executing-plans

Executing plans with review checkpoints involves breaking implementation into controlled batches with mandatory code reviews after each batch. This skill executes tasks sequentially or in parallel—depending on whether a manifest.json exists—and stops for review gates between batches to catch blockers early and prevent rework.

Parallel execution is automatically enabled when your plan includes a manifest.json file decomposing tasks into parallel-executable units. Plans without a manifest execute sequentially. This parallelization detection lets you accelerate completion on complex multi-task plans without manual coordination overhead.

Batch execution with per-task status tracking provides visibility into progress. Each task is logged and verified individually, with code review gates inserted after every batch. This structure prevents silent failures and ensures blockers surface immediately rather than compounding through later stages.

Critical review gates enforce stopping to ask for help when blockers are encountered, eliminating guessing that causes rework. Combining batch execution, per-task verification, and mandatory code reviews creates checkpoints that catch issues before they propagate downstream.

A manifest.json is optional. Plans with 4+ tasks benefit from manifest-driven parallel execution for speed, but simple 1–3 task plans execute successfully without one, running sequentially instead. The skill adapts its execution strategy to your plan's structure.

Plan execution concludes with a completion announcement and requires finishing-a-development-branch as the final step, ensuring your implementation branch is properly closed out and integrated into your workflow.