just is a handy way to save and run project-specific commands.

This readme is also available as a book. The book reflects the latest release, whereas the readme on GitHub reflects latest master.

(中文文档在 这里, 快看过来!)

Commands, called recipes, are stored in a file called justfile with syntax inspired by make:

You can then run them with just RECIPE:

$ just test-all
cc *.c -o main
./test --all
Yay, all your tests passed!

just has a ton of useful features, and many improvements over make:

• just is a command runner, not a build system, so it avoids much of make's complexity and idiosyncrasies. No need for .PHONY recipes!

• Linux, macOS, Windows, and other reasonable unixes are supported with no additional dependencies. (Although if your system doesn't have an sh, you'll need to choose a different shell.)

• Errors are specific and informative, and syntax errors are reported along with their source context.

• Recipes can accept command line arguments, including flags and options.

• just has a rich expression language and many built-in-functions.

• Wherever possible, errors are resolved statically. Unknown recipes and circular dependencies are reported before anything runs.

• just loads .env files, making it easy to populate environment variables.

• Recipes can be listed from the command line.

• Command line completion scripts are available for most popular shells.

• Recipes can be written in arbitrary languages, like Python or Node.js.

• just can be invoked from any subdirectory, not just the directory that contains the justfile.

• justfiles can be organized into multiple files using modules and imports.

• And much more!

If you need help with just, please feel free to open an issue or ping me on Discord. Feature requests and bug reports are always welcome!

Just can be installed using your favorite package manager, by downloading pre-built binaries, or building from source with cargo install just.

just should run on any system with a reasonable sh, including Linux, macOS, and the BSDs.

On Windows, just works with the sh provided by Git for Windows, GitHub Desktop, or Cygwin. After installation, sh must be available in the PATH of the shell you want to invoke just from.

If you'd rather not install sh, you can use the shell setting to use the shell of your choice.

Like PowerShell:

# use PowerShell instead of sh:
set shell := ["powershell.exe", "-c"]

hello:
  Write-Host "Hello, world!"

…or cmd.exe:

# use cmd.exe instead of sh:
set shell := ["cmd.exe", "/c"]

list:
  dir

You can also set the shell using command-line arguments. For example, to use PowerShell, launch just with --shell powershell.exe --shell-arg -c.

(PowerShell is installed by default on Windows 7 SP1 and Windows Server 2008 R2 SP1 and later, and cmd.exe is quite fiddly, so PowerShell is recommended for most Windows users.)

Pre-built binaries for Linux, macOS, and Windows can be found on the releases page.

You can use the following command on Linux, macOS, or Windows to download the latest release, just replace DEST with the directory where you'd like to put just:

curl --proto '=https' --tlsv1.2 -sSf https://just.systems/install.sh | bash -s -- --to DEST

For example, to install just to ~/bin:

# create ~/bin
mkdir -p ~/bin

# download and extract just to ~/bin/just
curl --proto '=https' --tlsv1.2 -sSf https://just.systems/install.sh | bash -s -- --to ~/bin

# add `~/bin` to the paths that your shell searches for executables
# this line should be added to your shell's initialization file,
# e.g. `~/.bashrc` or `~/.zshrc`
export PATH="$PATH:$HOME/bin"

# just should now be executable
just --help

Note that install.sh may fail on GitHub Actions, or in other environments where many machines share IP addresses. install.sh calls GitHub APIs in order to determine the latest version of just to install, and those API calls are rate-limited on a per-IP basis. To make install.sh more reliable in such circumstances, pass a specific tag to install with --tag.

Another way to avoid rate-limiting is to pass a GitHub authentication token to install.sh as an environment variable named GITHUB_TOKEN, allowing it to authenticate its requests.

Releases include a SHA256SUM file which can be used to verify the integrity of pre-built binary archives.

To verify a release, download the pre-built binary archive along with the SHA256SUM file and run:

shasum --algorithm 256 --ignore-missing --check SHA256SUMS

just can be installed on GitHub Actions in a few ways.

Using package managers pre-installed on GitHub Actions runners on macOS with brew install just, and on Windows with choco install just.

With extractions/setup-just:

- uses: extractions/setup-just@v3
  with:
    just-version: 1.5.0  # optional semver specification, otherwise latest

Or with taiki-e/install-action:

- uses: taiki-e/install-action@just

just is available as a Docker image from the GitHub Container Registry.

To copy just into a Docker image, add the following line to your Dockerfile:

COPY --from=ghcr.io/casey/just:latest /just /usr/local/bin/

After copying, just may also be used as part of a docker build:

RUN just

An RSS feed of just releases is available here.

just-install can be used to automate installation of just in Node.js applications.

just is a great, more robust alternative to npm scripts. If you want to include just in the dependencies of a Node.js application, just-install will install a local, platform-specific binary as part of the npm install command. This removes the need for every developer to install just independently using one of the processes mentioned above. After installation, the just command will work in npm scripts or with npx. It's great for teams who want to make the setup process for their project as easy as possible.

For more information, see the just-install README file.

The just repository includes a flake.nix that defines a nix flake, allowing you to use just as an input to another flake:

{
  inputs = {
    just.url = "github:casey/just";
  }

  outputs = {self, nixpkgs, just}: {
  }
}

With the release of version 1.0, just features a strong commitment to backwards compatibility and stability.

Future releases will not introduce backwards incompatible changes that make existing justfiles stop working, or break working invocations of the command-line interface.

This does not, however, preclude fixing outright bugs, even if doing so might break justfiles that rely on their behavior.

There will never be a just 2.0. Any desirable backwards-incompatible changes will be opt-in on a per-justfile basis, so users may migrate at their leisure.

Features that aren't yet ready for stabilization are marked as unstable and may be changed or removed at any time. Using unstable features produces an error by default, which can be suppressed by passing the --unstable flag, set unstable, or setting the environment variable JUST_UNSTABLE to any value other than false, 0, or the empty string.

If you use features of just which require a particular version, you may use the minimum-version^1.55.0 setting to make it an error to use older versions of just:

set minimum-version := '1.55.0'

If just encounters a minimum version greater than its own version, it will print an error message with the required version, which is hopefully better than the confused error message it would have otherwise produced.

The minimum-version setting should be placed at the top of the justfile, before any usage of the new feature that it guards.

Any features which change the lexer in forward-incompatible ways will still produce an unhelpful error message, as the minimum version check is implemented in the parser, which runs after the lexer.

justfile syntax is close enough to make that you may want to tell your editor to use make syntax highlighting for just.

Vim version 9.1.1042 or better and Neovim version 0.11 or better support Justfile syntax highlighting out of the box, thanks to pbnj.

The vim-just plugin provides syntax highlighting for justfiles.

Install it with your favorite package manager, like Plug:

call plug#begin()

Plug 'NoahTheDuke/vim-just'

call plug#end()

Or with Vim's built-in package support:

mkdir -p ~/.vim/pack/vendor/start
cd ~/.vim/pack/vendor/start
git clone https://github.com/NoahTheDuke/vim-just.git

tree-sitter-just is an Nvim Treesitter plugin for Neovim.

just-mode provides syntax highlighting and automatic indentation of justfiles. It is available on MELPA as just-mode.

justl provides commands for executing and listing recipes.

You can add the following to an individual justfile to enable make mode on a per-file basis:

# Local Variables:
# mode: makefile
# End:

An extension for VS Code is available here.

Unmaintained VS Code extensions include skellock/vscode-just and sclu1034/vscode-just.

A plugin for JetBrains IDEs by linux_china is available here.

Kakoune supports justfile syntax highlighting out of the box, thanks to TeddyDD.

Helix supports justfile syntax highlighting out-of-the-box since version 23.05.

The Just package by nk9 with just syntax and some other tools is available on PackageControl.

Micro supports Justfile syntax highlighting out of the box, thanks to tomodachi94.

The zed-just extension by jackTabsCode is available on the Zed extensions page.

Feel free to send me the commands necessary to get syntax highlighting working in your editor of choice so that I may include them here.

just-lsp provides a language server protocol implementation, enabling features such as go-to-definition, inline diagnostics, and code completion.

just-mcp provides a model context protocol adapter to allow LLMs to query the contents of justfiles and run recipes.

See the installation section for how to install just on your computer. Try running just --version to make sure that it's installed correctly.

For an overview of the syntax, check out this cheatsheet.

Once just is installed and working, create a file named justfile in the root of your project with the following contents:

recipe-name:
  echo 'This is a recipe!'

# this is a comment
another-recipe:
  @echo 'This is another recipe.'

When you invoke just, it looks for a file named justfile in the current directory and upwards, so you can invoke it from any subdirectory of your project.

The search for a justfile is case insensitive, so any case, like Justfile, JUSTFILE, or JuStFiLe, will work. just will also look for files with the name .justfile, in case you'd like to hide a justfile.

By default, running just with no arguments runs the first recipe in the justfile:

$ just
echo 'This is a recipe!'
This is a recipe!

One or more arguments specify the recipe(s) to run:

$ just another-recipe
This is another recipe.

just prints each command to standard error before running it, which is why echo 'This is a recipe!' was printed. This is suppressed for lines starting with @, which is why echo 'This is another recipe.' was not printed.

Recipes stop running if a command fails. Here cargo publish will only run if cargo test succeeds:

publish:
  cargo test
  # tests passed, time to publish!
  cargo publish

Recipes can depend on other recipes. Here the test recipe depends on the build recipe, so build will run before test:

build:
  cc main.c foo.c bar.c -o main

test: build
  ./test

sloc:
  @echo "`wc -l *.c` lines of code"

$ just test
cc main.c foo.c bar.c -o main
./test
testing… all tests passed!

Recipes without dependencies will run in the order they're given on the command line:

$ just build sloc
cc main.c foo.c bar.c -o main
1337 lines of code

Dependencies will always run first, even if they are passed after a recipe that depends on them:

$ just test build
cc main.c foo.c bar.c -o main
./test
testing… all tests passed!

Recipes may depend on recipes in submodules:

mod foo

baz: foo::bar

A variety of justfiles can be found in the examples directory and on GitHub.

When just is invoked without a recipe, it runs the recipe with the [default] attribute, or the first recipe in the justfile if no recipe has the [default] attribute.

This recipe might be the most frequently run command in the project, like running the tests:

test:
  cargo test

You can also use dependencies to run multiple recipes by default:

default: lint build test

build:
  echo Building…

test:
  echo Testing…

lint:
  echo Linting…

If no recipe makes sense as the default recipe, you can use default-list^1.52.0 to list the available recipes instead:

set default-list := true

Recipes may have parameters. Here recipe build has a parameter called target:

build target:
  @echo 'Building {{target}}…'
  cd {{target}} && make

To pass arguments on the command line, put them after the recipe name:

$ just build my-awesome-project
Building my-awesome-project…
cd my-awesome-project && make

To pass arguments to a dependency, put the dependency in parentheses along with the arguments:

default: (build "main")

build target:
  @echo 'Building {{target}}…'
  cd {{target}} && make

Variables can also be passed as arguments to dependencies:

target := "main"

_build version:
  @echo 'Building {{version}}…'
  cd {{version}} && make

build: (_build target)

A command's arguments can be passed to a dependency by putting the dependency in parentheses along with the arguments:

build target:
  @echo "Building {{target}}…"

push target: (build target)
  @echo 'Pushing {{target}}…'

Parameters may have default values:

default := 'all'

test target tests=default:
  @echo 'Testing {{target}}:{{tests}}…'
  ./test --tests {{tests}} {{target}}

Parameters with default values may be omitted:

$ just test server
Testing server:all…
./test --tests all server

Or supplied:

$ just test server unit
Testing server:unit…
./test --tests unit server

Default values may be arbitrary expressions, but expressions containing the +, &&, ||, or / operators must be parenthesized:

arch := "wasm"

test triple=(arch + "-unknown-unknown") input=(arch / "input.dat"):
  ./test {{triple}}

The last parameter of a recipe may be variadic, indicated with either a + or a * before the argument name:

backup +FILES:
  scp {{FILES}} me@server.com:

Variadic parameters prefixed with + accept one or more arguments and expand to a string containing those arguments separated by spaces:

$ just backup FAQ.md GRAMMAR.md
scp FAQ.md GRAMMAR.md me@server.com:
FAQ.md                  100% 1831     1.8KB/s   00:00
GRAMMAR.md              100% 1666     1.6KB/s   00:00

Variadic parameters prefixed with * accept zero or more arguments and expand to a string containing those arguments separated by spaces, or an empty string if no arguments are present:

commit MESSAGE *FLAGS:
  git commit {{FLAGS}} -m "{{MESSAGE}}"

Variadic parameters can be assigned default values. These are overridden by arguments passed on the command line:

test +FLAGS='-q':
  cargo test {{FLAGS}}

The number of arguments a variadic parameter accepts may be limited with the [arg(ARG, min=MIN)] and [arg(ARG, max=MAX)] attributes^1.56.0, which require lists to be enabled:

set unstable
set lists

[arg('FILES', min='2', max='4')]
backup +FILES:
  scp {{FILES}} me@server.com:

min and max also apply to default values.

{{…}} substitutions may need to be quoted if they contain spaces. For example, if you have the following recipe:

search QUERY:
  lynx https://www.google.com/?q={{QUERY}}

And you type:

$ just search "cat toupee"

just will run the command lynx https://www.google.com/?q=cat toupee, which will get parsed by sh as lynx, https://www.google.com/?q=cat, and toupee, and not the intended lynx and https://www.google.com/?q=cat toupee.

You can fix this by adding quotes:

search QUERY:
  lynx 'https://www.google.com/?q={{QUERY}}'

Parameters prefixed with a $ will be exported as environment variables:

foo $bar:
  echo $bar

Parameters may be constrained to match regular expression patterns using the [arg("name", pattern=PATTERN)] attribute^1.45.0:

[arg('n', pattern='\d+')]
double n:
  echo $(({{n}} * 2))

The value of pattern may be a const expression^1.55.0.

A leading ^ and trailing $ are added to the pattern, so it must match the entire argument value.

You may constrain the pattern to a number of alternatives using the | operator:

[arg('flag', pattern='--help|--version')]
info flag:
  just {{flag}}

Regular expressions are provided by the Rust regex crate. See the syntax documentation for usage examples.

Usage information for a recipe may be printed with the --usage subcommand^1.46.0:

$ just --usage foo
Usage: just foo [OPTIONS] bar

Arguments:
  bar

Help strings may be added to arguments using the [arg(ARG, help=HELP)] attribute:

[arg("bar", help="hello")]
foo bar:

The value help may be a const expression^1.55.0.

$ just --usage foo
Usage: just foo bar

Arguments:
  bar hello

Recipe parameters are positional by default.

In this justfile:

@foo bar:
  echo bar={{bar}}

The parameter bar is positional:

$ just foo hello
bar=hello

The [arg(ARG, long=OPTION)]^1.46.0 attribute can be used to make a parameter a long option.

In this justfile:

[arg("bar", long="bar")]
foo bar:

The parameter bar is given with the --bar option:

$ just foo --bar hello
bar=hello

Options may also be passed with --name=value syntax:

$ just foo --bar=hello
bar=hello

The value of long may be omitted, in which case the option defaults to the name of the parameter. With the following justfile, bar may be passed with --bar:

[arg("bar", long)]
foo bar:

The [arg(ARG, short=OPTION)]^1.46.0 attribute can be used to make a parameter a short option.

In this justfile:

[arg("bar", short="b")]
foo bar:

The parameter bar is given with the -b option:

$ just foo -b hello
bar=hello

The value of short may be omitted, in which case the option defaults to the first character of the name of the parameter. With the following justfile, bar may be passed with -b:

[arg("bar", short)]
foo bar:

If a parameter has both a long and short option, it may be passed using either.

Multiple short options may be combined^1.55.0, for example -abc is equivalent to -a -b -c. A short option which takes a value may appear last, for example -abcd VALUE.

Variadic * and + parameters may be options, in which case the option is repeatable, with each occurrence contributing one value:

[arg('file', long)]
backup +file:
  scp {{file}} me@server.com:

$ just backup --file FAQ.md --file GRAMMAR.md
scp FAQ.md GRAMMAR.md me@server.com:

As with positional variadic parameters, + options must be passed at least once, whereas * options may be omitted.

The [arg(ARG, value=VALUE, …)]^1.46.0 attribute can be used with long or short to make a parameter a flag which does not take a value. VALUE may be an expression^1.54.0.

In this justfile:

[arg("bar", long="bar", value="hello")]
foo bar:

The parameter bar is given with the --bar option, but does not take a value, and instead takes the value given in the [arg] attribute:

$ just foo --bar
bar=hello

This is useful for unconditionally requiring a flag like --force on dangerous commands.

A flag is optional if its parameter has a default:

[arg("bar", long="bar", value="hello")]
foo bar="goodbye":

Causing it to receive the default when not passed in the invocation:

$ just foo
bar=goodbye

Given this justfile:

foo argument:
  touch {{argument}}

The following command will create two files, some and argument.txt:

$ just foo "some argument.txt"

The user's shell will parse "some argument.txt" as a single argument, but when just replaces touch {{argument}} with touch some argument.txt, the quotes are not preserved, and touch will receive two arguments.

There are a few ways to avoid this: quoting, positional arguments, and exported arguments.

Quotes can be added around the {{argument}} interpolation:

foo argument:
  touch '{{argument}}'

This preserves just's ability to catch variable name typos before running, for example if you were to write {{argument}}, but will not do what you want if the value of argument contains single quotes.

The positional-arguments setting causes all arguments to be passed as positional arguments, allowing them to be accessed with $1, $2, …, and $@, which can then be double-quoted to avoid further splitting by the shell:

set positional-arguments

foo argument:
  touch "$1"

This defeats just's ability to catch typos, for example if you type $2 instead of $1, but works for all possible values of argument, including those with double quotes.

All arguments are exported when the export setting is set:

set export

foo argument:
  touch "$argument"

Or individual arguments may be exported by prefixing them with $:

foo $argument:
  touch "$argument"

This defeats just's ability to catch typos, for example if you type $argument, but works for all possible values of argument, including those with double quotes.

If positional-arguments is true, recipe arguments will be passed as positional arguments to commands. For shell recipes, argument $0 will be the name of the recipe.

For example, running this recipe:

set positional-arguments

@foo bar:
  echo $0
  echo $1

Will produce the following output:

$ just foo hello
foo
hello

When using an sh-compatible shell, such as bash or zsh, $@ expands to the positional arguments given to the recipe, starting from one. When used within double quotes as "$@", arguments including whitespace will be passed on as if they were double-quoted. That is, "$@" is equivalent to "$1" "$2"… When there are no positional parameters, "$@" and $@ expand to nothing (i.e., they are removed).

This example recipe will print arguments one by one on separate lines:

set positional-arguments

@test *args='':
  bash -c 'while (( "$#" )); do echo - $1; shift; done' -- "$@"

Running it with two arguments:

$ just test foo "bar baz"
- foo
- bar baz

Positional arguments may also be turned on a per-recipe basis with the [positional-arguments] attribute^1.29.0:

[positional-arguments]
@foo bar:
  echo $0
  echo $1

Note that PowerShell does not handle positional arguments in the same way as other shells, so turning on positional arguments will likely break recipes that use PowerShell.

If using PowerShell 7.4 or better, the -CommandWithArgs flag will make positional arguments work as expected:

set shell := ['pwsh.exe', '-CommandWithArgs']
set positional-arguments

print-args a b c:
  Write-Output @($args[1..($args.Count - 1)])

Dependencies run before recipes that depend on them:

a: b
  @echo A

b:
  @echo B

$ just a
B
A

In a given invocation of just, a recipe with the same arguments will only run once, regardless of how many times it appears in the command-line invocation, or how many times it appears as a dependency:

a:
  @echo A

b: a
  @echo B

c: a
  @echo C

$ just a a a a a
A
$ just b c
A
B
C

Multiple recipes may depend on a recipe that performs some kind of setup, and when those recipes run, that setup will only be performed once:

build:
  cc main.c

test-foo: build
  ./a.out --test foo

test-bar: build
  ./a.out --test bar

$ just test-foo test-bar
cc main.c
./a.out --test foo
./a.out --test bar

Recipes in a given run are only skipped when they receive the same arguments:

build:
  cc main.c

test TEST: build
  ./a.out --test {{TEST}}

$ just test foo test bar
cc main.c
./a.out --test foo
./a.out --test bar

Normal dependencies of a recipe always run before a recipe starts. That is to say, the dependee always runs before the depender. These dependencies are called "prior dependencies".

A recipe can also have subsequent dependencies, which run immediately after the recipe and are introduced with an &&:

a:
  echo 'A!'

b: a && c d
  echo 'B!'

c:
  echo 'C!'

d:
  echo 'D!'

…running b prints:

$ just b
echo 'A!'
A!
echo 'B!'
B!
echo 'C!'
C!
echo 'D!'
D!

just doesn't support running recipes in the middle of another recipe, but you can call just recursively in the middle of a recipe. Given the following justfile:

a:
  echo 'A!'

b: a
  echo 'B start!'
  just c
  echo 'B end!'

c:
  echo 'C!'

…running b prints:

$ just b
echo 'A!'
A!
echo 'B start!'
B start!
echo 'C!'
C!
echo 'B end!'
B end!

This has limitations, since recipe c is run with an entirely new invocation of just: Assignments will be recalculated, dependencies might run twice, and command line arguments will not be propagated to the child just process.

Dependencies may be run in parallel with the [parallel] attribute.

In this justfile, foo, bar, and baz will execute in parallel when main is run:

[parallel]
main: foo bar baz

foo:
  sleep 1

bar:
  sleep 1

baz:
  sleep 1

The number of simultaneously running recipes may be limited with the --jobs option^1.56.0. The num_jobs() function returns the number of jobs, falling back to the empty list if --jobs was not passed.

GNU parallel may be used to run recipe lines concurrently:

parallel:
  #!/usr/bin/env -S parallel --shebang --ungroup --jobs {{ num_cpus() }}
  echo task 1 start; sleep 3; echo task 1 done
  echo task 2 start; sleep 3; echo task 2 done
  echo task 3 start; sleep 3; echo task 3 done
  echo task 4 start; sleep 3; echo task 4 done

Comments immediately preceding a recipe will appear in just --list:

# build stuff
build:
  ./bin/build

# test stuff
test:
  ./bin/test

$ just --list
Available recipes:
    build # build stuff
    test # test stuff

The [doc] attribute can be used to set or suppress a recipe's doc comment:

# This comment won't appear
[doc('Build stuff')]
build:
  ./bin/build

# This one won't either
[doc]
test:
  ./bin/test

$ just --list
Available recipes:
    build # Build stuff
    test

The value of [doc] may be a const expression^1.56.0.

Recipes and modules may be annotated with one or more group names:

[group('lint')]
js-lint:
    echo 'Running JS linter…'

[group('rust recipes')]
[group('lint')]
rust-lint:
    echo 'Running Rust linter…'

[group('lint')]
cpp-lint:
  echo 'Running C++ linter…'

# not in any group
email-everyone:
    echo 'Sending mass email…'

Recipes are listed by group:

$ just --list
Available recipes:
    email-everyone # not in any group

    [lint]
    cpp-lint
    js-lint
    rust-lint

    [rust recipes]
    rust-lint

just --list --unsorted prints recipes in their justfile order within each group:

$ just --list --unsorted
Available recipes:
    (no group)
    email-everyone # not in any group

    [lint]
    js-lint
    rust-lint
    cpp-lint

    [rust recipes]
    rust-lint

Groups can be listed with --groups:

$ just --groups
Recipe groups:
  lint
  rust recipes

Use just --groups --unsorted to print groups in their justfile order.

Aliases allow recipes to be invoked on the command line with alternative names:

alias b := build

build:
  echo 'Building!'

$ just b
echo 'Building!'
Building!

The target of an alias may be a recipe in a submodule:

mod foo

alias baz := foo::bar

Or a module^1.55.0:

mod frontend

alias f := frontend

$ just f build

Recipes and aliases whose name starts with a _ are omitted from just --list:

test: _test-helper
  ./bin/test

_test-helper:
  ./bin/super-secret-test-helper-stuff

$ just --list
Available recipes:
    test

And from just --summary:

$ just --summary
test

The [private] attribute^1.10.0 may also be used to hide recipes or aliases without needing to change the name:

[private]
foo:

[private]
alias b := bar

bar:

$ just --list
Available recipes:
    bar

This is useful for helper recipes which are only meant to be used as dependencies of other recipes.

The [android], [dragonfly], [freebsd], [linux], [macos], [netbsd], [openbsd], [unix], and [windows] attributes are conditional attributes. By default, items are always enabled. An item with one or more conditional attributes will only be enabled when one or more of those conditional attributes is active.

The conditional attributes originally applied only to recipes, but may now be applied to all top-level items^1.56.0.

This can be used to write justfiles that behave differently depending on which operating system they run on. The run recipe in this justfile will compile and run main.c, using a different C compiler and using the correct output binary name for that compiler depending on the operating system:

[unix]
run:
  cc main.c
  ./a.out

[windows]
run:
  cl main.c
  main.exe

Similarly, a setting can be made conditional on the current operating system:

[unix]
set shell := ['sh', '-cu']

[windows]
set shell := ['cmd', '/c']

If allow-duplicate-recipes is set to true, defining multiple recipes with the same name is not an error and the last definition is used. Defaults to false.

set allow-duplicate-recipes

@foo:
  echo foo

@foo:
  echo bar

$ just foo
bar

Module-level variables may be created by assigning them a value with :=:

foo := "hello"
bar := "world"

baz:
  echo {{ foo + " " + bar }}

All variables in a module may be printed:

$ just --evaluate
bar := "world"
foo := "hello"

Or the value of a single variable:

$ just --evaluate foo
hello

All variables in a submodule or a single variable in a submodule may be printed with a path to the submodule or variable^1.49.0:

$ just --evaluate bob::bar
x := "world"
y := "hello"
$ just --evaluate bob::bar::y
hello

The format of exported variables may be controlled with --evaluate-format^1.49.0:

$ just --evaluate --evaluate-format shell
bar="world"
foo="hello"

The default format is --evaluate-format just:

$ just --evaluate --evaluate-format just
bar := "world"
foo := "hello"

If allow-duplicate-variables is set to true, defining multiple variables with the same name is not an error and the last definition is used. Defaults to false.

set allow-duplicate-variables

a := "foo"
a := "bar"

@foo:
  echo {{a}}

$ just foo
bar

The lazy setting^1.47.0 causes the evaluator to skip evaluating unused variables. This can be beneficial when a justfile contains variables that are expensive to evaluate but only sometimes used.

In the following justfile, token will be skipped when only invoking bar:

set lazy

token := `expensive-script-to-get-credentials`

foo:
  curl -H "Authorization: Bearer {{ token }}" https://example.com/foo

bar:
  cargo test

Because just cannot determine when exported variables are used, assignments with export and assignments in a module with set export will always be evaluated.

Various operators and function calls are supported in expressions, which may be used in assignments, default recipe arguments, and inside recipe body {{…}} substitutions.

tmpdir  := `mktemp -d`
version := "0.2.7"
tardir  := tmpdir / "awesomesauce-" + version
tarball := tardir + ".tar.gz"
config  := quote(config_dir() / ".project-config")

publish:
  rm -f {{tarball}}
  mkdir {{tardir}}
  cp README.md *.c {{ config }} {{tardir}}
  tar zcvf {{tarball}} {{tardir}}
  scp {{tarball}} me@server.com:release/
  rm -rf {{tarball}} {{tardir}}

The + operator returns the left-hand argument concatenated with the right-hand argument:

foobar := 'foo' + 'bar'

The logical operators && and || can be used to coalesce values^1.37.0, similar to Python's and and or. The only false value is the empty list []; every other value, including the empty string '', is true.

These operators require set lists^1.53.0, which is currently unstable.

The && operator returns the empty list if the left-hand argument is false, otherwise it returns the right-hand argument:

foo := [] && 'goodbye'      # []
bar := 'hello' && 'goodbye' # 'goodbye'

The || operator returns the left-hand argument if it is true, otherwise it returns the right-hand argument:

foo := [] || 'goodbye'      # 'goodbye'
bar := 'hello' || 'goodbye' # 'hello'

The / operator can be used to join two strings with a slash:

foo := "a" / "b"

$ just --evaluate foo
a/b

Note that a / is added even if one is already present:

foo := "a/"
bar := foo / "b"

$ just --evaluate bar
a//b

Absolute paths can also be constructed^1.5.0:

foo := / "b"

$ just --evaluate foo
/b

The / operator uses the / character, even on Windows. Thus, using the / operator should be avoided with paths that use universal naming convention (UNC), i.e., those that start with \\, since forward slashes are not supported with UNC paths.

To write a recipe containing {{, use {{{{:

braces:
  echo 'I {{{{LOVE}} curly braces!'

(An unmatched }} is ignored, so it doesn't need to be escaped.)

Another option is to put all the text you'd like to escape inside of an interpolation:

braces:
  echo '{{'I {{LOVE}} curly braces!'}}'

Yet another option is to use {{ "{{" }}:

braces:
  echo 'I {{ "{{" }}LOVE}} curly braces!'

'single', "double", and '''triple''' quoted string literals are supported. Unlike in recipe bodies, {{…}} interpolations are not supported inside strings.

Double-quoted strings support escape sequences:

carriage-return   := "\r"
double-quote      := "\""
newline           := "\n"
no-newline        := "\
"
slash             := "\\"
tab               := "\t"
unicode-codepoint := "\u{1F916}"

$ just --evaluate
"arriage-return   := "
double-quote      := """
newline           := "
"
no-newline        := ""
slash             := "\"
tab               := "     "
unicode-codepoint := "🤖"

The unicode character escape sequence \u{…}^1.36.0 accepts up to six hex digits.

Strings may contain line breaks:

single := '
hello
'

double := "
goodbye
"

Single-quoted strings do not recognize escape sequences:

escapes := '\t\n\r\"\\'

$ just --evaluate
escapes := "\t\n\r\"\\"

Indented versions of both single- and double-quoted strings, delimited by triple single- or double-quotes, are supported. Indented string lines are stripped of a leading line break, and leading whitespace common to all non-blank lines:

# this string will evaluate to `foo\nbar\n`
x := '''
  foo
  bar
'''

# this string will evaluate to `abc\n  wuv\nxyz\n`
y := """
  abc
    wuv
  xyz
"""

Similar to unindented strings, indented double-quoted strings process escape sequences, and indented single-quoted strings ignore escape sequences. Escape sequence processing takes place after unindentation. The unindentation algorithm does not take escape-sequence produced whitespace or newlines into account.

Strings prefixed with x are shell expanded^1.27.0:

foobar := x'~/$FOO/${BAR}'

This expansion is performed at compile time, so variables from .env files and exported just variables cannot be used. However, this allows shell expanded strings to be used in places like settings and import paths, which cannot depend on just variables and .env files.

Strings prefixed with f are format strings^1.44.0:

name := "world"
message := f'Hello, {{name}}!'

Format strings may contain interpolations delimited with {{…}} that contain expressions. Format strings evaluate to the concatenated string fragments and evaluated expressions.

Use {{{{ to include a literal {{ in a format string:

foo := f'I {{{{LOVE} curly braces!'

The lists setting^1.53.0 allows values that are lists of strings. It is currently unstable and will change in backwards incompatible ways. This section documents changes in behavior when set lists is enabled.

It has not yet been decided how lists should behave with many of the built-in functions. Functions that have been updated to accept lists are mentioned in this section. Using lists with any other function is an error. The join_list() function can be used to convert lists into space-separated strings for use with un-upgraded functions. Feedback on how built-in functions should behave with lists, and on lists in general, is most welcome! Feel free to open an issue or leave a comment in the set lists tracking issue.

Variadic recipe parameters are lists of strings instead of single space-separated strings.

List literals are written [a, b, c]. List literals flatten their arguments, since lists may only contain strings and not other lists. For example, [["a", "b"], [], "c"] evaluates to ["a", "b", "c"].

Lists in recipe and f-string interpolations are joined with spaces into a single string.

Each argument to a dependency binds to exactly one parameter, and supplying extra arguments to a variadic dependency is an error.

Dependencies may be invoked once per element of a list with *(recipe *argument).

A parameter evaluates to the default when the argument is the empty list.

Passing an empty list to a non-* parameter without a default is an error.

The else of an if may be omitted, in which case the if evaluates to [] when its condition is false.

Message values in assert(condition, message) and [confirm(message)] are space-joined for display.

The + and / operators combine strings and lists. A string and a non-empty list are combined by concatenating the string with each element of the list. Two lists of the same length are combined into a list containing the pairwise concatenated elements of both operands. Combining two lists of different lengths is an error.

The ++ operator performs list concatenation.

The canonical boolean true value is the string "true", and the canonical boolean false value is the empty list []. All values other than the empty list are truthy, including ''.

The condition of an if or assert() may be any expression, which is evaluated for truthiness.

The comparison operators ==, !=, =~, and !~ may be used anywhere, not just in if and assert(), and evaluate to "true" or [].

value =~ regexes is true if any element in value matches any regex in regexes. It is false if either value or regexes is empty.

value !~ regexes is true if no element in value matches any regex in regexes. It is true if either value or regexes is empty.

Values may be negated with !. !expression evaluates to "true" if expression is [], otherwise it evaluates to [].

The script-interpreter, shell, and windows-shell settings flatten their elements like list literals.

When positional-arguments is set, list arguments are space-joined unless they are variadic, in which case they are passed as one positional argument per element.

The --dotenv-filename and --dotenv-path options may be passed multiple times, and the dotenv-filename and dotenv-path settings accept lists, in which case multiple environment files may be loaded. The values of dotenv-path are tried first. If none are found the current directory is searched for the names in dotenv-filename, followed by its ancestors, stopping in the first directory that contains any of them and loading all matching files in that directory. If multiple environment files are loaded, variables in files later in list take precedence over earlier ones.

Each element of the value of set dotenv-command is run as a command, with variables from commands later in the list taking precedence over variables from commands earlier in the list.

The [arg(flag)] attribute makes the parameter a flag which does not take a value on the command line. For example, with [arg('foo', long, flag)], foo will be "true" when --foo is passed, and [] otherwise. Flag parameters may not have a default.

The [arg(multiple)] attribute allows an option or flag to be passed more than once, assigning the list of passed values to the parameter. When combined with flag or value=VALUE, "true" or VALUE, respectively, are repeated for each occurance of the flag.

The [arg(min=MIN)] and [arg(max=MAX)] attributes^1.56.0 can be used to limit the number of values an option or flag may receive.

The value of [arg(help)] may be a list, in which case the help string is the elements of the list joined with spaces. If the list is empty, the argument has no help string.

The value of [arg(pattern)] may be a list, in which case the argument is accepted if it matches any pattern in the list. If the value is the empty list, any argument is accepted. For example, with [arg('foo', pattern=['--help', '--version'])], foo may be --help or --version.

In [env(variable, value)] if value is [], variable is not set. Otherwise it is set to value joined with spaces.

• absolute_path() - Applies to each list element individually.
• append() - Applies to each list element individually and does not split elements on whitespace.
• assert(condition, message) - Evaluates to condition.
• bool(value) Converts value to the canonical boolean values. Returns [] when value is "" "0" "false", or [], and "true" when value is "1" or "true". All other values are an error. Can be used to parse booleans passed as arguments or environment variables.
• env(keys, default) Checks for the environment variables named in keys in order and returns the value of the first that is set. Returns default if none are set or an error if default is omitted.
• is_dependency() - Returns the canonical booleans.
• join_list(value, separator) - Joins value into a single string. Elements are joined with separator, or with a single space if separator is omitted.
• len(value) - Returns the number of elements in value.
• path_exists() - Returns the canonical booleans.
• prepend() - Applies to each list element individually and does not split elements on whitespace.
• quote() - Applies to each list element individually.
• semver_matches() - Returns the canonical booleans.
• show(value) - Converts value into a string containing its literal representation. Brackets are used for empty and multi-element lists, e.g., "[]" and "["foo", "bar"]", but not single-element lists, e.g., "foo".
• split(string, separator) - Splits string into a list on each occurrence of separator. If separator is omitted, string is split on whitespace, with leading and trailing whitespace trimmed.
• which() - Returns the empty list when no executable is found.

Each list element is quote()'ed separately:

set unstable
set lists

@foo *args:
  printf '%s\n' {{ quote(args) }}

$ just foo bar 'baz bob'
bar
baz bob

The return value of quote(args) is 'bar' 'baz bob', instead of 'bar baz bob', as would be the case without set lists.

Variadic positional arguments:

set unstable
set lists
set positional-arguments

foo *args: (bar args 'bob') (baz args)

@bar first second:
  echo first=$1
  echo second=$2

@baz *args:
  echo '$1='$1
  echo '$2='$2

$ just foo one two
first=one two
second=bob
$1=one
$2=two

A mapped dependency is invoked once per element of its starred argument, with [parallel] to run them in parallel:

set unstable
set lists

[parallel]
build target *platform: *(compile target *platform)

@compile target platform:
  echo compiling {{ target }} for {{ platform }}…

$ just build x86 foo bar
compiling foo for x86…
compiling bar for x86…

The canonical false value [] is recommended as a default for options:

set unstable
set lists

[arg('bar', long)]
foo bar=[]:

if/else expressions evaluate different branches depending on if two expressions evaluate to the same value:

foo := if "2" == "2" { "Good!" } else { "1984" }

bar:
  @echo "{{foo}}"

$ just bar
Good!

It is also possible to test for inequality:

foo := if "hello" != "goodbye" { "xyz" } else { "abc" }

bar:
  @echo {{foo}}

$ just bar
xyz

And match against regular expressions:

foo := if "hello" =~ 'hel+o' { "match" } else { "mismatch" }

bar:
  @echo {{foo}}

$ just bar
match

Regular expressions are provided by the regex crate, whose syntax is documented on docs.rs. Since regular expressions commonly use backslash escape sequences, consider using single-quoted string literals, which will pass slashes to the regex parser unmolested.

Conditional expressions short-circuit, which means they only evaluate one of their branches. This can be used to make sure that backtick expressions don't run when they shouldn't.

foo := if env_var("RELEASE") == "true" { `get-something-from-release-database` } else { "dummy-value" }

Conditionals can be used inside of recipes:

bar foo:
  echo {{ if foo == "bar" { "hello" } else { "goodbye" } }}

Multiple conditionals can be chained:

foo := if "hello" == "goodbye" {
  "xyz"
} else if "a" == "a" {
  "abc"
} else {
  "123"
}

bar:
  @echo {{foo}}

$ just bar
abc

Backticks can be used to store the result of commands:

localhost := `dumpinterfaces | cut -d: -f2 | sed 's/\/.*//' | sed 's/ //g'`

serve:
  ./serve {{localhost}} 8080

Indented backticks, delimited by three backticks, are de-indented in the same manner as indented strings:

# This backtick evaluates the command `echo foo\necho bar\n`, which produces the value `foo\nbar\n`.
stuff := ```
    echo foo
    echo bar
  ```

See the Strings section for details on unindenting.

Backticks may not start with #!. This syntax is reserved for a future upgrade.

The shell(…) function provides a more general mechanism to invoke external commands, including the ability to execute the contents of a variable as a command, and to pass arguments to a command.

Execution can be halted with the error function. For example:

foo := if "hello" == "goodbye" {
  "xyz"
} else if "a" == "b" {
  "abc"
} else {
  error("123")
}

Which produces the following error when run:

error: Call to function `error` failed: 123
   |
16 |   error("123")

just provides many built-in functions for use in expressions, including recipe body {{…}} substitutions, assignments, and default parameter values.

All functions ending in _directory can be abbreviated to _dir. So home_directory() can also be written as home_dir(). In addition, invocation_directory_native() can be abbreviated to invocation_dir_native().

New functions may be defined^1.49.0:

set unstable

hello(name) := f"Hello, {{ name }}!"

foo:
  echo '{{ hello("World") }}'

User-defined functions are currently unstable.

Functions may reference assignments in the same module:

set unstable

base := "foo"

join(extension) := base + "." + extension

create:
  touch {{ join("c") }}
  touch {{ join("html") }}
  touch {{ join("txt") }}

Commands in shell recipes may be prefixed with any combination of the sigils -, @, and ?.

The @ sigil toggles command echoing:

foo:
  @echo "This line won't be echoed!"
  echo "This line will be echoed!"

@bar:
  @echo "This line will be echoed!"
  echo "This line won't be echoed!"

The - sigil causes recipe execution to continue even if the command returns a nonzero exit status:

# execution will continue, even if bar doesn't exist
foo:
  -rmdir bar
  mkdir bar
  echo 'so much good stuff' > bar/stuff.txt

The ? sigil^1.47.0 causes the current recipe to stop executing if the command exits with status code 1, however execution of other recipes will continue. Exit status 0 causes the current recipe to continue execution as normal. All other exit codes are reserved and should not be used, as they may be given meaning in a future version of just.

If the guards setting is unset or false, ? sigils are ignored and instead treated as part of the command.

set guards

@foo: bar
  echo FOO

@bar:
  ?[[ -f baz ]]
  echo BAR

$ just foo
FOO
$ touch baz
$ just foo
BAR
FOO

A recipe name may be prefixed with @ to invert the meaning of @ before each line:

@quiet:
  echo hello
  echo goodbye
  @# all done!

Now only the lines starting with @ will be echoed:

$ just quiet
hello
goodbye
# all done!

All recipes in a justfile can be made quiet with set quiet:

set quiet

foo:
  echo "This is quiet"

@foo2:
  echo "This is also quiet"

The [no-quiet] attribute overrides this setting:

set quiet

foo:
  echo "This is quiet"

[no-quiet]
foo2:
  echo "This is not quiet"

Shebang recipes are quiet by default:

foo:
  #!/usr/bin/env bash
  echo 'Foo!'

$ just foo
Foo!

Adding @ to a shebang recipe name makes just print the recipe before executing it:

@bar:
  #!/usr/bin/env bash
  echo 'Bar!'

$ just bar
#!/usr/bin/env bash
echo 'Bar!'
Bar!

just normally prints error messages when a recipe line fails. These error messages can be suppressed using the [no-exit-message]^1.7.0 attribute on individual recipes, or module-wide with set no-exit-message^1.39.0. You may find this especially useful with a recipe that wraps a tool:

git *args:
    @git {{args}}

$ just git status
fatal: not a git repository (or any of the parent directories): .git
error: Recipe `git` failed on line 2 with exit code 128

Add the attribute to suppress the exit error message when the tool exits with a non-zero code:

[no-exit-message]
git *args:
    @git {{args}}

$ just git status
fatal: not a git repository (or any of the parent directories): .git

Recipes that start with #! are called shebang recipes, and are executed by saving the recipe body to a file and running it. This lets you write recipes in different languages:

polyglot: python js perl sh ruby nu

python:
  #!/usr/bin/env python3
  print('Hello from python!')

js:
  #!/usr/bin/env node
  console.log('Greetings from JavaScript!')

perl:
  #!/usr/bin/env perl
  print "Larry Wall says Hi!\n";

sh:
  #!/usr/bin/env sh
  hello='Yo'
  echo "$hello from a shell script!"

nu:
  #!/usr/bin/env nu
  let hello = 'Hola'
  echo $"($hello) from a nushell script!"

ruby:
  #!/usr/bin/env ruby
  puts "Hello from ruby!"

$ just polyglot
Hello from python!
Greetings from JavaScript!
Larry Wall says Hi!
Yo from a shell script!
Hola from a nushell script!
Hello from ruby!

On Unix-like operating systems, including Linux and macOS, shebang recipes are executed by saving the recipe body to a file in a temporary directory, marking the file as executable, and executing it. The OS then parses the shebang line into a command line and invokes it, including the path to the file. For example, if a recipe starts with #!/usr/bin/env bash, the final command that the OS runs will be something like /usr/bin/env bash /tmp/PATH_TO_SAVED_RECIPE_BODY.

Shebang line splitting is operating system dependent. When passing a command with arguments, you may need to tell env to split them explicitly by using the -S flag:

run:
  #!/usr/bin/env -S bash -x
  ls

Windows does not support shebang lines. On Windows, just splits the shebang line into a command and arguments, saves the recipe body to a file, and invokes the split command and arguments, adding the path to the saved recipe body as the final argument. For example, on Windows, if a recipe starts with #! py, the final command the OS runs will be something like py C:\Temp\PATH_TO_SAVED_RECIPE_BODY.

Recipes with a [script(COMMAND)]^1.32.0 attribute are run as scripts interpreted by COMMAND. This avoids some of the issues with shebang recipes, such as the use of cygpath on Windows, the need to use /usr/bin/env, inconsistencies in shebang line splitting across Unix OSs, and requiring a temporary directory from which files can be executed.

Recipes with an empty [script] attribute are executed with the value of set script-interpreter := […]^1.33.0, defaulting to sh -eu, and not the value of set shell.

The body of the recipe is evaluated, written to disk in the temporary directory, and run by passing its path as an argument to COMMAND.

With set default-script := true^1.52.0, recipes default to script recipes instead of shell recipes, unless overridden with the [shell] attribute^1.52.0.

Both script and shebang recipes write the recipe body to a temporary file for execution. Script recipes execute that file by passing it to a command, while shebang recipes execute the file directly. Shebang recipe execution will fail if the filesystem containing the temporary file is mounted with noexec or is otherwise non-executable.

The directory that just writes temporary files to may be configured in a number of ways, from highest to lowest precedence:

• Globally with the --tempdir command-line option or the JUST_TEMPDIR environment variable^1.41.0.

• On a per-module basis with the tempdir setting.

• Globally on Linux with the XDG_RUNTIME_DIR environment variable.

• Falling back to the directory returned by std::env::temp_dir.

If you're writing a bash shebang recipe, consider adding set -euxo pipefail:

foo:
  #!/usr/bin/env bash
  set -euxo pipefail
  hello='Yo'
  echo "$hello from Bash!"

It isn't strictly necessary, but set -euxo pipefail turns on a few useful features that make bash shebang recipes behave more like normal, shell just recipes:

• set -e makes bash exit if a command fails.

• set -u makes bash exit if a variable is undefined.

• set -x makes bash print each script line before it's run.

• set -o pipefail makes bash exit if a command in a pipeline fails. This is bash-specific, so isn't turned on in normal shell just recipes.

Together, these avoid a lot of shell scripting gotchas.

On Windows, shebang interpreter paths containing a / are translated from Unix-style paths to Windows-style paths using cygpath, a utility that ships with Cygwin.

For example, to execute this recipe on Windows:

echo:
  #!/bin/sh
  echo "Hello!"

The interpreter path /bin/sh will be translated to a Windows-style path using cygpath before being executed.

If the interpreter path does not contain a / it will be executed without being translated. This is useful if cygpath is not available, or you wish to pass a Windows-style path to the interpreter.

Recipe lines can be indented with spaces or tabs, but not a mix of both. All of a recipe's lines must have the same type of indentation, but different recipes in the same justfile may use different indentation.

Each recipe must be indented at least one level from the recipe-name but after that may be further indented.

Here's a justfile with a recipe indented with spaces, represented as ·, and tabs, represented as →.

[windows]
set shell := ["pwsh", "-NoLogo", "-NoProfileLoadTime", "-Command"]

set ignore-comments

list-space directory:
··#!pwsh
··foreach ($item in $(Get-ChildItem {{directory}} )) {
····echo $item.Name
··}
··echo ""

# indentation nesting works even when newlines are escaped
list-tab directory:
→ @foreach ($item in $(Get-ChildItem {{directory}} )) { \
→ → echo $item.Name \
→ }
→ @echo ""

PS > just list-space ~
Desktop
Documents
Downloads

PS > just list-tab ~
Desktop
Documents
Downloads

Recipes without an initial shebang are evaluated and run line-by-line, which means that multi-line constructs probably won't do what you want.

For example, with the following justfile:

conditional:
  if true; then
    echo 'True!'
  fi

The extra leading whitespace before the second line of the conditional recipe will produce a parse error:

$ just conditional
error: Recipe line has extra leading whitespace
  |
3 |         echo 'True!'
  |     ^^^^^^^^^^^^^^^^

To work around this, you can write conditionals on one line, escape newlines with slashes, or add a shebang to your recipe. Some examples of multi-line constructs are provided for reference.

conditional:
  if true; then echo 'True!'; fi

conditional:
  if true; then \
    echo 'True!'; \
  fi

conditional:
  #!/usr/bin/env sh
  if true; then
    echo 'True!'
  fi

for:
  for file in `ls .`; do echo $file; done

for:
  for file in `ls .`; do \
    echo $file; \
  done

for:
  #!/usr/bin/env sh
  for file in `ls .`; do
    echo $file
  done

while:
  while `server-is-dead`; do ping -c 1 server; done

while:
  while `server-is-dead`; do \
    ping -c 1 server; \
  done

while:
  #!/usr/bin/env sh
  while `server-is-dead`; do
    ping -c 1 server
  done

Parenthesized expressions can span multiple lines:

abc := ('a' +
        'b'
         + 'c')

abc2 := (
  'a' +
  'b' +
  'c'
)

foo param=('foo'
      + 'bar'
    ):
  echo {{param}}

bar: (foo
        'Foo'
     )
  echo 'Bar!'

Lines ending with a backslash continue on to the next line as if the lines were joined by whitespace^1.15.0:

a := 'foo' + \
     'bar'

foo param1 \
  param2='foo' \
  *varparam='': dep1 \
                (dep2 'foo')
  echo {{param1}} {{param2}} {{varparam}}

dep1: \
    # this comment is not part of the recipe body
  echo 'dep1'

dep2 \
  param:
    echo 'Dependency with parameter {{param}}'

Backslash line continuations can also be used in interpolations. The line following the backslash must be indented.

recipe:
  echo '{{ \
  "This interpolation " + \
    "has a lot of text." \
  }}'
  echo 'back to recipe body'

Recipe lines are interpreted by the shell, not just, so it's not possible to set just variables in the middle of a recipe:

foo:
  x := "hello" # This doesn't work!
  echo {{x}}

It is possible to use shell variables, but there's another problem. Every recipe line is run by a new shell instance, so variables set in one line won't be set in the next:

foo:
  x=hello && echo $x # This works!
  y=bye
  echo $y            # This doesn't, `y` is undefined here!

The best way to work around this is to use a shebang recipe. Shebang recipe bodies are extracted and run as scripts, so a single shell instance will run the whole thing:

foo:
  #!/usr/bin/env bash
  set -euxo pipefail
  x=hello
  echo $x

There are a number of ways to configure the shell for shell recipes, which are the default when a recipe does not start with a #! shebang. Their precedence, from highest to lowest, is:

1. The --shell and --shell-arg command line options. Passing either of these will cause just to ignore any settings in the current justfile.
2. set windows-shell := [...] (deprecated)
3. set windows-powershell (deprecated)
4. set shell := [...]

Use the [windows] and [unix] attributes with set shell to use different a shells on Windows.

The shell setting controls the command used to invoke recipe lines and backticks. Shebang recipes are unaffected. The default shell is sh -cu.

# use python3 to execute recipe lines and backticks
set shell := ["python3", "-c"]

# use print to capture result of evaluation
foos := `print("foo" * 4)`

foo:
  print("Snake snake snake snake.")
  print("{{foos}}")

just passes the command to be executed as an argument. Many shells will need an additional flag, often -c, to make them evaluate the first argument.

just uses sh on Windows by default. To use a different shell on Windows, use the [windows] attribute on the shell setting:

[windows]
set shell := ["powershell.exe", "-NoLogo", "-Command"]

hello:
  Write-Host "Hello, world!"

See powershell.just for a justfile that uses PowerShell on all platforms.

set shell := ["python3", "-c"]

set shell := ["bash", "-uc"]

set shell := ["zsh", "-uc"]

set shell := ["fish", "-c"]

set shell := ["nu", "-c"]

If you want to change the default table mode to light:

set shell := ['nu', '-m', 'light', '-c']

Nushell was written in Rust, and has cross-platform support for Windows / macOS and Linux.

Assignments prefixed with the export keyword will be exported to recipes as environment variables:

export RUST_BACKTRACE := "1"

test:
  # will print a stack trace if it crashes
  cargo test

Parameters prefixed with a $ will be exported as environment variables:

test $RUST_BACKTRACE="1":
  # will print a stack trace if it crashes
  cargo test

You can also use the [env(NAME, VALUE)] attribute to export environment variables to a specific recipe:

[env("RUST_BACKTRACE", "1")]
test:
  # will print a stack trace if it crashes
  cargo test

Exported variables and parameters are not exported to backticks in the same scope.

export WORLD := "world"
# This backtick will fail with "WORLD: unbound variable"
BAR := `echo hello $WORLD`

# Running `just a foo` will fail with "A: unbound variable"
a $A $B=`echo $A`:
  echo $A $B

When export is set, all just variables are exported as environment variables.

Environment variables can be unexported with the unexport keyword^1.29.0:

unexport FOO

@foo:
  echo $FOO

$ export FOO=bar
$ just foo
sh: FOO: unbound variable

Environment variables from the environment are passed automatically to the recipes.

print_home_folder:
  @echo "HOME is: '${HOME}'"

$ just
HOME is: '/home/myuser'

Environment variables can be propagated to just variables using the env() function. See environment-variables.

If any of dotenv-load, dotenv-filename, dotenv-override, dotenv-path, or dotenv-required are set, just will try to load environment variables from a file.

If dotenv-path is set, just will look for a file at the given path, which may be absolute, or relative to the working directory.

The command-line option --dotenv-path, short form -E, can be used to set or override dotenv-path at runtime.

If dotenv-filename is set, just will look for a file at the given path, relative to the working directory and each of its ancestors.

The command-line option --dotenv-filename, short form -F, can be used to set or override dotenv-filename at runtime.

If dotenv-filename is not set, but dotenv-load or dotenv-required are set, just will look for a file named .env, relative to the working directory and each of its ancestors.

dotenv-filename and dotenv-path are similar, but dotenv-path is only checked relative to the working directory, whereas dotenv-filename is checked relative to the working directory and each of its ancestors.

It is not an error if an environment file is not found, unless dotenv-required is set.

The loaded variables are environment variables, not just variables, and so must be accessed using $VARIABLE_NAME in recipes and backticks.

If dotenv-override is set, variables from the environment file will override existing environment variables.

If dotenv-command is set, just runs it with the configured shell and loads its standard output as an environment file.

This is useful for sourcing secrets from a secret manager or vault:

set dotenv-command := 'sops -d .enc.env'

The command-line option --dotenv-command can be used to set or override dotenv-command at runtime, and may be passed multiple times. Each value is run as a command, with variables from later commands taking precedence over variables from earlier commands.

For example, if your .env file contains:

# a comment, will be ignored
DATABASE_ADDRESS=localhost:6379
SERVER_PORT=1337

And your justfile contains:

set dotenv-load

serve:
  @echo "Starting server with database $DATABASE_ADDRESS on port $SERVER_PORT…"
  ./server --database $DATABASE_ADDRESS --port $SERVER_PORT

just serve will output:

$ just serve
Starting server with database localhost:6379 on port 1337…
./server --database $DATABASE_ADDRESS --port $SERVER_PORT

Variables in environment files loaded in parent modules are inherited by submodules.

Environment files are loaded in submodules^1.49.0 and may override variables defined in parent module environment files.

The export setting causes all just variables to be exported as environment variables. Defaults to false.

set export

a := "hello"

@foo b:
  echo $a
  echo $b

$ just foo goodbye
hello
goodbye

Each line of each recipe is executed by a fresh shell, so it is not possible to share environment variables between recipes.

Some tools, like Python's venv, require loading environment variables in order to work, making them challenging to use with just. As a workaround, you can execute the virtual environment binaries directly:

venv:
  [ -d foo ] || python3 -m venv foo

run: venv
  ./foo/bin/python3 main.py

By default, recipes run with the working directory set to the directory that contains the justfile.

The [no-cd] attribute can be used to make recipes run with the working directory set to the directory in which just was invoked.

@foo:
  pwd

[no-cd]
@bar:
  pwd

$ cd subdir
$ just foo
/
$ just bar
/subdir

Note that this does not affect dotenv file search, or the working directory of backtick and shell() invocations.

Use set no-cd^1.51.0 to make all recipes in the current module default to the same behavior.

set no-cd and set working-directory can be overridden on a per-recipe basis with the [no-cd] and [working-directory] attributes.

You can override the working directory for all recipes with set working-directory := '…':

set working-directory := 'bar'

@foo:
  pwd

$ pwd
/home/bob
$ just foo
/home/bob/bar

You can override the working directory for a specific recipe with the working-directory attribute^1.38.0:

[working-directory: 'bar']
@foo:
  pwd

$ pwd
/home/bob
$ just foo
/home/bob/bar

The argument to the working-directory setting or working-directory attribute may be an expression^1.51.0 whose value is absolute or relative. If it is relative it is interpreted relative to the default working directory.

Each recipe line is executed by a new shell, so if you change the working directory on one line, it won't have an effect on later lines:

foo:
  pwd    # This `pwd` will print the same directory…
  cd bar
  pwd    # …as this `pwd`!

There are a couple ways around this. One is to call cd on the same line as the command you want to run:

foo:
  cd bar && pwd

The other is to use a shebang recipe. Shebang recipe bodies are extracted and run as scripts, so a single shell instance will run the whole thing, and thus a cd on one line will affect later lines, just like a shell script:

foo:
  #!/usr/bin/env bash
  set -euxo pipefail
  cd bar
  pwd

just normally executes recipes with the current directory set to the directory that contains the justfile. This can be disabled using the [no-cd] attribute^1.9.0. This can be used to create recipes which use paths relative to the invocation directory, or which operate on the current directory.

For example, this commit recipe:

[no-cd]
commit file:
  git add {{file}}
  git commit

Can be used with paths that are relative to the current directory, because [no-cd] prevents just from changing the current directory when executing commit.

Note that this does not affect dotenv file search, or the working directory of backtick and shell() invocations.

One justfile can include the contents of another using import statements.

If you have the following justfile:

import 'foo/bar.just'

a: b
  @echo A

And the following text in foo/bar.just:

b:
  @echo B

foo/bar.just will be included in justfile and recipe b will be defined:

$ just b
B
$ just a
B
A

The import path can be absolute or relative to the location of the justfile containing it. A leading ~/ in the import path is replaced with the current user's home directory.

Justfiles are insensitive to order, so included files can reference variables and recipes defined after the import statement.

Imported files can themselves contain imports, which are processed recursively.

allow-duplicate-recipes and allow-duplicate-variables allow duplicate recipes and variables, respectively, to override each other, instead of producing an error.

Within a module, later definitions override earlier definitions:

set allow-duplicate-recipes

foo:

foo:
  echo 'yes'

When imports are involved, things unfortunately get much more complicated and hard to explain.

Shallower definitions always override deeper definitions, so recipes at the top level will override recipes in imports, and recipes in an import will override recipes in an import which itself imports those recipes.

When two duplicate definitions are imported and are at the same depth, the one from the earlier import will override the one from the later import.

This is because just uses a stack when processing imports, pushing imports onto the stack in source-order, and always processing the top of the stack next, so earlier imports are actually handled later by the compiler.

This is definitely a bug, but since just has very strong backwards compatibility guarantees and we take enormous pains not to break anyone's justfile, see issue #2540 for discussion on whether or not we can actually fix it.

Imports may be made optional by putting a ? after the import keyword:

import? 'foo/bar.just'

Importing the same source file multiple times is not an error^1.37.0. This allows importing multiple justfiles, for example foo.just and bar.just, which both import a third justfile containing shared recipes, for example baz.just, without the duplicate import of baz.just being an error:

# justfile
import 'foo.just'
import 'bar.just'

# foo.just
import 'baz.just'
foo: baz

# bar.just
import 'baz.just'
bar: baz

# baz
baz:

A justfile can declare modules using mod statements^1.19.0.

mod statements were stabilized in just^1.31.0.

If you have the following justfile:

mod bar

a:
  @echo A

And the following text in bar.just:

b:
  @echo B

bar.just will be included in justfile as a submodule. Recipes, aliases, and variables defined in one submodule cannot be used in another, and each module uses its own settings.

Recipes in submodules can be invoked as subcommands:

$ just bar b
B

Or with path syntax:

$ just bar::b
B

If a module is named foo, just will search for the module file in foo.just, foo/mod.just, foo/justfile, and foo/.justfile. In the latter two cases, the module file may have any capitalization.

Module statements may be of the form:

mod foo 'PATH'

Which loads the module's source file from PATH, instead of from the usual locations. A leading ~/ in PATH is replaced with the current user's home directory. PATH may point to the module source file itself, or to a directory containing the module source file with the name mod.just, justfile, or .justfile. In the latter two cases, the module file may have any capitalization.

Environment files are loaded for each module, respecting that module's related settings. Environment variables from parent modules are visible in child modules.

Recipes in submodules without the [no-cd] attribute run with the working directory set to the directory containing the submodule source file.

justfile() and justfile_directory() always return the path to the root justfile and the directory that contains it, even when called from submodule recipes.

Modules may be made optional by putting a ? after the mod keyword:

mod? foo

Missing source files for optional modules do not produce an error. If a recipe or alias depends on a missing optional module, directly, or transitively, it will be disabled. Attempting to invoke a disabled recipe or alias is an error, but other non-disabled recipes can still be run.^1.52.0

Optional modules with no source file do not conflict, so you can have multiple mod statements with the same name, but with different source file paths, as long as at most one source file exists:

mod? foo 'bar.just'
mod? foo 'baz.just'

Modules may be given doc comments which appear in --list output^1.30.0:

# foo is a great module!
mod foo

$ just --list
Available recipes:
    foo ... # foo is a great module!

Modules are still missing a lot of features, for example, the ability to refer to variables in other modules. See the module improvement tracking issue for more information.

If the first argument passed to just contains a /, then the following occurs:

1. The argument is split at the last /.

2. The part before the last / is treated as a directory. just will start its search for the justfile there, instead of in the current directory.

3. The part after the last slash is treated as a normal argument, or ignored if it is empty.

This may seem a little strange, but it's useful if you wish to run a command in a justfile that is in a subdirectory.

For example, if you are in a directory which contains a subdirectory named foo, which contains a justfile with the recipe build, which is also the default recipe, the following are all equivalent:

$ (cd foo && just build)
$ just foo/build
$ just foo/

Additional recipes after the first are sought in the same justfile. For example, the following are both equivalent:

$ just foo/a b
$ (cd foo && just a b)

And will both invoke recipes a and b in foo/justfile.

If a recipe is not found in a justfile and the fallback setting is set, just will look for justfiles in the parent directory and up, until it reaches the root directory. just will stop after it reaches a justfile in which the fallback setting is false or unset.

As an example, suppose the current directory contains this justfile:

set fallback
foo:
  echo foo

And the parent directory contains this justfile:

bar:
  echo bar

$ just bar
Trying ../justfile
echo bar
bar

If you wish to include a mod or import source file in many justfiles without needing to duplicate it, you can use an optional mod or import, along with a recipe to fetch the module source:

import? 'foo.just'

fetch:
  curl https://raw.githubusercontent.com/casey/just/master/justfile > foo.just

Given the above justfile, after running just fetch, the recipes in foo.just will be available.

If you want some recipes to be available everywhere, you have a few options.

just --global-justfile, or just -g for short, searches the following paths, in-order, for a justfile:

• $XDG_CONFIG_HOME/just/justfile
• $HOME/.config/just/justfile
• $HOME/justfile
• $HOME/.justfile

You can put recipes that are used across many projects in a global justfile to easily invoke them from any directory.

You can also adopt some of the following workflows. These tips assume you've created a justfile at ~/.user.justfile, but you can put this justfile at any convenient path on your system.

If you want to call the recipes in ~/.user.justfile by name, and don't mind creating an alias for every recipe, add the following to your shell's initialization script:

for recipe in `just --justfile ~/.user.justfile --summary`; do
  alias $recipe="just --justfile ~/.user.justfile --working-directory . $recipe"
done

Now, if you have a recipe called foo in ~/.user.justfile, you can just type foo at the command line to run it.

It took me way too long to realize that you could create recipe aliases like this. Notwithstanding my tardiness, I am very pleased to bring you this major advance in justfile technology.

If you'd rather not create aliases for every recipe, you can create a single alias:

alias .j='just --justfile ~/.user.justfile --working-directory .'

Now, if you have a recipe called foo in ~/.user.justfile, you can just type .j foo at the command line to run it.

I'm pretty sure that nobody actually uses this feature, but it's there.

¯\_(ツ)_/¯

You can customize the above aliases with additional options. For example, if you'd prefer to have the recipes in your justfile run in your home directory, instead of the current directory:

alias .j='just --justfile ~/.user.justfile --working-directory ~'

If the argument to --justfile ends in .md, just extracts the contents of unindented just fenced code blocks and writes them to a justfile in a temporary directory^1.53.0:

# Project

Build the project:

```just
build:
  echo Building…
```

$ just --justfile README.md build
echo Building…
Building…

By adding a shebang line to the top of a justfile and making it executable, just can be used as an interpreter for scripts:

$ cat > script <<EOF
#!/usr/bin/env just --justfile

foo:
  echo foo
EOF
$ chmod +x script
$ ./script foo
echo foo
foo

When a script with a shebang is executed, the system supplies the path to the script as an argument to the command in the shebang. So, with a shebang of #!/usr/bin/env just --justfile, the command will be /usr/bin/env just --justfile PATH_TO_SCRIPT.

With the above shebang, just will change its working directory to the location of the script. If you'd rather leave the working directory unchanged, use #!/usr/bin/env just --working-directory . --justfile.

Note: Shebang line splitting is not consistent across operating systems. The previous examples have only been tested on macOS. On Linux, you may need to pass the -S flag to env:

#!/usr/bin/env -S just --justfile

default:
  echo foo

just looks for justfiles named justfile and .justfile, which can be used to keep a justfile hidden.

Recipes can be listed in alphabetical order with just --list:

$ just --list
Available recipes:
    build
    test
    deploy
    lint

Recipes in submodules can be listed with just --list PATH, where PATH is a space- or ::-separated module path:

$ cat justfile
mod foo
$ cat foo.just
mod bar
$ cat bar.just
baz:
$ just --list foo bar
Available recipes:
    baz
$ just --list foo::bar
Available recipes:
    baz

just --summary is more concise:

$ just --summary
build test deploy lint

Pass --unsorted to print recipes in the order they appear in the justfile:

test:
  echo 'Testing!'

build:
  echo 'Building!'

$ just --list --unsorted
Available recipes:
    test
    build

$ just --summary --unsorted
test build

If you'd like just to default to listing the recipes in the justfile, set default-list^1.52.0:

set default-list := true

The setting is per-module, so invoking a module path with default-list enabled lists that module's recipes.

You can also default to listing recipes this behavior by settting the environment variable JUST_DEFAULT_LIST=true or passing --default-list^1.52.0.

The heading text can be customized with --list-heading:

$ just --list --list-heading $'Cool stuff…\n'
Cool stuff…
    test
    build

And the indentation can be customized with --list-prefix:

$ just --list --list-prefix ····
Available recipes:
····test
····build

The argument to --list-heading replaces both the heading and the newline following it, so it should contain a newline if non-empty. It works this way so you can suppress the heading line entirely by passing the empty string:

$ just --list --list-heading ''
    test
    build

Multiple recipes may be invoked on the command line at once:

build:
  make web

serve:
  python3 -m http.server -d out 8000

$ just build serve
make web
python3 -m http.server -d out 8000

Keep in mind that recipes with parameters will swallow arguments, even if they match the names of other recipes:

build project:
  make {{project}}

serve:
  python3 -m http.server -d out 8000

$ just build serve
make: *** No rule to make target `serve'.  Stop.

The --one flag can be used to restrict command-line invocations to a single recipe:

$ just --one build serve
error: Expected 1 command-line recipe invocation but found 2.

Variables can be overridden from the command line.

os := "linux"

test: build
  ./test --test {{os}}

build:
  ./build {{os}}

$ just
./build linux
./test --test linux

Any number of arguments of the form NAME=VALUE can be passed before recipes:

$ just os=plan9
./build plan9
./test --test plan9

Or you can use the --set flag:

$ just --set os bsd
./build bsd
./test --test bsd

Variables in submodules can be overridden using the ::-separated path to the variable. A variable named bar in a submodule named foo may be overridden with foo::bar=VALUE or --set foo::bar VALUE.

just supports a number of useful command-line options for listing, dumping, and debugging recipes and variables:

$ just --list
Available recipes:
  js
  perl
  polyglot
  python
  ruby
$ just --show perl
perl:
  #!/usr/bin/env perl
  print "Larry Wall says Hi!\n";
$ just --show polyglot
polyglot: python js perl sh ruby

Some command-line options can be set with environment variables

For example, unstable features can be enabled either with the --unstable flag:

$ just --unstable

Or by setting the JUST_UNSTABLE environment variable:

$ export JUST_UNSTABLE=1
$ just

Since environment variables are inherited by child processes, command-line options set with environment variables are inherited by recursive invocations of just, whereas command-line options set with arguments are not.

Consult just --help for which options can be set with environment variables.

The --choose subcommand makes just invoke a chooser to select which recipes to run. Choosers should read lines containing recipe names from standard input and print one or more of those names separated by spaces to standard output.

Because there is currently no way to run a recipe that requires arguments with --choose, such recipes will not be given to the chooser. Private recipes and aliases are also skipped.

The chooser can be overridden with the --chooser flag. If --chooser is not given, then just first checks if $JUST_CHOOSER is set. If it isn't, then the chooser defaults to fzf, a popular fuzzy finder.

Arguments can be included in the chooser, i.e. fzf --exact.

The chooser is invoked in the same way as recipe lines. For example, if the chooser is fzf, it will be invoked with sh -cu 'fzf', and if the shell, or the shell arguments are overridden, the chooser invocation will respect those overrides.

If you'd like just to default to selecting recipes with a chooser, you can use this as your default recipe:

default:
  @just --choose

just normally executes all recipes unless there is an error. The [confirm] attribute^1.17.0 allows recipes to require confirmation in the terminal prior to running. This can be overridden by passing --yes to just, which will automatically confirm any recipes marked by this attribute.

Recipes dependent on a recipe that requires confirmation will not be run if the relied upon recipe is not confirmed, as well as recipes passed after any recipe that requires confirmation.

[confirm]
delete-all:
  rm -rf *

The default confirmation prompt can be overridden with [confirm(PROMPT)]^1.23.0:

[confirm("Are you sure you want to delete everything?")]
delete-everything:
  rm -rf *

The confirmation prompt may also be an expression^1.49.0 which may reference assignments or recipe arguments:

[confirm("Deploy to " + env + "?")]
deploy env:
  echo 'Deploying to {{env}}...'

just can print timestamps before each recipe command:

recipe:
  echo one
  sleep 2
  echo two

$ just --timestamp recipe
[07:28:46] echo one
one
[07:28:46] sleep 2
[07:28:48] echo two
two

By default, timestamps are formatted as HH:MM:SS. The format can be changed with --timestamp-format:

$ just --timestamp recipe --timestamp-format '%H:%M:%S%.3f %Z'
[07:32:11:.349 UTC] echo one
one
[07:32:11:.350 UTC] sleep 2
[07:32:13:.352 UTC] echo two
two

The argument to --timestamp-format is a strftime-style format string, see the chrono library docs for details.

The [timestamp] attribute^1.58.0 can be used to enable timestamps for a specific recipe:

[timestamp]
foo:
  echo hello

$ just foo
[07:28:46] echo hello
hello

Which may include a format string:

[timestamp('%H:%M:%S%.3f')]
foo:
  echo hello

$ just foo
[07:28:46.487] echo hello
hello

Signals are messages sent to running programs to trigger specific behavior. For example, SIGINT is sent to all processes in the terminal foreground process group when ctrl-c is pressed.

just tries to exit when requested by a signal, but it also tries to avoid leaving behind running child processes, two goals which are somewhat in conflict.

If just exits leaving behind child processes, the user will have no recourse but to ps aux | grep for the children and manually kill them, a tedious endeavor.

SIGHUP, SIGINT, and SIGQUIT are generated when the user closes the terminal, types ctrl-c, or types ctrl-\, respectively, and are sent to all processes in the foreground process group.

SIGTERM is the default signal sent by the kill command, and is delivered only to its intended victim.

When a child process is not running, just will exit immediately on receipt of any of the above signals.

When a child process is running, just will wait until it terminates, to avoid leaving it behind.

Additionally, on receipt of SIGTERM, just will forward SIGTERM to any running children^1.41.0, since unlike other fatal signals, SIGTERM, was likely sent to just alone.

Regardless of whether a child process terminates successfully after just receives a fatal signal, just halts execution.

The [continue]^1.54.0 attribute can be used to make just continue execution even if it receives a fatal signal as long as the child process it's running exits successfully.

With no arguments, [continue] handles SIGINT (ctrl-c) so SIGQUIT (ctrl-\) can still be used to stop execution.

With arguments, one or more signals to handle may be given explicitly, as "SIGHUP", "SIGINT", and "SIGQUIT".

In this example, if main.py catches SIGINT and exits successfully, cleanup will still run and just will exit successfully:

[continue]
test: && cleanup
  python3 main.py

cleanup:
  echo cleanup

SIGINFO is sent to all processes in the foreground process group when the user types ctrl-t on BSD-derived operating systems, including macOS, but not Linux.

just responds by printing a list of all child process IDs and commands^1.41.0.

On Windows, just behaves as if it had received SIGINT when the user types ctrl-c. Other signals are unsupported.

Shell completion scripts for Bash, Elvish, Fish, Nushell, PowerShell, and Zsh are available in release archives.

The just binary can also generate the same completion scripts at runtime using just --completions SHELL:

$ just --completions bash > just

The recommended approach is to use the bash-completions package to lazy-load the completion script:

mkdir -p ~/.local/share/bash-completion/completions
just --completions bash > ~/.local/share/bash-completion/completions/just

If bash-completions is not installed, you can source the completion script in your .bashrc:

source <(just --completions bash)

If you use an alias like alias j=just, you should also save the completion script with the name j when lazy-loading:

just --completions bash > ~/.local/share/bash-completion/completions/j

Or if not lazy-loading, add this line after sourcing the completion script in your .bashrc:

complete -F _clap_complete_just -o bashdefault -o default j

In your rc.elv:

set edit:completion:arg-completer[just] = { |@args|
  eval (just --completions elvish | slurp)
  set @result = (edit:completion:arg-completer[just] $@args)
  put $@result
}

Save the completion script to the completions directory to lazy-load it:

mkdir -p ~/.config/fish/completions
just --completions fish > ~/.config/fish/completions/just.fish

First save the completion script:

just --completions nushell | save -f ($nu.default-config-dir | path join just.nu)

Then in config.nu:

source just.nu

In your PowerShell $PROFILE:

just --completions powershell | Out-String | Invoke-Expression

First save the completion script:

mkdir -p ~/.zsh/completions
just --completions zsh > ~/.zsh/completions/_just

Then in your .zshrc:

fpath=(~/.zsh/completions $fpath)
autoload -U compinit
compinit

just can print its own man page with just --man. Man pages are written in roff, a venerable markup language and one of the first practical applications of Unix. If you have groff installed you can view the man page with just --man | groff -mandoc -Tascii | less.

Each justfile has a canonical formatting with respect to whitespace and newlines.

You can overwrite the current justfile with a canonically-formatted version using the --fmt flag:

$ cat justfile
# A lot of blank lines





some-recipe:
  echo "foo"
$ just --fmt
$ cat justfile
# A lot of blank lines

some-recipe:
    echo "foo"

When the justfile is read from standard input with --justfile - or extracted from a markdown file, --fmt prints the formatted justfile to stdout.

Note that formatting is not covered by any backwards compatibility guarantee and is subject to change from time to time.

Recipe bodies are indented with four spaces by default. This can be changed with the --indentation command-line option, the JUST_INDENTATION environment variable, or the indentation setting:

set indentation := "  "

Invoking just --fmt --check runs --fmt in check mode. Instead of overwriting the justfile, just will exit with an exit code of 0 if it is formatted correctly, and will exit with 1 and print a diff if it is not.

You can use the --dump command to output a formatted version of the justfile to stdout:

$ just --dump > formatted-justfile

The --dump command can be used with --dump-format json to print a JSON representation of a justfile.

just will skip invocations of recipes with the [cache] attribute^1.54.0 if it finds an entry matching the invocation in the cache. The [cache] attribute may only be used with script recipes and is currently unstable.

For example, this recipe will be skipped if image.jpg exists and the contents of image.png and the output of convert -version haven't changed since the last run:

set unstable

[script]
[cache(inputs = "image.png", outputs = "image.jpg", extra = `convert -version`)]
convert:
  convert image.png image.jpg

Unlike many other features of just, which are, hopefully, well thought-out and user-friendly, cached recipes are inherently fragile. It is important to understand their limitations before relying on them. Please read this section thoroughly, including the friendly admonitions below.

just will happily skip cached recipes, but it is your responsibility to make sure that this is safe, and that the contents of the cache key capture enough information about recipe invocations for caching to make sense in the first place.

In particular, there are many details about the context in which a recipe runs that are not captured by cache keys.

These include the time, input files, output files, system binaries, operating system version, databases, systems over the network, the DNS, and any of the myriad other things which may change the execution of a computer program.

Attempting to skip execution based on the type of crude heuristics that just employs has a long and sordid history. However, it is an undeniably convenient and powerful tool, and it is provided in the hopes that you find it useful.

The cache is a directory named .justcache alongside the justfile and should not be committed to version control systems. It contains cache entries named HASH.json, where HASH is the BLAKE3 hash of a serialized cache key JSON object.

The keys of the cache key object are:

• body: evaluated recipe body
• environment: map of environment variable names to values
• executor: script interpreter or shebang
• extension: script file extension
• extra: user-supplied value
• inputs: map of file paths to content hashs
• positional: positional arguments
• recipe: ::-separated module path to invoked recipe
• working_directory: current working directory

All keys other than extra and inputs are populated automatically.

Cache key objects for invoked recipes can be printed to standard error with just -vv.

The environment variables included in environment may be controlled with [cache(environment = EXPRESSION])^master.

The value of extra may be supplied with [cache(extra = EXPRESSION)], where EXPRESSION is an arbitrary expression evaluated with recipe arguments in scope. Changes to the value of extra will cause a cache miss.

Before just runs a cached recipe, it creates a cache key, hashes it, and looks for the corresponding cache entry.

If the cache entry is non-empty, it skips the invocation.

If the cache entry does not exist or is empty, it runs the invocation and writes {} to the cache entry.

File locks are taken on cache entries, so concurrent execution of cached recipes by multiple just processes is safe. If two processes run a recipe invocation with the same cache key, the first will take the lock, run the recipe, write to the cache entry, and relinquish the lock. The second will block until the first relinquishes the lock, see that the entry is non-empty, and skip the invocation.

The cache can be bypassed entirely with the --no-cache flag.

The recipe cache is stored in a directory named .justcache alongside the justfile. Deleting it will clear the cache.

The cache can also be cleared with just --clean, which can selectively clear cache entries:

# clear all cache entries
just --clean

# clear cache entries for recipe `foo`
just --clean foo

# clear cache entries for recipe `baz` in submodule `bar`
just --clean bar baz

# clear cache entries for recipes in submodule module `bar`
just --clean bar

# clear cache entries for recipes in submodule module `bar::bob`
just --clean bar bob

# '::'-separated paths may also be used
just --clean bar::bob

By default, environment variables which are exported or unexported in the justfile are included in the cache key, and environment variables exported outside the justfile and inherited by just are not included.

Environment variables can be added or removed from the cache key using [cache(environment = EXPRESSION)]^master.

In this example, the environment variable foo will not be included in the cache key, but PATH will:

set unstable
set lists

export foo := 'bar'

[script]
[cache(environment = ['PATH'])]
build:
  cc lib.c main.c -o main

Input files can be provided with [cache(inputs = FILES)], where FILES is an expression that is evaluated with recipe arguments in scope and whose evaluated elements are paths. Paths may be absolute or relative to the recipe's working directory.

Each input file is hashed with BLAKE3 and added to the inputs cache key, which contains a map of paths to hashes.

Any changes to the contents of an input file changes the cache key, which causes the next invocation to miss the cache and re-run.

Missing inputs and paths to directories are errors.

In this example, the build recipe will re-run if lib.c or main.c change:

set unstable
set lists

[script]
[cache(inputs = ["lib.c", "main.c"])]
build:
  cc lib.c main.c -o main

Output files can be provided with [cache(outputs = FILES)], where FILES is an expression that is evaluated with recipe arguments in scope and whose evaluated elements are paths. Paths may be absolute or relative to the recipe's working directory.

Outputs are not part of the cache key.

All output files must exist for an invocation to be skipped, and after an invocation runs successfully, it is an error if any output file does not exist.

In this example, build re-runs whenever main is missing, and errors if it runs without producing main:

set unstable
set lists

[script]
[cache(inputs = ["lib.c", "main.c"], outputs = "main")]
build:
  cc lib.c main.c -o main

clean:
  rm -f main

This forces build to re-run if main is deleted by clean.

Recipes, mod statements, and aliases may be annotated with attributes that change their behavior.

A recipe can have multiple attributes, either on multiple lines:

[no-cd]
[private]
foo:
    echo "foo"

Or separated by commas on a single line^1.14.0:

[no-cd, private]
foo:
    echo "foo"

Attributes with a single argument may be written with a colon:

[group: 'bar']
foo:

Settings control interpretation and execution. Each setting may be specified at most once, anywhere in the justfile.

For example:

set shell := ["zsh", "-cu"]

foo:
  # this line will be run as `zsh -cu 'ls **/*.txt'`
  ls **/*.txt

Boolean settings can be written as:

set NAME

Which is equivalent to:

set NAME := true

Non-boolean settings can be set to both strings and expressions^1.46.0.

However, because settings affect the behavior of backticks and many functions, those expressions may not contain backticks or function calls, directly or transitively via reference.

• arch() — Instruction set architecture. Possible values are: "aarch64", "arm", "asmjs", "hexagon", "mips", "msp430", "powerpc", "powerpc64", "s390x", "sparc", "wasm32", "x86", "x86_64", and "xcore".
• num_cpus()^1.15.0 — Number of logical CPUs.
• os() — Operating system. Possible values are: "android", "bitrig", "dragonfly", "emscripten", "freebsd", "haiku", "ios", "linux", "macos", "netbsd", "openbsd", "solaris", and "windows".
• os_family() — Operating system family; possible values are: "unix" and "windows".

For example:

system-info:
  @echo "This is an {{arch()}} machine."

$ just system-info
This is an x86_64 machine.

The os_family() function can be used to create cross-platform justfiles that work on various operating systems. For an example, see cross-platform.just file.

• shell(command, args...)^1.27.0 returns the standard output of shell script command with zero or more positional arguments args. The shell used to interpret command is the same shell that is used to evaluate recipe lines, and can be changed with set shell := […].

  command is passed as the first argument, so if the command is 'echo $@', the full command line, with the default shell command sh -cu and args 'foo' and 'bar' will be:

  'sh' '-cu' 'echo $@' 'echo $@' 'foo' 'bar'
  

  This is so that $@ works as expected, and $1 refers to the first argument. $@ does not include the first positional argument, which is expected to be the name of the program being run.

# arguments can be variables or expressions
file := '/sys/class/power_supply/BAT0/status'
bat0stat := shell('cat $1', file)

# commands can be variables or expressions
command := 'wc -l'
output := shell(command + ' "$1"', 'main.c')

# arguments referenced by the shell command must be used
empty := shell('echo', 'foo')
full := shell('echo $1', 'foo')
error := shell('echo $1')

# Using python as the shell. Since `python -c` sets `sys.argv[0]` to `'-c'`,
# the first "real" positional argument will be `sys.argv[2]`.
set shell := ["python3", "-c"]
olleh := shell('import sys; print(sys.argv[2][::-1])', 'hello')

• env(key)^1.15.0 — Retrieves the environment variable with name key, aborting if it is not present.

home_dir := env('HOME')

test:
  echo "{{home_dir}}"

$ just
/home/user1

• env(key, default)^1.15.0 — Retrieves the environment variable with name key, returning default if it is not present.
• env_var(key) — Deprecated alias for env(key).
• env_var_or_default(key, default) — Deprecated alias for env(key, default).

• require(name)^1.39.0 — Search directories in the PATH environment variable for the executable name and return its full path, or halt with an error if no executable with name exists.

  bash := require("bash")
  
  @test:
      echo "bash: '{{bash}}'"

  $ just
  bash: '/bin/bash'

• which(name)^1.39.0 — Search directories in the PATH environment variable for the executable name and return its full path, or the empty list if not found. Requires set lists^1.53.0.

  set unstable
  set lists
  
  bosh := which("bosh")
  
  @test:
      echo "bosh: '{{bosh}}'"

  $ just
  bosh: ''

• is_dependency() - Returns the string true if the current recipe is being run as a dependency of another recipe, rather than being run directly, otherwise returns the string false.

• num_jobs()^1.56.0 — The value of --jobs or the empty list, [], if it was not passed.

• recipe_name()^1.53.0 - Returns the name of the current recipe.

• invocation_directory() - Retrieves the absolute path to the current directory when just was invoked, before just changed it (chdir'd) prior to executing commands. On Windows, invocation_directory() uses cygpath to convert the invocation directory to a Cygwin-compatible /-separated path. Use invocation_directory_native() to return the verbatim invocation directory on all platforms.

For example, to call rustfmt on files just under the "current directory" (from the user/invoker's perspective), use the following recipe:

rustfmt:
  find {{invocation_directory()}} -name \*.rs -exec rustfmt {} \;

Alternatively, if your command needs to be run from the current directory, you could use (e.g.):

build:
  cd {{invocation_directory()}}; ./some_script_that_needs_to_be_run_from_here

• invocation_directory_native() - Retrieves the absolute path to the current directory when just was invoked, before just changed it (chdir'd) prior to executing commands.

• justfile() - Retrieves the path of the current justfile.

• justfile_directory() - Retrieves the path of the parent directory of the current justfile.

For example, to run a command relative to the location of the current justfile:

script:
  {{justfile_directory()}}/scripts/some_script

• source_file()^1.27.0 - Retrieves the path of the current source file.

• source_directory()^1.27.0 - Retrieves the path of the parent directory of the current source file.

source_file() and source_directory() behave the same as justfile() and justfile_directory() in the root justfile, but will return the path and directory, respectively, of the current import or mod source file when called from within an import or submodule.

• module_file() - Returns the path of the current module file.

• module_directory() - Returns the path of the parent directory of the current module file.

• module_path() - Returns the ::-separated path to the current module.

module_file() and module_directory() behave the same as justfile() and justfile_directory() in the root justfile, but will return the path and directory, respectively, of the current mod source file when called from within a submodule.

• just_executable() - Absolute path to the just executable.
• just_pid() - Process ID of the just executable.
• just_version()^1.55.0 - Version of the just executable.

For example:

just-info:
  @echo The executable is at: {{just_executable()}}
  @echo The process ID is: {{ just_pid() }}
  @echo The version is: {{ just_version() }}

$ just
The executable is at: /bin/just
The process ID is: 420
The version is: 1.54.0

• append(suffix, s)^1.27.0 - Append suffix to whitespace-separated strings in s. append('/src', 'foo bar baz') → 'foo/src bar/src baz/src'
• prepend(prefix, s)^1.27.0 - Prepend prefix to whitespace-separated strings in s. prepend('src/', 'foo bar baz') → 'src/foo src/bar src/baz'
• encode_uri_component(s)^1.27.0 - Percent-encode characters in s except [A-Za-z0-9_.!~*'()-], matching the behavior of the JavaScript encodeURIComponent function.
• quote(s) - Replace all single quotes with '\'' and prepend and append single quotes to s. This is sufficient to escape special characters for many shells, including most Bourne shell descendants.
• replace(s, from, to) - Replace all occurrences of from in s with to.
• replace_regex(s, regex, replacement) - Replace all occurrences of regex in s with replacement. Regular expressions are provided by the Rust regex crate. See the syntax documentation for usage examples. Capture groups are supported. The replacement string uses Replacement string syntax.
• trim(s) - Remove leading and trailing whitespace from s.
• trim_end(s) - Remove trailing whitespace from s.
• trim_end_match(s, substring) - Remove suffix of s matching substring.
• trim_end_matches(s, substring) - Repeatedly remove suffixes of s matching substring.
• trim_start(s) - Remove leading whitespace from s.
• trim_start_match(s, substring) - Remove prefix of s matching substring.
• trim_start_matches(s, substring) - Repeatedly remove prefixes of s matching substring.

• capitalize(s)^1.7.0 - Convert first character of s to uppercase and the rest to lowercase.
• kebabcase(s)^1.7.0 - Convert s to kebab-case.
• lowercamelcase(s)^1.7.0 - Convert s to lowerCamelCase.
• lowercase(s) - Convert s to lowercase.
• shoutykebabcase(s)^1.7.0 - Convert s to SHOUTY-KEBAB-CASE.
• shoutysnakecase(s)^1.7.0 - Convert s to SHOUTY_SNAKE_CASE.
• snakecase(s)^1.7.0 - Convert s to snake_case.
• titlecase(s)^1.7.0 - Convert s to Title Case.
• uppercamelcase(s)^1.7.0 - Convert s to UpperCamelCase.
• uppercase(s) - Convert s to uppercase.

• absolute_path(path) - Absolute path to relative path in the working directory. absolute_path("./bar.txt") in directory /foo is /foo/bar.txt.
• canonicalize(path)^1.24.0 - Canonicalize path by resolving symlinks and removing ., .., and extra /s where possible.
• extension(path) - Extension of path. extension("/foo/bar.txt") is txt.
• file_name(path) - File name of path with any leading directory components removed. file_name("/foo/bar.txt") is bar.txt.
• file_stem(path) - File name of path without extension. file_stem("/foo/bar.txt") is bar.
• parent_directory(path) - Parent directory of path. parent_directory("/foo/bar.txt") is /foo.
• without_extension(path) - path without extension. without_extension("/foo/bar.txt") is /foo/bar.

These functions can fail, for example if a path does not have an extension, which will halt execution.

• clean(path) - Simplify path by removing extra path separators, intermediate . components, and .. where possible. clean("foo//bar") is foo/bar, clean("foo/..") is ., clean("foo/./bar") is foo/bar.
• join(a, b…) - This function uses / on Unix and \ on Windows, which can lead to unwanted behavior. The / operator, e.g., a / b, which always uses /, should be considered as a replacement unless \s are specifically desired on Windows. Join path a with path b. join("foo/bar", "baz") is foo/bar/baz. Accepts two or more arguments.

• path_exists(path) - Returns the string true if the path points at an existing entity and the string false otherwise. Traverses symbolic links, and returns the string false if the path is inaccessible or points to a broken symlink.
• read(path)^1.39.0 - Returns the content of file at path as a string.

• assert(CONDITION, EXPRESSION)^1.27.0 - Error with message EXPRESSION if CONDITION is false. EXPRESSION may be omitted^1.53.0,
• error(message) - Abort execution and report error message to user.

• blake3(string)^1.25.0 - Return BLAKE3 hash of string as hexadecimal string.
• blake3_file(path)^1.25.0 - Return BLAKE3 hash of file at path as hexadecimal string.
• sha256(string) - Return the SHA-256 hash of string as hexadecimal string.
• sha256_file(path) - Return SHA-256 hash of file at path as hexadecimal string.
• uuid() - Generate a random version 4 UUID.

• choose(n, alphabet)^1.27.0 - Generate a string of n randomly selected characters from alphabet, which may not contain repeated characters. For example, choose('64', HEX) will generate a random 64-character lowercase hex string.

• datetime(format)^1.30.0 - Return local time with format.
• datetime_utc(format)^1.30.0 - Return UTC time with format.

The arguments to datetime and datetime_utc are strftime-style format strings, see the chrono library docs for details.

• semver_matches(version, requirement)^1.16.0 - Check whether a semantic version, e.g., "0.1.0" matches a requirement, e.g., ">=0.1.0", returning the string "true" if so and the string "false" otherwise.

• style(styles)^1.37.0 - Return a terminal escape sequence combining the named styles in styles.

  The styles supported by version 1.37.0 and later can be used to duplicate just's own styles:

  • command: echoed recipe lines
  • error: errors
  • warning: warnings

  Additional styles supported by ^1.55.0 and later include named colors:

  • black
  • blue
  • cyan
  • green
  • magenta
  • red
  • white
  • yellow

  The 256 indexed colors, written as integers between 0 and 255, e.g., 1 or 67.

  The 24-bit colors, written as #RRGGBB or #RGB hex codes, e.g., #065535 or #AAA.

  And display properties:

  • blink
  • bold
  • dim
  • hidden
  • italic
  • reverse
  • strikethrough
  • underline

  Two stream names^1.55.0 gate the style on whether just would color the output stream, determined by --color, JUST_COLOR, and whether the stream is connected to a terminal:

  • stdout
  • stderr

  All color styles color the foreground by default, and come in explicit foreground variants prefixed with fg: and background variants prefixed with bg:, e.g., bg:blue, fg:133, and #FFF.

  styles may be a list of styles, in which case all listed styles are combined to produce the final escape sequence.

  Note that the escape sequence returned by style(styles) are prefixes. You can use the NORMAL constant to reset the style after use:

  error message:
    echo '{{style("error") + message + NORMAL}}'

• style(styles, text)^1.55.0 Style text with styles as in the one-argument form. The style is reset automatically, so use of NORMAL to reset the terminal is not needed:

  error message:
    echo '{{style("error", message)}}'

These functions^1.23.0 return paths to user-specific directories for things like configuration, data, caches, executables, and the user's home directory.

On Unix, these functions follow the XDG Base Directory Specification.

On macOS and Windows, these functions return the system-specified user-specific directories. For example, cache_directory() returns ~/Library/Caches on macOS and {FOLDERID_LocalAppData} on Windows.

See the dirs crate for more details.

• cache_directory() - The user-specific cache directory.
• config_directory() - The user-specific configuration directory.
• config_local_directory() - The local user-specific configuration directory.
• data_directory() - The user-specific data directory.
• data_local_directory() - The local user-specific data directory.
• executable_directory() - The user-specific executable directory.
• home_directory() - The user's home directory.
• runtime_directory() - The user-specific runtime directory. Only defined on Linux.

If you would like to use XDG base directories on all platforms you can use the env(…) function with the appropriate environment variable and fallback, although note that the XDG specification requires ignoring non-absolute paths, so for full compatibility with spec-compliant applications, you would need to do:

xdg_config_dir := if env('XDG_CONFIG_HOME', '') =~ '^/' {
  env('XDG_CONFIG_HOME')
} else {
  home_directory() / '.config'
}

A number of constants are predefined:

@foo:
  echo {{HEX}}

$ just foo
0123456789abcdef

Constants starting with \e are ANSI escape sequences.

CLEAR clears the screen, similar to the clear command. The rest are of the form \e[Nm, where N is an integer, and set terminal display attributes.

Terminal display attribute escape sequences can be combined, for example text weight BOLD, text style STRIKETHROUGH, foreground color CYAN, and background color BG_BLUE. They should be followed by NORMAL, to reset the terminal back to normal.

Escape sequences should be quoted, since [ is treated as a special character by some shells.

@foo:
  echo '{{BOLD + STRIKETHROUGH + CYAN + BG_BLUE}}Hi!{{NORMAL}}'

A changelog for the latest release is available in CHANGELOG.md. Changelogs for previous releases are available on the releases page. just --changelog can also be used to make a just binary print its changelog.

watchexec can re-run any command when files change.

To re-run the recipe foo when any file changes:

watchexec just foo

See watchexec --help for more info, including how to specify which files should be watched for changes.

For lightning-fast command running, put alias j=just in your shell's configuration file.

A non-normative grammar of justfiles can be found in GRAMMAR.md.

Before just was a fancy Rust program it was a tiny shell script that called make. You can find the old version in contrib/just.sh.

The following export statement gives just recipes access to local Node module binaries, and makes just recipe commands behave more like script entries in Node.js package.json files:

export PATH := "./node_modules/.bin:" + env_var('PATH')

On Windows, all functions that return paths, except invocation_directory(), will return \-separated paths. When not using PowerShell or cmd.exe, these paths should be quoted to prevent the \s from being interpreted as character escapes:

ls:
    echo '{{absolute_path(".")}}'

cygpath.exe is an executable included in some distributions of Unix userlands for Windows, including Cygwin and Git for Windows.

just uses cygpath.exe in two places:

For backwards compatibility, invocation_directory(), uses cygpath.exe to convert the invocation directory into a unix-style /-separated path. Use invocation_directory_native() to get the native, Windows-style path. On unix, invocation_directory() and invocation_directory_native() both return the same unix-style path.

cygpath.exe is also used to convert Unix-style shebang lines into Windows paths. As an alternative, the [script] attribute can be used, which does not depend on cygpath.exe.

If cygpath.exe is available, you can use it to convert between path styles:

foo_unix := '/hello/world'
foo_windows := shell('cygpath --windows $1', foo_unix)

bar_windows := 'C:\hello\world'
bar_unix := shell('cygpath --unix $1', bar_windows)

echo can be used to print strings, but because it processes escape sequences, like \n, and different implementations of echo recognize different escape sequences, using printf is often a better choice.

printf takes a C-style format string and any number of arguments, which are interpolated into the format string.

This can be combined with indented, triple quoted strings to emulate shell heredocs.

Substituting complex strings into recipe bodies with {{…}} can also lead to trouble as it may be split by the shell into multiple arguments depending on the presence of whitespace and quotes. Exporting complex strings as environment variables and referring to them with "$NAME", note the double quotes, can also help.

Putting all this together, to print a string verbatim to standard output, with all its various escape sequences and quotes undisturbed:

export FOO := '''
  a complicated string with
  some dis\tur\bi\ng escape sequences
  and "quotes" of 'different' kinds
'''

bar:
  printf %s "$FOO"

A skill for agents is available in skills/just and may be installed manually or with npx skills add casey/just --global.

There is no shortage of command runners! Some more or less similar alternatives to just include:

• make: The Unix build tool that inspired just. There are a few different modern day descendants of the original make, including FreeBSD Make and GNU Make.
• task: A YAML-based command runner written in Go.
• maid: A Markdown-based command runner written in JavaScript.
• microsoft/just: A JavaScript-based command runner written in JavaScript.
• cargo-make: A command runner for Rust projects.
• mmake: A wrapper around make with a number of improvements, including remote includes.
• robo: A YAML-based command runner written in Go.
• mask: A Markdown-based command runner written in Rust.
• makesure: A simple and portable command runner written in AWK and shell.
• haku: A make-like command runner written in Rust.
• mise: A development environment tool manager written in Rust supporting tasks in TOML files and standalone scripts.

Metadata in the form of lists of strings may be attached to recipes with the [metadata(METADATA)] attribute^1.42.0:

[metadata("hello", "goodbye")]
foo:

Metadata can be read using just --dump --dump-format json.

uv is an excellent cross-platform python project manager, written in Rust.

Using the [script] attribute and script-interpreter setting, just can easily be configured to run Python recipes with uv:

set script-interpreter := ['uv', 'run', '--script']

[script]
hello:
  print("Hello from Python!")

[script]
goodbye:
  # /// script
  # requires-python = ">=3.11"
  # dependencies=["sh"]
  # ///
  import sh
  print(sh.echo("Goodbye from Python!"), end='')

Of course, a shebang also works:

hello:
  #!/usr/bin/env -S uv run --script
  print("Hello from Python!")

Some tools require an activation step, such as Python virtual environments:

. .venv/bin/activate

Because these tools modify the environment of a running shell, it is not possible for just to perform this activation step for you. However, there are some workarounds.

The best workaround for Python environment management is to switch to uv. uv sets up the correct environment for each command, so no activation step is needed.

If that isn't possible, and for other tools, you can create a shared prelude and include it in script recipes that need it. It can span multiple lines and include any number of steps:

prelude := '''
  set -eux
  . .venv/bin/activate
'''

[script]
run:
  {{ prelude }}
  python script.py

This workaround doesn't work with shell recipes, which spawn a new shell for each command.

just welcomes your contributions! just is released under the maximally permissive CC0 public domain dedication and fallback license, so your changes must also be released under this license.

just is written in Rust. Use rustup to install a Rust toolchain.

just is extensively tested. All new features must be covered by unit or integration tests. Unit tests are under src, live alongside the code being tested, and test code in isolation. Integration tests are in the tests directory and test the just binary from the outside by invoking just on a given justfile and set of command-line arguments, and checking the output.

You should write whichever type of tests are easiest to write for your feature while still providing good test coverage.

Unit tests are useful for testing new Rust functions that are used internally and as an aid for development. A good example are the unit tests which cover the unindent() function, used to unindent triple-quoted strings and backticks. unindent() has a bunch of tricky edge cases which are easy to exercise with unit tests that call unindent() directly.

Integration tests are useful for making sure that the final behavior of the just binary is correct. unindent() is also covered by integration tests which make sure that evaluating a triple-quoted string produces the correct unindented value. However, there are no integration tests for all possible cases. These are covered by faster, more concise unit tests that call unindent() directly.

Integration tests use the Test struct, a builder which allows for easily invoking just with a given justfile, arguments, and environment variables, and checking the program's stdout, stderr, and exit code.

1. Make sure the feature is wanted. There should be an open issue about the feature with a comment from @casey saying that it's a good idea or seems reasonable. If there isn't, open a new issue and ask for feedback.

   There are lots of good features which can't be merged, either because they aren't backwards compatible, have an implementation which would overcomplicate the codebase, or go against just's design philosophy.

2. Settle on the design of the feature. If the feature has multiple possible implementations or syntaxes, make sure to nail down the details in the issue.

3. Clone just and start hacking. The best workflow is to have the code you're working on in an editor alongside a job that re-runs tests whenever a file changes. You can run such a job by installing cargo-watch with cargo install cargo-watch and running just watch test.

4. Add a failing test for your feature. Most of the time this will be an integration test which exercises the feature end-to-end. Look for an appropriate file to put the test in tests, or add a new file in tests and add a mod statement importing that file in tests/lib.rs.

5. Implement the feature.

6. Run just ci to make sure that all tests, lints, and checks pass. Requires mdBook and mdbook-linkcheck.

7. Open a PR with the new code that is editable by maintainers. PRs often require rebasing and minor tweaks. If the PR is not editable by maintainers, each rebase and tweak will require a round trip of code review. Your PR may be summarily closed if it is not editable by maintainers.

8. Incorporate feedback.

9. Enjoy the sweet feeling of your PR getting merged!

Feel free to open a draft PR at any time for discussion and feedback.

Here are some hints to get you started with specific kinds of new features, which you can use in addition to the contribution workflow above.

1. Write a new integration test in tests/attributes.rs.

2. Add a new variant to the Attribute enum.

3. Implement the functionality of the new attribute.

4. Run just ci to make sure that all tests pass.

Janus is a tool for checking whether a change to just breaks or changes the interpretation of existing justfiles. It collects and analyzes public justfiles on GitHub.

Before merging a particularly large or gruesome change, Janus should be run to make sure that nothing breaks. Don't worry about running Janus yourself, Casey will happily run it for you on changes that need it.

The minimum supported Rust version, or MSRV, is current stable Rust. It may build on older versions of Rust, but this is not guaranteed.

New releases of just are made frequently so that users quickly get access to new features.

Release commit messages use the following template:

Release x.y.z

- Bump version: x.y.z → x.y.z
- Update changelog
- Update changelog contributor credits
- Update dependencies
- Update version references in readme

make has some behaviors which are confusing, complicated, or make it unsuitable for use as a general command runner.

One example is that under some circumstances, make won't actually run the commands in a recipe. For example, if you have a file called test and the following makefile:

test:
  ./test

make will refuse to run your tests:

$ make test
make: `test' is up to date.

make assumes that the test recipe produces a file called test. Since this file exists and the recipe has no other dependencies, make thinks that it doesn't have anything to do and exits.

To be fair, this behavior is desirable when using make as a build system, but not when using it as a command runner. You can disable this behavior for specific targets using make's built-in .PHONY target name, but the syntax is verbose and can be hard to remember. The explicit list of phony targets, written separately from the recipe definitions, also introduces the risk of accidentally defining a new non-phony target. In just, all recipes are treated as if they were phony.

Other examples of make's idiosyncrasies include the difference between = and := in assignments, the confusing error messages that are produced if you mess up your makefile, needing $$ to use environment variables in recipes, and incompatibilities between different flavors of make.