autodev-parallel

To run parallel AutoDev experiments, this orchestrator coordinates multiple git worktrees and agents concurrently. It automatically generates ideas, collects outcomes, scores against a baseline, and selects the best-performing branch while keeping the main branch stable.

Evaluating multiple coding experiments safely requires isolated git worktrees. This orchestration approach enforces explicit isolation per worktree, runs configurable rounds, and applies a final best-branch cherry-pick workflow to preserve main branch stability throughout the comparison.

Git worktree parallelism works by assigning each agent a separate working directory linked to the same repository. The orchestrator enforces configurable parallelism limits across these worktrees, coordinating rounds and automatic fallback handling to manage concurrent execution.

Yes, you can configure both the number of rounds and the level of parallelism for AutoDev orchestration. These settings control how many concurrent worktrees execute experiments and how many iteration cycles run before collecting results.

Best results are selected through baseline scoring and results collection. After all parallel experiments complete, the orchestrator compares outcomes across branches and applies a final best-branch cherry-pick workflow to merge the highest-scoring changes.

You should avoid parallel worktree orchestration for simple, linear feature development that lacks multiple viable implementation paths. The overhead of managing concurrent agents, baseline scoring, and isolated worktrees is designed for exploring and comparing several distinct ideas concurrently.