Nate Builds retweeted
People of pi. RFC. Already did so on Discord. Seems like pretty much everyone is OK with that. Happy to answer questions.
70
16
618
60,108
Nate Builds retweeted
AT LAST! Custom Status Lines are coming to Codex 🙌
This is just one of many highly requested features in the pipeline.
You can now run /statusline to configure exactly what shows in your terminal. Choose from 15 data points, toggle them on/off, or even reorder them.
One thing future is to be able to install custom extensions that would allow you to customize their styling or run custom scripts (like weather updates).
More updates coming soon, but this is the most mature imminent release (0.99 Alpha).
Codex team is absolutely cooking.
This is just one of many highly requested features in the pipeline.
You can now run /statusline to configure exactly what shows in your terminal. Choose from 15 data points, toggle them on/off, or even reorder them.
One thing future is to be able to install custom extensions that would allow you to customize their styling or run custom scripts (like weather updates).
More updates coming soon, but this is the most mature imminent release (0.99 Alpha).
Codex team is absolutely cooking.
13
7
152
38,869
Recent addition to claude code:
/debug
helpful for cleaning up configuration issues (run at start) or solving issues that happen during a session
/debug
helpful for cleaning up configuration issues (run at start) or solving issues that happen during a session
0
0
0
68
Nate Builds retweeted
I mentioned recently that I've been... busy. Lots of projects in the oven, in various stages of completeness.
Well, I'm now pleased to introduce one I've worked very hard on, because it's so near-and-dear to my heart:
beads_rust, or br for short. You can get it here:
N
It's a fast, minimal Rust port of @Steve_Yegge's amazing Beads project that I've built so many of my workflows around.
Discovering Beads and seeing how well it worked together with my MCP Agent Mail was a truly transformative moment in my agent coding workflows and professional life more generally.
This quickly also led to my beads_viewer (bv) project, which added another layer of analysis to beads that gives swarms of agents the insight into what beads they should work on next to de-bottleneck the development process and increase velocity.
It's beads (and mail) all the way down.
I'm very grateful for finding beads when I did and to Steve for making it. But at this point, my Agent Flywheel System is built around beads operating in a specific way.
As Steve continues evolving beads toward GasTown and beyond, our use cases have naturally diverged.
The hybrid SQLite + JSONL-git architecture that I built my tooling around (and independently mirrored in MCP Agent Mail) is being replaced with approaches better suited to Steve's vision.
Rather than ask Steve to maintain a legacy mode for my niche use case, I created this Rust port that freezes the "classic beads" architecture I depend on. The command is br to distinguish it from the original bd.
This isn't a criticism of beads; Steve's taking it in exciting directions. It's simply that my tooling needs a stable snapshot of the architecture I built around, and maintaining my own fork is the right solution for that.
Steve has given his full endorsement of this project.
Well, I'm now pleased to introduce one I've worked very hard on, because it's so near-and-dear to my heart:
beads_rust, or br for short. You can get it here:
N
It's a fast, minimal Rust port of @Steve_Yegge's amazing Beads project that I've built so many of my workflows around.
Discovering Beads and seeing how well it worked together with my MCP Agent Mail was a truly transformative moment in my agent coding workflows and professional life more generally.
This quickly also led to my beads_viewer (bv) project, which added another layer of analysis to beads that gives swarms of agents the insight into what beads they should work on next to de-bottleneck the development process and increase velocity.
It's beads (and mail) all the way down.
I'm very grateful for finding beads when I did and to Steve for making it. But at this point, my Agent Flywheel System is built around beads operating in a specific way.
As Steve continues evolving beads toward GasTown and beyond, our use cases have naturally diverged.
The hybrid SQLite + JSONL-git architecture that I built my tooling around (and independently mirrored in MCP Agent Mail) is being replaced with approaches better suited to Steve's vision.
Rather than ask Steve to maintain a legacy mode for my niche use case, I created this Rust port that freezes the "classic beads" architecture I depend on. The command is br to distinguish it from the original bd.
This isn't a criticism of beads; Steve's taking it in exciting directions. It's simply that my tooling needs a stable snapshot of the architecture I built around, and maintaining my own fork is the right solution for that.
Steve has given his full endorsement of this project.
27
14
236
65,127
I was wondering why claude code hadn't released a new version in a while, then checked npm ...
2.1 imminent? 👀
2.1 imminent? 👀
0
0
1
123
Nate Builds retweeted
One of the longest running schisms in programming is that of static vs dynamic typing. I've heard a million arguments from both sides throughout my entire career, but seen very few of them ever convinced anyone of anything. As rationalizations masquerading as reason rarely do in matters of faith. The rider will always justify the way of the elephant.
That's not to say there aren't people who've switched camps. In fact, such individuals usually shout the loudest rationalizations of all. Often with the charm of a livelong omnivore suddenly turning vegan or a traditional banker who caught a lucky ride on crypto. The shorter the faith, the brighter the flame.
Personally, I'm unashamedly a dynamic typing kind of guy. That's why I love Ruby so very much. It takes full advantage of dynamic typing to allow the poetic syntax that results in such beautiful code. To me, Ruby with explicit, static typing would be like a salad with a scoop of ice cream. They just don't go together.
I'll also confess to having embraced the evangelical position for dynamic typing in the past. To the point of suffering from a One True Proposition affliction. Seeing the lack of enthusiasm for dynamic typing as a reflection of missing education, experience, or perhaps even competence.
Oh what folly.
Like trying to convince an introvert that they'd really like parties if they'd just loosen up a bit. That actually it's really fun to be in crowded rooms, for hours on end, shouting to communicate, because how about that VIBE!
These days, I've come to appreciate the magnificence of multiplicity. Programming would be an awful endeavor if we were all confined to the same paradigm. Human nature is much too varied to accept such constraint on its creativity.
Could you imagine if all visual art had to be rendered in the style of cubism? Or realism? Or all novels written in the short, direct flavor of Hemmingway? What a bore it would all quickly be!
It would ruin the magic of programming. This unique fusion of art and engineering.
But it took a while for me to come to these conclusions. I'm a recovering solutionist. So when I see folks cross their heart in disbelief that anyone, anywhere might fancy JavaScript over TypeScript, I smile, and I remember the days when I'd recognize their zeal in the mirror.
That's not to say all matters of programming approaches boil down to equal but different mindsets. There are limits to this relativism. But dynamic vs static typing is certain within its confines. So too is functional vs object-oriented programming. Poles on both these axes have shown to deliver excellent software over the decades (and awful stuff too!).
Now people are blessed with a capacity to express themselves equally well from either end. Such ambidextrous nature seems rare, though, as evidenced by the utter disbelief so frequently expressed by either side that the other might hold a reasonable position.
I am not ambidextrous. I do not enjoy static typing, and objects animate my mind's eye. But I have come to appreciate the fact that others illuminate their creativity with just as much intensity as I do mine, using functional programming constraints and explicitly spelled out types.
As long as I never have to soil my Ruby with type hints or write all front-end code in TypeScript, I can so happily coexist with those who adore Go or can't stand JavaScript. Vive la différence!
That's not to say there aren't people who've switched camps. In fact, such individuals usually shout the loudest rationalizations of all. Often with the charm of a livelong omnivore suddenly turning vegan or a traditional banker who caught a lucky ride on crypto. The shorter the faith, the brighter the flame.
Personally, I'm unashamedly a dynamic typing kind of guy. That's why I love Ruby so very much. It takes full advantage of dynamic typing to allow the poetic syntax that results in such beautiful code. To me, Ruby with explicit, static typing would be like a salad with a scoop of ice cream. They just don't go together.
I'll also confess to having embraced the evangelical position for dynamic typing in the past. To the point of suffering from a One True Proposition affliction. Seeing the lack of enthusiasm for dynamic typing as a reflection of missing education, experience, or perhaps even competence.
Oh what folly.
Like trying to convince an introvert that they'd really like parties if they'd just loosen up a bit. That actually it's really fun to be in crowded rooms, for hours on end, shouting to communicate, because how about that VIBE!
These days, I've come to appreciate the magnificence of multiplicity. Programming would be an awful endeavor if we were all confined to the same paradigm. Human nature is much too varied to accept such constraint on its creativity.
Could you imagine if all visual art had to be rendered in the style of cubism? Or realism? Or all novels written in the short, direct flavor of Hemmingway? What a bore it would all quickly be!
It would ruin the magic of programming. This unique fusion of art and engineering.
But it took a while for me to come to these conclusions. I'm a recovering solutionist. So when I see folks cross their heart in disbelief that anyone, anywhere might fancy JavaScript over TypeScript, I smile, and I remember the days when I'd recognize their zeal in the mirror.
That's not to say all matters of programming approaches boil down to equal but different mindsets. There are limits to this relativism. But dynamic vs static typing is certain within its confines. So too is functional vs object-oriented programming. Poles on both these axes have shown to deliver excellent software over the decades (and awful stuff too!).
Now people are blessed with a capacity to express themselves equally well from either end. Such ambidextrous nature seems rare, though, as evidenced by the utter disbelief so frequently expressed by either side that the other might hold a reasonable position.
I am not ambidextrous. I do not enjoy static typing, and objects animate my mind's eye. But I have come to appreciate the fact that others illuminate their creativity with just as much intensity as I do mine, using functional programming constraints and explicitly spelled out types.
As long as I never have to soil my Ruby with type hints or write all front-end code in TypeScript, I can so happily coexist with those who adore Go or can't stand JavaScript. Vive la différence!
178
50
1,163
446,454
Nate Builds retweeted
Remember sitegeist.ai, my funky browser agent? Well, decided to make it free and OSS. It did summarize YouTube videos before it was cool!
OSS will take a bit of clean-up. If you can't wait, you can already install it from here: sitegeist.ai/install
OSS will take a bit of clean-up. If you can't wait, you can already install it from here: sitegeist.ai/install
25
25
373
49,829
Does Codex CLI have plan mode or anything that will emulate it?
I just keep telling it "give me a plan of what to do, but don't make any changes" ... kind of annoying
I just keep telling it "give me a plan of what to do, but don't make any changes" ... kind of annoying
4
0
4
237
workaround (thanks @Road_Kill11):
/approvals --> read-only mode
not exactly plan mode, and a bit clunky to switch back & forth, but it does stop codex from automatically making changes
/approvals --> read-only mode
not exactly plan mode, and a bit clunky to switch back & forth, but it does stop codex from automatically making changes
0
0
2
83
Using a mac and need to paste an image into Codex CLI?
Use ctrl-v instead of cmd-v 🫡
harkens back to the early claude code days
Use ctrl-v instead of cmd-v 🫡
harkens back to the early claude code days
1
0
10
815
Anyone else using learning mode?
I love it ❤️
I love it ❤️
I just tried out the claude code "learning" mode 🤯
use command:
/output-style learning
Claude stops in the middle of changes and ask you to complete them
It forces you to really understand what's going on
Just learned how to wire up react-hook-form for the first time 💪
5
0
12
1,514
Nate Builds retweeted
Hey guys!
CurseX is now live on Product Hunt.
Let’s hit it, and sorry in advance for polluting your feed with AI inspired shitposts 🙈
(It’s also my wife’s birthday 🥳)
Gogogogogo!!
I’d really love a retweet. If you’re into that kind of thing 😅
producthunt.com/products/curse…
CurseX is now live on Product Hunt.
Let’s hit it, and sorry in advance for polluting your feed with AI inspired shitposts 🙈
(It’s also my wife’s birthday 🥳)
Gogogogogo!!
I’d really love a retweet. If you’re into that kind of thing 😅
producthunt.com/products/curse…
29
3
41
2,329
Why are claude code subagents so bad at coding ...
but so good at other arbitrary tasks like researching?
It's because each subagent starts with a clean context.
The context required to code well is large. CC needs to build up a thorough understanding of the codebase and what you want changed. (as well as API specs, coding conventions etc)
It's hard to give the subagent enough context to do that well.
But to research something, you just need to tell it what you want to learn, the context is (relatively) tiny.
but so good at other arbitrary tasks like researching?
It's because each subagent starts with a clean context.
The context required to code well is large. CC needs to build up a thorough understanding of the codebase and what you want changed. (as well as API specs, coding conventions etc)
It's hard to give the subagent enough context to do that well.
But to research something, you just need to tell it what you want to learn, the context is (relatively) tiny.
2
0
5
843
The best developers are:
just lazy enough to want to automate everything
and
just motivated enough to actually do it
just lazy enough to want to automate everything
and
just motivated enough to actually do it
4
0
4
175
Software architecture skills are the key to effective agentic dev
Those skills come from writing a lot of code. 🧑💻
How does the next gen of builders develop them if agents do most of the writing?
Those skills come from writing a lot of code. 🧑💻
How does the next gen of builders develop them if agents do most of the writing?
6
0
10
1,428
How to try out Codex CLI if you've been using Claude Code
(and reuse your CLAUDE․md rules)
Just move your CLAUDE․md contents into AGENTS․md and @ mention AGENTS․md inside of CLAUDE․md
Codex CLI uses AGENTS․md
Claude uses CLAUDE․md (and you can embed files with @)
(and reuse your CLAUDE․md rules)
Just move your CLAUDE․md contents into AGENTS․md and @ mention AGENTS․md inside of CLAUDE․md
Codex CLI uses AGENTS․md
Claude uses CLAUDE․md (and you can embed files with @)
3
0
6
290
getting MCPs migrated is a manual migration, since codex config is done in a toml file and not json
github.com/openai/codex/b…
github.com/openai/codex/b…
0
0
1
73
Anyone know how to shrink the amount of context a MCP consumes?
(in claude code)
I love the linear MCP, but I really only use 2-3 tools out of the 24 it offers...
Ideally I'd pick the tools I want and only include those in the context or something like that
(in claude code)
I love the linear MCP, but I really only use 2-3 tools out of the 24 it offers...
Ideally I'd pick the tools I want and only include those in the context or something like that
That's good enough for now, but I'd also like to do something about the Linear & Playwright MCPs, since they consume a lot of context
Will save that for another day ...
3
0
0
216