nethum.dev

nethum but dev

[nethum.dev @nethumdev]

@JackEllis Have you heard of computer?
No what’s that?

You have no idea how much of a bubble we’re in

[nethum.dev @nethumdev]

@aeroxzero @herdrdev One line in AGENTS.md and CLAUDE.md does this

[nethum.dev @nethumdev]

Whoever said use the best models for orchestrating was wrong

[nethum.dev @nethumdev]

@avrldotdev GitHub sends webhooks for certain events, you can track events related to whatever you are working on, have something that listens for these events. Additionally If curr SHA != expected SHA find the diff then report to the agent (automated) and force a re-plan?

[nethum.dev @nethumdev]

@cheatyyyy Let me guess, opus 5

[nethum.dev @nethumdev]

@zwurv @alexeheath Threads has a big international community oddly enough, have friends from Eastern Europe etc. that are on threads

[nethum.dev @nethumdev]

@jetfam @saad_mak4 @StevFredA @ClaudeDevs We all different experiences, look forward to seeing yours 👍

[nethum.dev @nethumdev]

Ohhh so that’s why I got a phone call asking me how to solve the Navier-Stokes problem.

[nethum.dev @nethumdev]

@saad_mak4 @StevFredA @ClaudeDevs You also shouldn’t compact your sessions too often (I don’t do it all). @kunchenguid has a good write up on why

[nethum.dev @nethumdev]

@0xmatti @danielbergholz Whoops

[nethum.dev @nethumdev]

@saad_mak4 @StevFredA @ClaudeDevs With skills, some skills can be invoked when not needed and might be bloated (just a couple months ago 100-200+ line skills were fine) nowadays, skills can be just 20-50 line and agents will implement just fine. Bloated skills that are invoked too often also eat up usage.

[nethum.dev @nethumdev]

@saad_mak4 @StevFredA @ClaudeDevs Check .Claude if you have a CLAUDE.md or AGENTS.md with an excessive amount of information then all that information is loaded every session (eating up usage)

[nethum.dev @nethumdev]

@iannuttall Herdr has something called herdr skill just search it up. Then mess around with agents spawning other agents. You can get Claude code agents to spawn codex agents all in their native harness

[nethum.dev @nethumdev]

@StevFredA @ClaudeDevs This might be a skill issue, I’m not an anthropic apologist but chances are your config is bloated with skills and memory that consume usage before ever doing anything. I have 5x and push nearly 200 contributions a day, and I don’t hit my weekly usage (mostly using fable)

[nethum.dev @nethumdev]

@gethackteam it’s to assess the models spatial reasoning abilities and harder tool use. 3d maps and models are objectively harder to make than simple sites for a number of reasons

[nethum.dev @nethumdev]

@VictorTaelin Massive pre train + recurrent depth reasoning

[nethum.dev @nethumdev]

@_AbdullahIslam @OmedVibeCodes Eh fable and sol implementer works well for me. Sol’s increased efficiency makes it very viable as subagents ($100 Claude and $100 Codex)

[nethum.dev @nethumdev]

@AminiAbdellah @kunchenguid I agree with you, I’m not so hardstuck in my ways that I’m not willing to adapt to the situation.