What problem does it solve? Running multiple AI coding agents in the same directory causes dotnet build failures from shared obj/ folders, file modification races, staged-file collisions in shared commits, and detached-HEAD push failures. This Skill provides verified operational rules for isolating each subagent in its own git worktree and recovering from the failure modes that appear when worktrees are shared. ## Core Features & Use Cases - Worktree isolation workflow: Create per-subagent worktrees with git worktree add, merge results back with fetch/merge, and clean up with git worktree remove. - Race and pitfall recovery: Diagnose and repair staged-file leaks across parallel lanes, detached HEAD states, vanished worktrees, patch-tool writes landing in the main checkout, and red origin/main merges. - Verification gates: Confirm commits with git show --stat HEAD, verify pushes with git rev-parse, and prove pre-existing test failures using throwaway base-commit worktrees. - Use Case: An orchestrator dispatches three parallel implementation subagents into one task worktree with disjoint file sets; each lane commits only its own paths using the pathspec commit form, and the orchestrator reviews at the seams before merging. ## Quick Start Dispatch each code-editing subagent into its own git worktree created with git worktree add, and verify every commit with git show --stat HEAD before merging it back.
