It helps that Codex is so much slower than Anthropic models, a 4.5 hours Codex session might as well be a 2 hour Claude Code one. I use both extensively FWIW.
- Good automated tests which the coding agent can run. I love pytest for this - one of my projects has 1500 tests and Claude Code is really good at selectively executing just tests relevant to the change it is making, and then running the whole suite at the end
- Give them the ability to interactively test the code they are writing too. Notes on how to start a development server (for web projects) are useful, then you can have them use Playwright or curl to try things out
- I'm having great results from maintaining a GitHub issues collection for projects and pasting URLs to issues directly into Claude Code
- I actually don't think documentation is too important: LLMs can read the code a lot faster than you to figure out how to use it. I have comprehensive documentation across all of my projects but I don't think it's the helpful for the coding agents, though they are good at helping me spot if it needs updating.
- Linters, type checkers, auto-formatters - give coding agents helpful tools to run and they'll use them.
For the most part anything that makes a codebase easier for humans to maintain turns out to help agents as well.
> I actually don't think documentation is too important
Trying to maintain a documentation was actually problematic in a recent project. I made some big changes that were not immediately updated in the documentation; then, in a new session, Claude Code was always referring to the stale markdown documentation file and getting confused about the codebase.
I just have a /fix-github-issue command that tells Claude to fix issue #xx with some extra definition of done stuff (test + build + update docs before finishing).
I also have a few other /commands for preparing and labeling issues.
Documentation by itself isn't useful (the human-readable kind), but for larger projects it's really useful to have a tree-style description on what is in each directory and a short description on where everything is in general. And this file is specifically LLM-optimised without any extra prose or formatting.
For larger projects you're wasting context if the LLM has to poke around in the dark and guess based on file/class names where everything is.
Regarding documentation, I’d love to hear you take on creating the CLAUDE.md/AGENTS.md type of files using the tools themselves, typically via some /init command. I think Claude Code, Codex and Copilot CLI all have a similar thing.
Those often seem to generate a snapshot of the current state of the codebase that to me seem to be just begging to get out of date, often with references to specific files. I sometimes start out with them and remove a bunch of stuff, or I just start empty and add things as they appear to be needed.
I have to admit I don't actually use those yet - I've been enjoying learning how to succinctly prompt in a way that gets the right results without them.
I've experimented with having LLMs write snippets for them - things like "Based on this session, update AGENTS.md with notes about running the tests" - but my stuff tends to be simple enough that I can just say "Run tests with pytest" and it will know what to do.
I did just confirm that you can use efficiently both AGENTS.md and CLAUDE.md in the same repo by having a CLAUDE.md with a single line:
My issue with that is that I don't really have any Python tooling installed globally (old trauma with overlapping package versions), each project is in an uv venv.
So every time an LLM starts firing blind they'll go python -> python3 -> damnit -> uv run python, wasting precious tokens.
That's why I have explicit instructions on how to use uv to manage things in AGENTS.md
Sysco isn't the only one, but it is one of the few.
If you start looking at the distribution centers of these companies and the competitors, you get a pretty clear picture of how concentrated things are.
The drop off for the distribution centers of the top 3 is stark. [1]
An astute and accurate observation. However, there is no numeric target set in the mandate you allude to:
"The Federal Reserve was created by Congress in 1913 to provide the nation with a safer, more
flexible, and more stable monetary and financial system. In 1977, Congress amended the Federal
Reserve Act (FRA) to provide greater clarity about the goals of monetary policy. The amended FRA
directs the Board of Governors and the FOMC to conduct monetary policy “so as to promote effectively the goals of maximum employment, stable prices, and moderate long-term interest rates.” [https://www.federalreserve.gov/aboutthefed/files/the-fed-exp...]
What he is doing is counter to the Fed charter, but if you're pro-Capital, you like some unemployment because it disciplines Labor.
> The chair's main responsibility is to carry out the mandate of the Fed, which is to promote the goals of maximum employment, stable prices, and moderate long-term interest rates.
> The chair's main responsibility is to carry out the mandate of the Fed, which is to promote the goals of maximum employment, stable prices, and moderate long-term interest rates
You’re correctly quoting your source. But this is crap, as their source [1] makes no reference to “moderate long-term interest rates”.
The Fed is mandated to promote “maximum employment” and “stable prices” [2]. (It defines the former as “the highest level of employment or lowest level of unemployment that the economy can sustain while maintaining a stable inflation rate.”) If inflation is unstable, the economy is above maximum employment.
Isn't the idea that maximum employment, stable prices, and moderate long-term interest rates are somewhat in tension with each other though? Which would mean the mandate is to balance those three things – e.g. maximize employment to the extent possible while maintaining stable prices and moderate interest rates.
That hasn’t been my experience, perhaps it’s state to state. I’ve been stopped without ID and had no problem. I’ve even boarded a plane through TSA without my license.
Slightly more than half have CoN laws and other states have a number of restrictions of facility construction that complicate building smaller clinics.
To add... The missing scrollbars feel painful to me. I don't even want them for scrolling, really. I want the visual reference for:
"How much content is in this window"
and
"Where am I in that content"
I still use it daily but it means I have to switch tools for certain things, and reading log files or log output is one of the more common reasons I switch.
I should probably look into trying to get the scrollback info into my statusline, No idea how easy or hard that is - so if someone has done it, feel free to shoot me pointers.
Centralized logging, a la Splunk is the gold (with what they charge, probably platinum) standard here. Watching someone who knows what they're doing at a Splunk console for a distributed system they understand is a thing of beauty. It's just a shame they charge so much.
Same here, scroll bars are literally the one missing feature keeping me from making Ghostty my daily driver. But the release notes roadmap mentions they will add them for 1.3. Waiting with anticipation.
Maybe I'm misunderstanding your request but do you not pipe outputs into a pager? (Less, more, batcat, etc) I feel like I'm probably not understanding the context though.
I'm still not quite sure what you mean. You can still {|,>} output from a program, so less and grep still help.
Or do you mean when you inevitably forget? Well then yeah you're at the mercy of your terminal but as others mentioned ghostty has a hack to help as well as some other terminals. But this should also help reinforce why you should pipe more often and write to files (or tee). It sucks but not making the same mistakes in the future and learning better habits will help you write better code and use better practices.
But that's the age old problem of "you can't analyze the data you didn't record" and that's a footgun you'll experience in every programming language, every experiment, and across many parts of life. Better to record and throw it away than not record and regret it.
It IS recorded. It's right there in scrollback (Literally the default buffer to record). It's easily accessible with most tooling, including nice scrolling, mouse support, find/search, etc...
Except in Ghostty, it's not so accessible. No find, no scrollbars.
Although deciding when to do that would be easier if I had a better indicator for just how much scrollback content exists. Ex - if it's 3 pages... I'll just scroll through it. If it's 3000 pages... time to dump to file.
So no - by default I use a pager... just about never. Why would I when I have absolutely everything in scrollback by default 99.9% of the time?
---
Don't confuse your preferences with "correct" :P
I'm doing just fine with code and best practices, I'm simply stating that this is a rough edge on an otherwise lovely tool.
In the latest version Shift + ctrl + alt + j copies the entire screen and opens it in what I assume is $EDITOR (maybe this can be configured). Or just shift + ctrl + j to copy the screen to temp file and pate the path. Either way, it then just becomes configuring it to use a pager by default, or `shift+ctrl+j` and then | less. This is similar to how kitty does search, by essentially outsourcing it to an external pager. I was put off by it at first, but it's very much in the UNIX philosophy.
There's a few things like this but I'll give a pass because how new it is. Were this a mature project I'd be upset at incompetence, but given how young it is I think it's too early to be harsh.
Plus, they're very open about what they're doing and prioritizing. As another commenter said, it's coming soon. For the rest, open feature requests, you might have needs that others didn't think of or even realize they needed
I find the search through a file actually makes sense.
If you don't know, you can send the log to a file, and open that file to look through it. More powerful than just a search next, as you can have instance counts, search with regexp and all the bells and whistles and it virtually stops the logging.
At first I thought the same as you, now I've become quite partial to this concept. I hope they don't remove this.
I honestly didn’t even realize it was missing until I was reading this thread, cause I’m always in a zellij session which has a “search in default editor” key.
Pop the whole scrollback in helix, where I can search, select, jump around, paste stuff into a scratch buffer. It’s slick. It’s got a normal search too. But yeah I haven’t used a built in emulator search in a while I guess!
