Sam L

97 posts

Sam L banner
Sam L

Sam L

@samelldev

building KruxOS - the operating system for AI agents.

Katılım Şubat 2022
125 Takip Edilen964 Takipçiler
Sam L
Sam L@samelldev·
made a little tool while reorganising my projects this week — cc-move: move/rename a Claude Code project dir without losing chat history, memory, or --resume functionality has been a pain a few times losing access to an existing conversation github.com/samellgit/cc-m…
English
0
0
2
13
Sam L
Sam L@samelldev·
sometimes i have a great long conversation with Claude Code/Codex agent and they are amazing and then we compact and they become my worse enemy
English
1
1
4
31
Sam L
Sam L@samelldev·
what's the most effective long term memory system you've used for AI agents? link it below 👇
English
0
0
3
12
Sam L
Sam L@samelldev·
never ask 2 people for design feedback, always ask an odd amount - then you'll get a majority 😂
English
0
0
1
13
Thariq
Thariq@trq212·
HTML is the new markdown. I've stopped writing markdown files for almost everything and switched to using Claude Code to generate HTML for me. This is why.
Thariq@trq212

x.com/i/article/2052…

English
887
1K
12.1K
4.3M
Ryan Carson
Ryan Carson@ryancarson·
Sheesh y'all - building new features with Claude Design is just SO freaking good. One huge problem though: they need to add pay-as-you-go pricing once you hit your quota.
Ryan Carson tweet media
English
22
3
141
15.1K
Sam L
Sam L@samelldev·
@LLMJunky i can't lie... I was expecting Mona Lisa or something 😅
English
1
0
2
189
Jaytel
Jaytel@Jaytel·
4.7 is completely unusable
English
449
173
5.9K
1M
Vaibhav Kumar
Vaibhav Kumar@vaibhavk97·
Every frontier model has its own UI taste, or lack thereof. Built a game to see if you can tell them apart.
English
51
24
545
75.3K
Sam L
Sam L@samelldev·
@naman_307 "spawn a sub agent to diagnose <insert failure>"
English
0
0
1
46
Naman
Naman@naman_307·
Claude stuck in a loop. Same fix. Same error. 4 iterations deep. You've burned 80K tokens going nowhere. What I do now, in this exact order: 1. Stop asking it to fix. Ask it to diagnose. 2. /clear, then paste only the failing 20 lines. 3. New session. Fresh context. Almost always works. Yours?
English
2
0
8
311
Sam L
Sam L@samelldev·
made a Just Chat mode in Krux a couple of weeks ago which strips tool use, found myself using it more and more and because KruxChat is model agnostic I can have a conversation with every model, in the same chat window, and they all have the full context
English
0
1
4
58
Sigrid Jin 🌈🙏
Sigrid Jin 🌈🙏@realsigridjin·
codex context forking is an cheat code claude can't do this set fork_context=true > take snapshot of exact state and spins up independent sub agents > run them in parallel eg explorer agent reads the codebase, worker agent writes code explicitly prompt fork subagent try it ngl
English
12
5
155
9.8K
Sam L
Sam L@samelldev·
/goal in Codex is making me lazy prompt still need to test full specs with it will do some big tests inside Krux this week to see how far it can be pushed
English
0
0
2
45
Sam L retweetledi
Tibo
Tibo@thsottiaux·
/goal might be the most consequential thing we have shipped in codex The value of good instructions has never been higher.
English
363
185
4.9K
376.2K
Sam L
Sam L@samelldev·
@thsottiaux @steipete how does it deal with auto compaction over multiple days? is there a recommended way to set out a structure for it at the start to keep it progressing on the right track?
English
0
0
0
1.3K
Tibo
Tibo@thsottiaux·
You can now keep codex going for days. With GPT-5.5 it will build an entire OS kernel for you if you ask, or find critical bugs in a codebase, or optimize your database schemas, or… the options are endless.
Felipe Coury 🦀@fcoury

/goal also lands in Codex CLI 0.128.0. Our take on the Ralph loop: keep a goal alive across turns. Don't stop until it's achieved. Built by my co-worker and OpenAI mentor Eric Traut, aka the Pyright guy. One of the GOATs I get to work with daily.

English
336
255
5.4K
704.5K
Greg Kamradt
Greg Kamradt@GregKamradt·
When OAI employees say, “I let codex run all night…” What framework they use? How do you set up the task so it has enough work for 8 hours?
English
298
18
1.8K
337.9K
Sam L retweetledi
Richard Wu
Richard Wu@0xrwu·
If you think about it 5 hour limits on your Codex/Claude plan is basically streaks for nerds.
English
1
1
10
1.2K
Sam L
Sam L@samelldev·
@KingBootoshi ok, i'm feeling the exact same 4.7 taking a year to do anything i found reducing the effort did help with speed but also negative on the results. ran a side by side review of our codebase with opus 4.7 and gpt5.5 with same prompt and effort level 5.5 was better and faster.
English
0
0
0
55
BOOTOSHI 👑
BOOTOSHI 👑@KingBootoshi·
@samelldev i was on high effort. medium is not that much faster though. my 5.5 is also on high effort by default
English
1
0
1
206
BOOTOSHI 👑
BOOTOSHI 👑@KingBootoshi·
using codex gpt 5.5 and getting used to its speed then going to claude code 4.7 sucks ass, why is 4.7 so slow in comparison ??? i can't do slow LLMs anymore. i need fast responses. i need to give my instant feedback and iteration. i have too much ADHD to prompt and wait MAN
English
26
3
102
6.7K

Keşfet

@trq212 @ryancarson @LLMJunky @Quentin_Adt @Jaytel @vaibhavk97 @naman_307 @realsigridjin