plan-implement

To execute a multi-step implementation plan with code reviews, you can use an orchestration tool that parses numbered or prose plans and runs a dedicated implementer/reviewer loop for each step. It applies incremental changes and only commits after receiving an approval verdict.

Yes, you can run specific steps from a plan file by specifying the desired step numbers during execution. The system supports flexible plan sourcing from a file path or conversation context and allows you to execute selected steps or all pending tasks sequentially.

Incremental development with an implementer and reviewer loop works by dispatching independent subagents to write and verify code changes for each task. The reviewer performs diff-verification, and if changes are rejected, the system uses bounded retry rounds before allowing the main session to proceed.

No, the git workflow does not automatically commit code changes after each task. It uses selective git staging and commits only after the reviewer subagent provides an APPROVED verdict, ensuring that unchecked or failed implementations are kept out of the repository.

The best way to prevent scope creep during a complex code refactor is to enforce task-based progress tracking and sequential step execution. By isolating each implementation phase behind a review gate, the workflow strictly bounds changes to the original drafted plan.

You should not use an automated plan execution workflow if your project lacks a clearly drafted implementation plan or if your coding task requires exploratory, unstructured changes. This approach relies on normalized step extraction and sequential execution, making it unsuitable for undefined tasks.