Preparation lifecycle¶
Shared infrastructure for interpreter and dependency preparation, outside the live
experiment pipeline. job.py owns the background worker, progress snapshots,
cancellation and close. process.py runs bounded, cancellable trusted subprocesses
with a reduced environment. POSIX commands use their own process group; Windows commands
are attached to the existing Job Object owner and release that job during cleanup. errors.py defines named preparation failures.
Runtime selection belongs in agents/runtimes; package resolution belongs in
agents/environments; filesystem locking and publication belong in persistence.
Maintained source: sim_platform/agents/preparation/README.md