Worktrees

.pitway/config.yaml controls how tasks execute, generated by pitway init with explanatory comments:

execution:
  strategy: parallel_worktrees   # independent tasks dispatch concurrently

sequential vs. parallel_worktrees

sequential runs one task at a time, inline.
parallel_worktrees (the generated default) lets independent, dependency-free, disjoint-write_scope tasks dispatch concurrently, each into its own temporary Git worktree. PitWay validates eligibility up front and integrates each result as a diff-apply, never a merge -- so the resulting mainline history stays indistinguishable from running the same tasks sequentially.

Commands

Command	Purpose
`pitway task-dispatch`	Prepare a parallel-eligible task for worktree execution.
`pitway task-integrate`	Apply a dispatched task's worktree commit to the main tree.
`pitway task-discard`	Abandon a dispatched task's worktree without integrating.

Each worktree worker receives only its own generated task-context bundle (task definition, acceptance criteria, contract excerpt, dependency results, relevant files, verification instructions) -- never full milestone history -- and its results are persisted as concise structured summaries, never full transcripts. This is the same context isolation used for sequential task execution; parallel dispatch doesn't relax it.

No branches, stashes, or merges mid-task

Whichever strategy is set, ordinary task work never creates branches, worktrees (outside the dispatch mechanism itself), stashes, or merges as a side effect -- integration is deliberate and explicit, through the commands above.