Streamlit prototype subprocesses are spawned with the following safeguards:

• Memory limit: RLIMIT_AS capped at MAX_MEMORY_BYTES (default 512 MB) via resource.setrlimit.
• Session isolation: Each subprocess runs in its own process group (os.setsid()), enabling clean group-level termination.
• Lifetime cap: Runtimes exceeding MAX_RUNTIME_SECONDS (default 3600) are auto-killed on next status check.
• Concurrency cap: At most MAX_CONCURRENT_RUNTIMES (default 5) subprocesses may run simultaneously.

All limits are configurable via environment variables.

• Every view that accesses a PrototypeProject filters by created_by=request.user.
• Users who do not own a project receive a 404 response (not 403), preventing project slug enumeration.
• The dashboard only shows projects owned by the authenticated user.
• Artifact downloads require that project.created_by matches the requesting user.

When DJANGO_SETTINGS_MODULE=config.settings.production:

• SECRET_KEY is required (startup fails if unset).
• DEBUG is always False.
• HSTS, secure cookies, CSRF cookie security, and SSL redirect are all enabled by default.
• SECURE_PROXY_SSL_HEADER is set for reverse proxy deployments.
• Static files are served via WhiteNoise with compressed manifest storage.

• Streamlit subprocesses run on the same host as the Django app. For stronger isolation, run prototypes in separate containers.
• RLIMIT_AS limits virtual memory, not resident memory. For hard RSS limits, use cgroups or container-level constraints.
• Port allocation uses double-bind verification but is not fully atomic; collisions are possible under extreme concurrency.
• No rate limiting is applied to prototype generation or runtime starts. Consider adding middleware or reverse proxy rate limits for public deployments.