safely grants write access only to user-specified directories while running an arbitrary command, preventing modification of files anywhere else. If you download the script sketchy.sh from a seedy website and want to run it while keeping it honest about only modifying the current directory like its documentation promises, you can run:

sketchy="$(realpath sketchy.sh)"
mkdir /tmp/throwaway
cd /tmp/throwaway
safely --writable . bash "$sketchy"

If sketchy.sh tries to modify any file outside of /tmp/throwaway, it will fail without doing any harm.

Multiple -w/--writable directories can be specified:

safely --writable /my/dir -w /touch/here -- touch /my/dir/my-file /touch/here/done

Write permission applies recursively to subdirectories. The following is thus equivalent to not using safely at all:

safely -w / command arg1 arg2

You may also be interested in pledge, which is more mature and has significantly broader scope.

apptainer must first be installed and able to bind-mount directories in containers. You can test this with:

mkdir it_works
apptainer exec --bind "$(realpath it_works)" docker://alpine ls -d it_works && echo SUCCESS || echo FAILURE

You can install safely with make install:

make install                  # install at /usr/bin/safely
make install PREFIX=/software # install at /software/bin/safely

If you have bats installed, you can run tests with make check. If you want to do testing with more than your home diretory and /tmp, you can specify extra directories with TESTDIRS:

make check                    # vanilla tests
make check TESTDIRS=/a/b,/cde # also make sure safely protects /a/b and /cde

safely was written to allow me to grade student assignments with some peace of mind, not to stand up to hostile commands. Since safely is just a fancy wrapper for apptainer, it's at most as secure as apptainer is--if a command can escape apptainer, it can escape safely. For the time being, don't assume that safely will stop a motivated attacker.

Pull requests that make safely more robust are welcome.