SandboxedCommand

Run commands and agent-written programs inside an operating-system boundary. See the guide.

class zett_agent.sandbox.base.SandboxedCommand(argv, env, cwd, limits=ResourceLimits(memory_bytes=None, cpu_seconds=None, cpu_percent=None, processes=None, file_size_bytes=None, open_files=None, core_dumps=False, memory_sample_seconds=0.25), preexec=None, warnings=())

A command rewritten to run inside a sandbox, ready for a process API.

Sandbox.wrap() returns one without running anything, so a host that owns its process handling (streaming output, a pty, its own timeout) can spawn argv itself with env and cwd. Start it in its own process group (start_new_session=True on POSIX) so the whole tree can be stopped.

Parameters:

• argv (tuple[str, ...])

• env (Mapping[str, str])

• cwd (Path)

• limits (ResourceLimits)

• preexec (Callable[[], None] | None)

• warnings (tuple[str, ...])

argv

The program and its arguments, including the sandbox launcher.

Type:

tuple[str, …]

env

The complete environment to start it with.

Type:

collections.abc.Mapping[str, str]

cwd

The working directory to start it in.

Type:

pathlib.Path

limits

The resource limits the policy asked for. A host that spawns the command itself applies the kernel limits with preexec; the ones only Sandbox.run() can enforce are named in warnings.

Type:

zett_agent.sandbox.policy.ResourceLimits

preexec

Apply the kernel resource limits in the child before exec, for subprocess’ preexec_fn. None when there is nothing to apply.

Type:

collections.abc.Callable[[], None] | None

warnings

Parts of the policy or the limits this backend could not enforce for this command. A caller that must not run with a weaker boundary checks it.

Type:

tuple[str, …]