Run commands across sandboxes, one array-shaped call [rate class: sandbox_api_rps]

Run shell commands in sandboxes - each inside one of your own forked microVMs, a single-tenant guest with its own kernel that holds only what you baked into it. Array-shaped ON PURPOSE: pass execs: [{sandbox, cmd, deadline_ms?}, ...] and one call runs across many sandboxes concurrently - the shape for rollouts, where one call per sandbox would only burn your rate budget. Results return in request order, each tagged with its sandbox id, carrying exit_code, stdout, stderr, duration_ms, and timed_out. Commands inside one sandbox run one at a time; different sandboxes run in parallel. For a single command, sandbox + cmd is accepted as shorthand for a one-entry array. A result carrying outcome: "void" is an infrastructure failure on OUR side - unbilled, and not a fact about your command: retry it on a fresh fork rather than scoring it as a failure. A result carrying cancelled: true means your own delete_sandbox won a race against this exec - the command's verdict is unknowable, the time is billed, and there is nothing to retry or score. oom_kills > 0 (with a note) means the guest kernel's OOM killer fired during the run: a bare "Killed" with exit 137 is the sandbox's own memory budget biting, billed - bake the env with more ram_gib or reduce memory use, and score it as your command's failure, never retry-as-void. The field absent means the env's baked agent predates measurement, not "no OOM".