What problem does it solve? Trying multiple solution approaches to a coding or content task sequentially is slow, and comparing alternatives manually is error-prone. This Skill runs N parallel agents on the same task in isolated git worktrees, evaluates their results by metric or LLM judge, and merges the best branch automatically. ## Core Features & Use Cases - Parallel agent competition: Spawn N subagents in isolated git worktrees, each working independently on the same task with no cross-communication. - Flexible evaluation: Rank results by numeric metrics (latency, coverage, size), LLM-judged quality, or a hybrid of both, then merge the winner with git merge --no-ff and archive losers as tags. - Predefined agent templates: Built-in templates for optimizer, refactorer, test-writer, and bug-fixer patterns, each with a structured iteration loop and strategy assignment. - Use Case: You want to reduce API response time below 100ms. Initialize a session with 3 agents, each assigned a different strategy (caching, algorithm optimization, I/O batching), let them iterate in parallel worktrees, then rank by p50_ms and merge the winner. ## Quick Start Ask the assistant to initialize an AgentHub session with your task description and agent count, then spawn the agents and evaluate their results.
