executing-parallel-phase

Parallel execution uses isolated git worktrees to run independent tasks concurrently. Each task gets its own worktree, preventing sequential bottlenecks and enabling true parallel development. Dependencies are installed separately in each worktree, and results are stacked linearly after completion.

Create separate git worktrees for each concurrent task, dispatch subagents to handle them in parallel, then stack the branches linearly. This approach maintains architectural consistency while isolating file changes and avoiding merge conflicts between parallel work.

Yes. Parallel execution handles non-overlapping files across tasks. Each worktree installs its own dependencies independently, so tasks can share the same packages without interference while operating on separate file sets.

The Skill detects and skips already-completed tasks, enabling efficient recovery. Resume the parallel phase to spawn only the remaining tasks while preserving results from successful worktrees.

Single-task phases still use worktrees to maintain architectural consistency. The Skill handles N=1 scenarios the same way as N≥2, ensuring uniform behavior across all parallel executions regardless of task count.

The Skill handles disk-space constraints by managing multiple worktrees efficiently. It performs precondition checks before execution and cleans up worktrees after stacking branches, minimizing resource overhead during concurrent operations.