Sandun Rathsara

I build software, automate chaos, overthink infrastructure, and occasionally break things (ethically).

[vlad/r @_architected]

Arch existed before. Hyprland existed before.

What didn’t exist was a user-friendly tutorial, a welcoming community, and AI agents running from the root folder.

I’m not praising Omarchy blindly, but I’ve been using it since its early releases – even going back to Omakub – and it has always had a friendly, helpful community that was excited to try, build, and develop new things.

All the people from so-called 'Linux Community' hating on it now could’ve done the same thing with CachyOS, Linux Mint, or any other distro. Instead, they chose gatekeeping and toxic behavior.

[Sandun Rathsara @srathsara1]

Thanks @thsottiaux

[Sandun Rathsara @srathsara1]

@RobertJBye @claudeai Have you considered giving a higher tier like $400 - 40x

[Sandun Rathsara @srathsara1]

“The Batman”

Please @danawhite

[Arman Tsarukyan UFC @ArmanUfc]

Title shot secured 👊🏼

[Sandun Rathsara @srathsara1]

@olegakbarov @opencode I’ve been using opencode for about 6 months nonstop now. Didn’t see something like this even in opencode2 beta. WTF is your problem 😂

[Sandun Rathsara @srathsara1]

@thdxr I will create an llm from scratch wasting 1000s of hours before centering a div again.

[Sandun Rathsara @srathsara1]

@thdxr Man. Finally

[Sandun Rathsara @srathsara1]

@MiaAI_lab I haven’t tried with opus, but did the same as you with luna max vs grok high. Luna always came out on top.

[Sandun Rathsara @srathsara1]

@wilsenhc @DorianDevelops @herdrdev @opencode Same combo + nvim

[Sandun Rathsara @srathsara1]

@maria_rcks Please @thdxr

[Sandun Rathsara @srathsara1]

@MiaAI_lab @thdxr can you?

[Sandun Rathsara @srathsara1]

@elonmusk @itslueul What will be the pricing for those future models

[Sandun Rathsara @srathsara1]

@AgomaMitchell Fair enough. Thanks for the feedback. will try that out.

[Sandun Rathsara @srathsara1]

Try this Prompt in your codebase.

/orchestrate a /ponytail review on the tests written in this project. Look for

1. Brittle Tests (or Low-Value Tests)
2. Mock-Heavy / Taillight-Chasing Tests
3. "Cargo Cult" Tests (or Coverage-Driven Tests)
4. Flaky Tests
5. Zombie Tests (or Obsolete Tests)

and remove those.

FYI, orchestrate & ponytail are skills. orchestrate is from my skill set called "bearings" and google for ponytail skills.

[Sandun Rathsara @srathsara1]

Grok 4.6 < GPT 5.6 Luna in every way.

[Sandun Rathsara @srathsara1]

Trying something with agent skills:

one charged word at the top, not a paragraph.

Scrupulosity. Martinet. Exhume.

See if the feeling drives the agent better than instructions.

[DHH @dhh]

It really isn't. Embrace Peterson's simple advice: Clean your room. Then clean your computer. Don't succumb to sloppy thinking, sloppy living, or sloppy computers. (It's also just gross, and disgust is an important human instinct).

[Sandun Rathsara @srathsara1]

@theo Waiting

[Sandun Rathsara @srathsara1]

Still nothing beats "ADHD" to make AI compress it's response.

`Note that I have ADHD and can't comprehend long descriptions. so shape your response to match my ability to read.`

[Sandun Rathsara @srathsara1]

Grok build cli is soo good. almost as claude code, might be even better. and grok 4.6 is intelligent enough and cost effective enough that I don't have to specify which model to spin up sub-agents on.