Worktree

Git worktree management, branch naming, parallel runs.

Overview

The worktree module (worktree.py) isolates each marker’s experiments in a separate git worktree, preventing interference between parallel runs.

Lifecycle

1. Create — create_worktree() sets up a new worktree with a dedicated branch
2. Execute — Engine runs experiments within the worktree directory
3. Commit/Reset — Changes are committed or discarded in the worktree
4. Remove — remove_worktree() cleans up after completion

Branch Naming

Branches follow the results.branch_prefix from the marker config (e.g., autoresearch/auth-flow).

Parallel Execution

Each marker gets its own worktree and branch. Multiple markers can run concurrently without git conflicts since worktrees provide filesystem-level isolation.

Error Handling

GitError exceptions are raised for worktree operations that fail (e.g., branch already exists, dirty working directory).