Code And Process Execution

Agent runs a build command inside a sandbox via the SDK's exec/code-run method and reads only stdout to decide success.

Branch on the process exit code, not on stdout presence: a zero exit is success, non-zero is failure (read stderr for the reason). Capture stdout, stderr, and exit code as three distinct fields. Do not infer success from non-empty stdout. [REQUIRES-VERIFICATION] for the exact result object field na…

AI-generated code contains an accidental infinite loop. The agent runs it in a sandbox with no timeout.

Set an explicit wall-clock timeout on every exec/code-run of untrusted generated code; on timeout, kill the process and reclaim the sandbox. Never run generated code unbounded — an infinite loop otherwise burns compute until quota/billing limits trip. [REQUIRES-VERIFICATION] for the SDK's timeout p…

A training/build step inside the sandbox runs for many minutes and emits incremental logs. The agent blocks on a single synchronous exec and shows nothing until it ends.

For long-running processes, use the SDK's streaming/log-follow API (or a background process handle) to surface incremental output and detect hangs, rather than blocking on one synchronous call. Distinguish 'still running' from 'stuck' via progress in the stream. [REQUIRES-VERIFICATION] for streamin…