Posted by benoitg 6 days ago
If I wanted to recommend to someone the min-maxed, highest density thing they could add to their prompt, it would simply be the time your current prompt appeared + the amount of time the last command you ran took.
These two pieces of information together make it very easy for you (or your local sysadmin (or an LLM looking over your digital shoulder)) to piece together a log of exactly what happened when. This kind of psuedo-non-repudiation can be invaluable for debugging sessions when you least expect it.
This was a tip I distilled from Michael W. Lucas's Networking for System Administrators a few years ago, which remains my preferred recommendation for any developers looking to learn just enough about networking to not feel totally lost when talking to an actual network engineer.
Bonus nerd points if you measure time in seconds since the UNIX epoch. Very easy and fast to run time delta calculations if you do that:
[0 1719242840] $ echo "foo"
[0 1719242905] $ echo "fell asleep before hitting enter" && sleep 5
[5 1719242910] $
[1]: https://github.com/hiAndrewQuinn/shell-bling-ubuntu > history | get 82076
╭─────────────────┬──────────────────╮
│ start_timestamp │ 2025-06-24 16:46 │
│ command │ mpc play │
│ cwd │ /home/work │
│ duration │ 1ms │
│ exit_status │ 0 │
╰─────────────────┴──────────────────╯
It's really nice, because it doesn't just tell you time between command executions (or rather time between commands finishing), but the actual runtime duration of the command. history | where exit_status == 127 | last | get duration
2sec 410ms
The syntax ends up very clean looking since the data going through the pipes is typed, and the error messages are top notch × Types 'duration' and 'int' are not compatible for the '-' operator.
╭─[entry #47:1:2]
1 │ (history | where exit_status == 127 | last | get duration) - 10
· ───┬─── ┬ ─┬
· │ │ ╰── int
· │ ╰── does not operate between 'duration' and 'int'
· ╰── duration
╰────
(fingers crossed hn doesn't mangle that...)In fact, I only set up Starship when I started to do more pairing. It wasn’t for my benefit as much as it was for those watching my screen and checking the work, especially when operating on prod and confirming what we wanted to run. I just load up a separate terminal app for that so I don’t have to walk people through my setup.
For example:
…/.config master on AWS_Prod (use2)
starship.toml: [aws]
format = 'on [($profile )(\($region\) )]($style)'
style = 'bold #B23D2F'
symbol = " " <- cloud symbol
[aws.region_aliases]
us-east-1 = 'use1'
us-east-2 = 'use2'
For example, I don’t expect running scripts/build.sh to modify tracked files in the repo. Seeing part of the prompt go from “” to “?2!3” (two untracked, three changed files) makes that glaringly obvious.
And the Branch is also an unintrusive reminder that you are in a path under versioncontrol.
`git status` to know git stuff. `pwd` for the current working directory, etc
I also don't use aliases like `gs` or `..`
One good thing about having a very minimal setup is that you feel at home anywhere.
It wasn't always like this. I used many, many prompts and shell tools over the decades. The only tool that stood the test of time is tmux.
I can sit down at (or ssh into) any machine and be basically just as productive, and it also turns out that I just always want to know more than nicely fits into the prompt anyways.
There's something to be said for accepting the defaults of a tool, and learning to use them well. Customization is powerful, but... I think most times it's not the right call until you're already an expert in the tool at hand.
For example if I have say 3 worktrees open in 3 seperate tmux tabs and are context switching between them (very common when reviewing multiple PRs from my devs) Sometimes i will get the tabs mixed up, which worktree is where etc and just autopilot a bunch of commands meant for one tree into a different one and its quite annoying to clean up.
The prompt has generally stopped me from doing that.
Usually, there will be from 2 to 8 panels of different sizes.
This gives me spacial short term memory: I know what each shell is by the panel position.
I can zoom on then to bring them full screen (ctrl+b z) if I'm going to do anything that requires more space, then zoom out to the panel arrangement when I'm done.
Sometimes I'll name prompts (eg `PS1='stg$ '`), specially when working with ssh, but that's rare.
What inspired me to work this way was this video on the acme editor: https://www.youtube.com/watch?v=dP1xVpMPn8M
tmux comes from BSD rather thsn GNU/Linux, or Windows
What is the default shell in OpenBSD
starship does not support it
starship init ksh
ksh is not yet supported by starship.
For the time being, we support the following shells:
* bash
* elvish
* fish
* ion
* powershell
* tcsh
* zsh
* nu
* xonsh
* cmd
Please open an issue in the starship repo if you would like to see support for ksh:
https://github.com/starship/starship/issues/new
Sort of contrary to that I really enjoy the maximalist shells. A computer should be fun to use!
So if you have
``` src/components/Button.vue src/components/ButtonGroup.vue ```
And you type `nvim s/c/G<Tab>` it’ll expand to the second file’s path.
I’m a researcher and work on small projects with 1-3 people (most of the time it’s just me prototyping stuff alone). I then tend to work on a branch for weeks at a time, so the git branch provides very little information compared to the space it takes in a prompt.
If I was switching branches every 5min, it would be useful.
I found out that with this verify-and-forget step, I work much more efficiently.
As a result, my workflow becomes independent of the machine I work on, because I become the tool, not my setup. After that point, only having a "$" at the beginning of the line is enough.
Of course everyone have their own choices, and YMMV.
Also, I somehow worked in special characters for Python and other things that get screwed up if I don’t have the right nerd font installed on the system.
Even for cases where I need to use old versions, I don’t need a reminder of that every time I run a command.
It isn't that useful but I do glance it when I'm working on dependencies and to ensure the context between a terminal session and pycharm's interpreter match.
The information doesn't cloud the prompt for me though, as it is right justified and I don't really think about time to load, as the machines are relatively recent Apple Silicon.
That way you only have to look at it when you need it, and you can also figure out what you were doing last week/month/year if necessary.
It's a personal machine and should be treated as disposable. Doing anything less is fairly irresponsible.
So sure - turn on timestamps for your ssh bastion (although it should be in the logs already...), or turn them on for the ci/cd pipeline (not that you should need them there anyways, since it should be dumping tons of timing info already).
But a personal machine? Plain ol' ">" is plenty.
Not that there's anything wrong with a maximal prompt either... I've definitely done the "configure all the powerline settings!" thing. But I also don't mind a simple ">" or "#".
You can always time your commands with "time".
Something like
export HISTFILESIZE=
export HISTSIZE=
export HISTTIMEFORMAT="[%F %T] "
shopt -s histappend
really ought to be default in bash.It's not as clear why you need it in the interactive prompt.
It sounds silly, but it has saved my butt more than once. Especially if you have bugs that e.g. only show up once per hour on the hour, and are otherwise fine.
Here's one zbell implementation, not sure it's the original but it looks like it does the trick: https://gist.github.com/oknowton/8346801
I developed the tool myself, and it's at https://git.sr.ht/~bayindirh/nudge if you feel like checking it out.
If you want to host the whole push notification infrastructure, you can look at https://ntfy.sh which also can be integrated with cURL.
int history_write_timestamps
If non-zero, timestamps are written to the history file, so they can be preserved between sessions. The default value is 0, [...]
So this isn't true by default on many machines unless it is explicitly turned on. Once you do have it on, of course, then I agree.Personally, over time, I have stopped caring too much about prompt customization. I concluded that, no matter how carefully you curate your prompt, 90% of the information shown will be irrelevant 90% of the time*. After a while, your brain will start perceiving this as visual clutter and filter it out, to the point you may even forget the information is there, right in front of your eyes.
And for the things that matter, you probably need more details than any prompt can show you. E.g. are there changes in your git branch? Ok there are, good to know, but which files have changed? Just knowing that there are changes is not really actionable information. You need to run additional commands to get actionable details.
* the numbers are completely arbitrary, but you get the picture
I played with Starship for an hour this morning - the joys of 50 person planning meetings - but ultimately uninstalled it. I did like some of its options like command timing and success/error, but all the tool versions ultimately just felt like noise. Not worth the effort to maintain a complex custom config to trim it down to what I'd want.
I agree there's a lot of noise that seems to be there by default.
For most of my career I used a very simple PS1:
export PS1="\[\033[1;32m\][\t \u@\h \w]\\$\[\033[0m\] "
timestamp, who I am, what box I'm on, where I am.I've tried prompts in the past, and they mostly annoyed me, or never showed me useful information. I've been a happy starship user for several years now. I've got the config tweaked so that it only shows me things I specifically care about. It's lightning fast.
I think these kinds of over-optimization rabbit holes are a good learning experience, but I compare it to woodworking. A woodworker just starting out will spend most of his/her time building or refining the tools they need, learning techniques, coming up with ideas/designs and testing them, etc. But eventually the point comes where you have to get Real Work done and the tools and jigs have to wait until the weekend.
Linux is still my favorite desktop OS, but these days I just run Debian and KDE because "free time" is not a thing I have anymore and I care more about getting things done than having the most optimal computing experience.
I only emit it if the prior command fails, too, so it doesn't clutter things the 90% of the time things are working.
» true
» false
(last command returned 1.)
»
I also translate signals, so I get "last command exited on SIGSEGV", or so.It's also useful the other way: when a program emits and error and exits with "success".
That was my prompt when it was written in zsh. Sort of like TFA, I've since moved to Rust:
https://github.com/thanatos/dotfiles/blob/master/zsh-prompt-...
I think (if I am reading TFA's code right) unlike the article, I'm using zsh's module functionality, so the Rust here is a .so that is loaded directly into the shell's memory. (I.e., I do not have to fork/exec a separate Rust bin to compute the prompt, though I think zsh might fork-but-not-exec for computing the prompt itself.)
The latter is, of course, somewhat more complicated in some senses. (Esp. on macOS, which work forces me to use, where dlopen(2) is just utterly insane.)
For example the following prints the exit code in green if zero, in red otherwise:
PS1='\[\e[$(($??31:32))m\]$? \[\e[39m\]'
My shell customization is largely throwing Starship in (so it looks the same on all the machines I use -- Ubuntu servers at work, macOS at home, nixOS/Fedora/etc. servers for personal use.) and a starship.toml I wrote once and now leave alone.
For a while, I tried a couple of Christmas tree prompts which included all kinds of condensed Git status and other bells and whistles, but eventually tired of them and settled on:
- Exit status of the previous command, if nonzero.
- Current time, HH:MM, 24 hour format.
- user@host, red if euid 0, green otherwise.
- Current directory, shortened if the path has three or more elements, with home directory recognition.
- Current directory, full path, echoed as hardstatus and hence appearing in the terminal window title.
- The name of the current branch if within a Git repo.
- Prompt character, dollar/hash sign.
All those elements are meaningful to me, inasmuch as I can quickly orient myself using that information and explore further if I notice anything out of the ordinary.
I'm pretty sure that megaprompt programs like Starship could produce the above, but I like obtaining a familiar prompt with a minimum of external dependencies, and so have written it all in Bash, then ported to Zsh and various Korn shells, which was quite tricky. It probably wouldn't work on Xenix 286, but anything newer has a fighting chance.
And now that you mention it, next year will be my 30th Unix-versary. Time flies... Still not a greybeard though.
I've tried prompts in the past, and they mostly annoyed me, or never showed me useful information. I've been a happy starship user for several years now. I've got the config tweaked so that it only shows me things I specifically care about. It's lightning fast.
That said my vimrc is 2 lines that i can configure manually, I don't touch bash config from Debian defaults and my fish config is vanilla save for a handful of functions because I'm a lazy. My ssh config is pretty heavily customized but mostly around what keys/usernames to default to for which hosts (see previous about lazy).
It also gives me the current branch in a directory that has git, just so I'm sure of what I'm working on-- but most of the time that's handled by whatever editor I'm using.
A few years ago I progressed to having the current directory in there.
The thought of running a child process to create my prompt every time I hit enter doesn't feel right.
I should know what git branch I'm in, and if I don't it's a simple command away.
By default my prompt is a shows me the current directory, the time, and a single character '%'. If I set something in my environment for which I need to be contextually aware - i.e if I have KUBECONFIG or OS_CLOUD - then the prompt is updated with the detail. Similar for languages - it'll automatically show me the version of Go or Python or whatever based on a few factors, all of which I can control.
The reason I love Starship is that it's made all this very, very easy to configure - instead of having to wade through arcane Zsh configuration or additional plugins, Starship makes it easy. It also adds negligible overhead to initialisation, especially when done so via evalcache [0]
Love the performance. Written in Rust and compiled to binary, it's _much_ faster than either python-based powerline, the bash-shell-based ohmybash and zshell-based ohmyzsh and spaceship.
You can use it for zsh, bash, sh, fish. but you can also use it for both MS Windows CMD and Powershell. I don't believe any other prompt tools can do all at the same time. And a single config file can control all of them on your system.
The default config is just that - a default. Too much information? you can change it. dont like icons? you can remove them.
At almost 100 modules to choose from, it's customization options are nearly limitless
My shell prompt is:
: ▶
You don’t need an entire shell prompt customisation framework to be minimal. format = """
$username\
$hostname\
$shlvl\
$directory\
$git_branch\
$git_commit\
$git_state\
$git_metrics\
$git_status\
$package\
$python\
$rust\
$env_var\
$custom\
$cmd_duration\
$jobs\
$time\
$status\
$shell\
$character"""
# clean, simple, minimal
PROMPT='%{%F{red}%}%~ %{%F{yellow}%}% › %{%F{reset_color}%}%'
Given that I do like my shiny prompt, which shows me:
The result of the last command (in green, red, or purple)
user@host:currentDirectory
current branch, if in a repo
with the last line showing summary git status, if in a repo, and background jobs, I suspect I might be their market, but I cannot see a why anywhere.(Green: Last command good, e.g., exit 0) (Red: Last command non-zero exit, with a special indicator if it was interrupted) (Purple: Last command suspended, and few other things)
It's very minimal while having useful features: - exit codes, even for pipelines - git branch, status (displayed as a dot if your tree is dirty) and ahead/behind counts - command execution time (if above some configurable threshold) - truncated/minified $CWD, always maintaining the git root's name (I sometimes like it, sometimes don't; fortunately, it's very easy to change) - current vi-like mode (I don't use that)
It's very fast and async (prompt repaints don't block your input or running commands), and totals 132 lines of fish (according to cloc[1]). It's also very customisable through variables, which can be declared as universal to instantly change on all sessions you have open.
If you're on fish and like this feature set, definitely give it a shot, or at least look at the code as a base for a bespoke prompt :P
[0]: https://github.com/jorgebucaran/hydro [1]: https://github.com/AlDanial/cloc
Git-aware prompts can't be recommended on Windows, imo.
I try to avoid emulation layers like MSYS2, as much as I'm able.
Also, yes, if git hangs on git show/git diff that sounds like an antivirus problem or a dying hard drive or the first one causing the other one.
git_status - 6ms - "[!?] "
directory - 4ms - "<redacted> "
python - 3ms - "via v3.12.9 (.venv) "
character - <1ms - " "
git_branch - <1ms - "on main "
hostname - <1ms - "<redacted> in "
If I go in to my checked out version of the linux kernel, probably the biggest git project I've got kicking around: git_status - 115ms - ""
directory - 4ms - "linux "
character - <1ms - " "
git_branch - <1ms - "on master "
hostname - <1ms - "<redacted> in "
That's typically the worst I see it.100ms optimization is a lot different for a CPU or a human brain. I'm not defending having the entire system log dumped out on every prompt but a few amenities are worth a few milliseconds computation time for a human.
Besides, I don't see how, for example , having your prompt take those 100ms to print a git branch or status breaks your "flow" yet having to type out the commands yourself and taking longer doing it doesn't.
Its a balance between bloat and and usability like so many other things, but, to me at least, being on either extreme of bloat or extreme-minimalism seems counterproductive.
FWIW, I switched from zsh default to starship and didn't notice any perceptible difference. But I certainly notice when I mess up my git commits!
Don't the layers of frameworks mean that the opposite is true.