agenthub

To run multiple AI agents in parallel, this skill spawns autonomous agents using isolated git worktrees, allowing each agent to explore different strategies on the same task without interfering with each other. A central coordinator then evaluates and selects the best outcome.

You can evaluate results from a multi-agent workflow using a metric, an LLM judge, or a hybrid approach. After evaluation, the central coordinator identifies the winning agent and safely merges the best outcome back into the main branch.

The best way to explore diverse solutions for code optimization is parallel exploration using multiple agents. By spawning autonomous agents in isolated worktrees, you can compare different optimization strategies concurrently and select the most effective one.

Yes, this multi-agent coordination approach requires a git repository to utilize git worktrees. Worktrees provide the necessary isolation for each agent to work independently and safely before the winning result is merged.

You should not use parallel agents for tasks requiring sequential execution or when isolated worktrees cannot be created. This approach is designed for parallel exploration, evaluation, and content variation rather than dependent, linear workflows.