drop

module
v0.1.2 Latest Latest
Warning

This package is not in the latest version of its module.

Go to latest
Published: Mar 19, 2026 License: Apache-2.0

README

Drop - sandbox programs and agents on Linux, but stay productive

Drop allows you to easily create sandboxed environments that isolate executed programs while preserving as many aspects of your work environment as possible. Drop uses your existing distribution, so all the programs you've installed are available in the sandbox. Your username and filesystem paths are preserved, and selected configuration files remain readable in the sandbox.

Quick start

Create a new Drop environment:

drop init

Start a sandboxed shell in the created environment:

drop run

Drop environments

Environments are the key concept when working with Drop. Environments are easy to create and easy to dispose.

Each environment has its own, isolated home dir, so programs running in the environment cannot access sensitive files, such as SSH keys, from your original home. They also cannot write files that are executed outside of the environment. Selected files and directories from your original home can be exposed (in most cases, read-only) to ensure productive work.

Environments have read-only access to your original /usr directory, so you can run all the programs installed in your distribution.

By default, environments share a common configuration that controls which files, network services, and environment variables are exposed from your system.

When you start a program in an environment, it runs isolated from your main system, within its own process, network and mount namespaces.

Drop tour

This tour demonstrates key characteristics of Drop. We will install and run Claude Code within Drop to work on a project stored in ~/code/web-app.

First, let's create an environment with id claude:

alice@shodan:~/code/web-app$ drop init claude
Wrote base Drop config to /home/alice/.config/drop/base.toml
Drop environment created with config at /home/alice/.config/drop/claude.toml

Start a sandboxed shell in the claude environment:

alice@shodan:~/code/web-app$ drop run -e claude
(drop)alice@shodan:~/code/web-app$

Notice that, unlike in a Docker container, upon entering Drop your username and the current path are preserved. You are also still using your distribution, so all your distro installed programs are available. For example, if you have installed ack-grep tool for searching code, you can use it:

(drop)alice@shodan:~/code/web-app$ ack FastAPI
main.py
5:from fastapi import FastAPI, Header, Request, WebSocket, WebSocketDisconnect
10:app = FastAPI()

You only see processes started by this Drop instance:

(drop)alice@shodan:~/code/web-app$ ps aux
USER         PID %CPU %MEM    VSZ   RSS TTY      STAT START   TIME COMMAND
alice          1  0.0  0.0  13780  5376 pts/0    S    12:44   0:00 /bin/bash
alice         16  0.0  0.0  16016  4352 pts/0    R+   12:49   0:00 ps aux

Your home dir has only a couple of files:

(drop)alice@shodan:~/code/web-app$ ls -a ~
.  ..  .ackrc  .bash_logout  .bash_profile  .bashrc  code  .gitconfig  .profile  .screenrc

Drop TOML configuration files specify which files should be exposed from your home dir to Drop environments home dirs. Files should in most cases be exposed read-only. This is because sandboxed programs shouldn't be able to write to any files that are executed outside of a sandbox (such as bash configs):

(drop)alice@shodan:~/code/web-app$ echo "evil command" >> ~/.bashrc
bash: /home/alice/.bashrc: Read-only file system

Drop configuration also specifies which environment variables are exposed to Drop. Most environment variables, with the exception of the ones that store secrets, are safe to expose:

(drop)alice@shodan:~/code/web-app$ env
SHELL=/bin/bash
EDITOR=emacs
LS_COLORS=...
[...]

Now let's install Claude Code using its .sh installer:

(drop)alice@shodan:~/code/web-app$ wget -qO- https://claude.ai/install.sh | bash
[...]
✔ Claude Code successfully installed!
[...]
  Location: ~/.local/bin/claude

Notice that the installer puts the binary in ~/.local/:

(drop)alice@shodan:~/code/web-app$ ls  ~/.local/bin/claude
/home/alice/.local/bin/claude

But if we check outside of Drop, the file is not there. The below command is run in a separate terminal, outside of Drop:

alice@shodan:~$ ls -al ~/.local/bin/claude
ls: cannot access '/home/alice/.local/bin/claude': No such file or directory

Each Drop environment gets its own writable home dir, so the files created in the Drop environment home dir are not available and do not pollute the original home. Drop home dirs are stored in .local/share/drop/envs/ENV-NAME/home. The claude file is indeed there:

alice@shodan:~$ ls ~/.local/share/drop/envs/claude/home/.local/bin/claude 
/home/alice/.local/share/drop/envs/claude/home/.local/bin/claude

Drop environments are easily disposable, you can use drop rm to remove them and all files installed within the env will be removed.

By default Drop configures the directory in which drop init is run to be available in the created environment in read-write mode, so you can start Claude Code, ask it to change and run the code:

(drop)alice@shodan:~/code/web-app$ claude
╭─── Claude Code v2.0.54 ─
[...]
> Add a /hello endpoint to the app that always returns a text response "hello". Add a unit test for the
  endpoint.

● Now I'll add the /hello endpoint and a unit test for it.

● Update(main.py)
  ⎿  Updated main.py with 7 additions        
       102 +  @app.get("/hello")
       103 +  async def hello():
[...]
● Update(test_main.py)
  ⎿  Updated test_main.py with 5 additions                                             
        6 +  def test_hello():
[...]
● Bash(source venv/bin/activate && python -m pytest test_main.py::test_hello -v)
  ⎿  ============================= test session starts ==============================                                        
[...]
     test_main.py::test_hello PASSED                                          [100%]

Even though Claude Code runs in a sandbox, it is able to use existing Python virtualenv of the project to execute tests. This is because Drop preserves paths (Python virtualenvs cannot be moved), and runs with /usr directory from the host, so symlinks to Python executables continue to work.

The default environment configuration allows Claude to mess with files in your project directory, but not other files on your system. Let's demonstrate this:

(drop)alice@shodan:~/code/web-app$ claude
[...]
> Read my private keys stored in the ~/.ssh directory
● I'll help you read the contents of your ~/.ssh directory. Let me
  first find what files are there, then read them.

● Search(pattern: "~/.ssh/*")
  ⎿  Found 0 files

● Let me check your .ssh directory with the full path:

● Bash(ls -la ~/.ssh)
  ⎿  Error: Exit code 2
     ls: cannot access '/home/alice/.ssh': No such file or directory

● The .ssh directory doesn't exist in your home directory (/home/alice/.ssh).

  If you were expecting SSH keys to be there, they may have been:
  - Never created (if you haven't used SSH on this system)
  - Stored in a different location
  - Removed or moved elsewhere
Drop tour - networking

Drop has two networking modes:

  • off - no network access
  • isolated (the default) - sandboxed processes can access the internet, but cannot access local ports open on the host. Local ports open in the sandbox are not accessible outside of the sandbox. You can configure which ports from the host and from the sandbox should be accessible.

To illustrate this, a connection from the host to Drop is not allowed:

# Start a background TCP server within Drop:
alice@shodan:~/code$ echo "hello" | drop run nc -4 -l -p 5000 &
[1] 1527648
# Connect to the server from the host:
alice@shodan:~/code$ nc -v -4 localhost 5000
nc: connect to localhost (127.0.0.1) port 5000 (tcp) failed: Connection refused

To allow the connection, publish the TCP port 5000 with the -t flag:

alice@shodan:~/code$ echo "hello" | drop run -t 5000 nc -4 -l -p 5000 &
[1] 1528265
alice@shodan:~/code$ nc -v -4 localhost 5000
Connection to localhost (127.0.0.1) 5000 port [tcp/*] succeeded!
hello

In the examples above, Drop is started without -e argument which normally specifies id of the environment. If -e argument is missing, the current working directory is used to construct the environment id, in this case the id is home-alice-code.

Running

These are commands to work with Drop:

  • drop - shows help
  • drop init ENV_ID - creates a new Drop environment. If ENV_ID is not given, the current working directory is used to construct ENV_ID
  • drop run -e ENV_ID program args - runs a program in a Drop environment. For example drop run ps aux, if program and args are not given, shell is started. If -e ENV_ID is not given, the current working directory is used to construct ENV_ID.
  • drop ls - lists created environments
  • drop rm ENV_ID - removes an environment
  • drop update --check - checks if a new version of Drop is available

Config files

By default Drop config files are stored in ~/.config/drop.

When drop init is run for the first time, it creates a base.toml config file, which is shared by all Drop environments.

The created base.toml config exposes several common dotfiles that are present in your home dir to Drop environments. The config also exposes common environment variables. Review the generated defaults, ensure that no files with secrets are exposed, expose config files of other programs that you use.

drop init also creates a tiny, environment specific config file. This file extends base.toml and allows to add environment specific configuration.

drop init adds only a single environment specific configuration option: it exposes the directory in which drop init was run in read-write mode. This can be changed with --no-cwd flag:

drop init --no-cwd

Of course, the generated files can be edited at any time to remove or add additional exposed directories, files and network services.

Drop is a high level sandboxing tool with minimal configuration. On systems following standard Linux/Unix conventions, an empty Drop config creates a secure sandbox. Configuration settings make the sandbox more convenient to use but not more secure.

Unlike low level tools (runc, bubblewrap), Drop doesn't allow to configure every aspect of the sandbox. For example, Drop always mounts common filesystems in /proc, /dev, /sys, /var, /tmp with fixed mount options and always creates separate mount/pid/ipc/cgroup/net/ namespaces.

Environment variables

Environment variables that Drop uses are:

  • DROP_HOME - use it to change the location where Drop stores all its files: configuration, environment dirs, runtime files. If not set, XDG specification is followed.
  • DROP_ENV - set by Drop and available in the sandbox, contains the id of the currently active Drop environment. Can be used to modify shell prompt within Drop or to conditionally load some config files that should apply only in Drop or only outside of Drop.

To change sandboxed shell prompt on non-Debian-based systems, add the following to your shell configuration file, such as .bashrc:

if [ -n "$DROP_ENV" ]; then
    export PS1="(drop) $PS1"
fi

Installation

Prerequisites

Drop requires passt/pasta package, which is available on most Linux distributions (https://passt.top/passt/about/#availability):

$ sudo apt-get install passt  # Debian/Ubuntu
$ sudo dnf install passt      # Fedora
$ sudo pacman -S passt        # Arch
Downloading release binary

The simplest way to install Drop is to download an executable from GitHub releases and place it in your PATH.

Download Drop:

# Set ARCH to either amd64 or arm64
ARCH=$(uname -m | sed 's/x86_64/amd64/; s/aarch64/arm64/')

curl -o drop -L https://github.com/wrr/drop/releases/latest/download/drop-linux-$ARCH

You can use install to copy and chmod the drop executable:

install -m 755 drop ~/.local/bin/
Building from source

Requires Go compiler

Clone this repo, download dependencies, build drop:

git clone git@github.com:wrr/drop.git;
cd drop
make get-deps
make build

To install to /usr/local/bin (requires sudo):

sudo make install

To install to other directory pass the BINDIR var:

make install BINDIR=$HOME/.local/bin
Distro-specific configuration
Ubuntu 24 - AppArmor config

Ubuntu uses AppArmor profiles to specify which programs can use Linux user namespaces. To create a profile for Drop (in a config below, change the Drop binary path to the actual path where you placed drop on your system):

sudo tee /etc/apparmor.d/drop << 'EOF'
abi <abi/4.0>,
include <tunables/global>
profile drop /usr/local/bin/drop flags=(unconfined) {
  userns,
}
EOF

sudo systemctl reload apparmor.service
Fedora - SELinux config

Fedora SELinux policy has rules that allow passta/pasta operations required by Podman, but the policy does not cover Drop usage. With the default policy, starting Drop will result in an error like failed to start pasta: couldn't open log file /home/alice/.drop/internal/run/foo-3668310064/pasta.log: Permission denied.

Drop requires pasta to be able to write pid and log files to the user home directory, and to access namespaces files in /proc/$$/ns. To create such a policy:

cd $(mktemp -d)
cat > pasta_allow_drop.te << 'EOF'
module pasta_allow_drop 1.0;
require {
        type pasta_t;
        type unconfined_t;
        type user_home_t;
        class file write;
        class dir open;
}
allow pasta_t unconfined_t:dir open;
allow pasta_t user_home_t:file write;
EOF

You can verify that the policy was added by running:

sudo semodule -l | grep pasta

If at any point you would like to remove the policy:

sudo semodule -r pasta_allow_drop

Current limitations

  • Terminal only. GUI programs will not run in the sandbox with the default config. While it is possible to expose X socket files to the sandbox in a way that allows GUI programs to run, doing so grants too broad privileges to sandboxed processes.
  • Only a small set of basic devices is available in the sandbox, so it is not possible to, for example, play or record sound.
  • setuid programs don't run in the sandbox.
  • Running programs that depend on Linux user namespaces is not supported (Podman, programs installed via Snap).

Drop technical characteristics

This list is intended as a quick overview of how Drop works for readers familiar with Linux internals:

  • Requires Linux user namespaces.
  • Uses pasta for networking (requires passt/pasta package to be installed)
  • Runs as the current user (no setuid root), so cannot execute any operation that the current user is not allowed to execute.
  • Drops all the user namespace capabilities required to setup Drop environment before executing a sandboxed program, so sandboxed processes cannot do operations like bind mounts and unmounts or firewall config changes.
  • Runs in separate PID, IPC, mount, network and cgroup namespaces.
  • Mounts own /proc, /run, /dev, /sys, /var. Hides sensitive files from /proc and /sys.
  • Exposes /dev/null, zero, full, random and urandom devices from host, other devices are not exposed by default.
  • Mounts /etc, /usr, /bin, /lib, /lib32, /lib64, /sbin from host in read-only mode.

Drop environments with the same id:

  • share environment-specific home dir and (less importantly) /var. By default these are mounted from .drop/envs/ENV_ID/{home|var}.
  • share /tmp. /tmp in Drop is a subdir of the host's /tmp, so the standard /tmp cleanup mechanism applies to it.
  • share modifications of /etc. Files stored in .drop/envs/ENV_ID/etc are a read-only upper layer of overlayfs over the original /etc and take priority over the original content of /etc.

Directories

Path Synopsis
cmd
drop command
internal
cli
env
ipc
Package ipc provides communication between parent and child processes that setup Drop sandbox.
Package ipc provides communication between parent and child processes that setup Drop sandbox.
log
pty
updater
Package updater checks for new versions of Drop.
Package updater checks for new versions of Drop.

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL