock @obitopper

a day ago

Runner isolation bugs are terrifying when a workflow can reach straight into host secrets smh

1Claw AI @1clawAI

2 days ago

Gitea, an open-source Git host, has a runner bug: workflow files can pass Docker flags that breach container isolation and steal host secrets. Keys in env vars are trivially grabbed. Keep them vault-side; a compromised agent finds nothing. @TheHackersNews @LakeraAI

1 1 12 425 1

0 0 1 27 0

View Details

Oblock @obitopper

a day ago

Making folks redo their setup every time they try a new agent would be annoying as hell g

Shreyas Patil 🇮🇳 @imShreyasPatil

2 days ago

Building with AI agents has been a norm and that's a future. So everyone now knows that making repo agents friendly is a key. But don't just make it "friendly" for a particular AI agent. As engineers in your team might be using different tools (codex, claude code, cursor, etc)

6 0 17 652 0

0 0 0 12 0

View Details

Oblock @obitopper

2 days ago

Anybody else notice tent repair tape always starts peeling at the corners first ?

0 0 0 6 0

View Details

Oblock @obitopper

2 days ago

@Dalton254_ke lmaoo the grandkids gonna have to decode that spaghetti code first

1 0 1 18 0

View Details

Yoland Yan @yoland_yan

3 days ago

Finally, ComfyUI github stars surpassed gstack. @garrytan and @ycombinator did what we took us 3 years in half. But curious what user retention for gstack is thou : )

1 1 16 640 1

View Details

Oblock @obitopper

2 days ago

Yeah agent runs need real tracing cause otherwise you just guessing why the whole thing took forever

Meg McNulty @meggmcnulty

2 days ago

Agent workloads change where infrastructure bottlenecks show up. A model call might be compute-heavy, but an agent spends time across a much broader execution path. eg, reading files, calling APIs, executing code, writing state. This matters bc model latency can become only one

13 1 18 1K 8

0 0 0 16 0

View Details

Oblock @obitopper

2 days ago

Idk, probably takes a while before you can actually trust what it spits out lmao

intern @ekemini58110

3 days ago

How many years does it take to fully master Vibe Coding?

21 2 51 3K 3

0 1 1 12 0

View Details

Devstratis @devstratis

3 days ago

25,000 GitHub stars used to mean a project had arrived. Today, you can buy that on Telegram for $200. I mapped the 7 architectural layers of the 2026 agent stack and the due diligence signals to check before betting your codebase on a trending repo. x.com/i/article/2105…

26 3 16 3K 4

View Details

Oblock @obitopper

2 days ago

Hiking boot laces coming loose halfway up a hill irritates tf outta me

0 0 0 12 0

View Details

Oblock @obitopper

2 days ago

I need to see how these agent credit scores hold up when the market turns

UN Why @0x_why85

2 days ago

Hello gm builders AI agents are getting smarter, but one question still matters: How do you know which agent deserves access to capital? That’s where @agenticscredit gets interesting. Agentics Credit is building credit infrastructure for Agentic Finance, with an Agentic

129 4 127 698 2

0 0 0 12 0

View Details

Oblock @obitopper

2 days ago

That production incident gonna turn the Slack channel into a crime scene

Mansi 👩‍💻 @MansiCodez

3 days ago

2026 software engineer starter pack: -IDE -GitHub -ChatGPT / Claude -Copilot -Slack -47 browser tabs -One production incident -One meeting that could've been an email We're not writing less code. We're just creating bugs faster. 💀