CodeRight treats "done" as a claim that needs evidence. The runtime does not stop just because a model produced a confident final message. It checks the work against the goal, review output, and verification signals.
1. You give CodeRight a goal
The unit of work is a goal, not a single chat response. A good goal includes what should change and what evidence should exist when the run is complete.
2. The Conductor splits the work
The Conductor turns the goal into a task graph. That lets CodeRight route planning, coding, review, and remediation through different roles instead of flattening the whole job into one prompt.
3. Specialist workers execute
A worker role can implement a change, inspect the repository, run commands, or prepare evidence. The model is a replaceable worker. The runtime owns the loop, permissions, logs, and state.
4. The jury reviews the artifact
The review layer is independent from the worker. It evaluates the actual artifact: diff, files, build output, tests, and stated goal. It is not enough for the worker to say that tests passed.
5. The Completion Arbiter gates "done"
The Arbiter decides whether the goal is complete. It can reject a finished-looking run when the verification evidence is missing, when the jury disagrees, or when the implementation solves only part of the request.
6. Failed verification triggers remediation
A rejected run is not a dead end. CodeRight can route the failure back into the task graph: add missing tests, repair the implementation, rerun checks, and ask the jury again.
7. The result includes an evidence trail
The output is not just a final answer. It includes the run history, tool output, review state, and verification record that explain why CodeRight considered the goal complete.
Core distinction: Cursor and Copilot help you write code inside an editor. CodeRabbit reviews pull requests in the cloud. CodeRight is the local runtime that supervises agent work and verifies completion.
Next: review the feature breakdown, check the security model, or compare the one-time Pro license against recurring code review subscriptions.