Skip to content

About

AI code-review agent combining deterministic tools with LLM reasoning, structured outputs, persistance, human in-the-loop workflow, Github integration, evals and observability

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

Development

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

    • runsc is Linux-only container runtime, for development runc is used
    • configured as environment variable SANDBOX_RUNTIME in .env.example
  • OpenAI API calls

    • 3 agent modes are implemented - live, mock & simulation
      • live - calls real OpenAI APIs (requires API token)
      • mock - skips the whole sandbox and worker setup and returns empty finding with approval
      • simulation - simulates whole end-to-end flow by using a fake-openai API service and returns based on configured simulation scenario
        • configured as environment variable SIMULATION_SCENARIO in .env.example (requires AGENT_MODE=simulation)

        • scenarios selected by SIMULATION_SCENARIO:

          Scenario Behavior
          success Tool call, then approved review
          findings Tool call, then one clearly synthetic finding
          api-error HTTP 429 before any completion; usage unavailable
          error-after-tool HTTP 429 after the tool response; retains partial usage/cost
          invalid-output Invalid final JSON; retains usage/cost from both responses
          missing-usage Final response omits usage; partial cost estimate
          step-limit Repeated tool calls until MAX_STEPS is exhausted
  • 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"

About

AI code-review agent combining deterministic tools with LLM reasoning, structured outputs, persistance, human in-the-loop workflow, Github integration, evals and observability

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages