This section describes how to run the whole stack in dev-mode and outlines the differences that are applied during development and compared to how they are run in production.
differences:
-
container runtime
runscis Linux-only container runtime, for developmentruncis used- configured as environment variable
SANDBOX_RUNTIMEin .env.example
-
OpenAI API calls
- 3 agent modes are implemented -
live,mock&simulationlive- calls real OpenAI APIs (requires API token)mock- skips the whole sandbox and worker setup and returns empty finding with approvalsimulation- simulates whole end-to-end flow by using a fake-openai API service and returns based on configuredsimulation scenario-
configured as environment variable
SIMULATION_SCENARIOin .env.example (requiresAGENT_MODE=simulation) -
scenarios selected by
SIMULATION_SCENARIO:Scenario Behavior successTool call, then approved review findingsTool call, then one clearly synthetic finding api-errorHTTP 429 before any completion; usage unavailable error-after-toolHTTP 429 after the tool response; retains partial usage/cost invalid-outputInvalid final JSON; retains usage/cost from both responses missing-usageFinal response omits usage; partial cost estimate step-limitRepeated tool calls until MAX_STEPSis exhausted
-
- 3 agent modes are implemented -
-
Sign Up (email + password)
- in production only OAuth sign ups are supported (with github & google providers)
- email + password is exposed during development for "low effort setup" purposes
- enforced through environment variable
NODE_ENV- when set to "development"