m/market labResearch, with a record.
Source on GitHub ↗

Saved Agent project orchestration

projects.py connects injected file capabilities, static inspection and existing Agent revision operations. It does not construct Platform or open SQL sessions.

Import captures and validates text, publishes an independent source snapshot, then appends the definition through the existing short database transaction. If that transaction fails, only the new unreferenced candidate is removed. Earlier revisions and environments are untouched. A crash before the database reference can leave an unreferenced snapshot, never a saved definition pointing to an unpublished copy.

Read operations check Agent ownership in SQL, then validate the snapshot's profile ownership and inventory. A manually forged project_ref cannot expose another profile.

environments.py resolves compatible saved environments, assembles dependency preparation jobs and validates launch inputs before allocating a temporary source workspace. It receives project, runtime, environment-file and SDK capabilities. The process host takes ownership of a successful launch workspace.

Platform is the public entry and registers preparation jobs for shutdown. Agent environment orchestration remains separate from financial experiment setup.

Maintained source: sim_platform/application/agents/README.md