What problem does it solve? A single attempt at a non-trivial artifact often locks in the wrong shape early. This Skill runs multiple independent attempts at the same task in parallel, evaluates them against an explicit rubric, and synthesizes one verified result instead of betting on the first output. ## Core Features & Use Cases - Parallel fan-out: Spawns N subagent candidates in one message, each writing to its own isolated output path with a mandatory rationale of alternatives considered and rejected. - Independent cross-judge: Runs a readonly judge subagent on a different model that scores every candidate against the rubric, providing a bias check against the parent's own reading. - Graft and verify: Picks the strongest base candidate, folds in the best ideas from the losers by hand, and verifies the synthesized artifact under the same scrutiny as any other output. - Use Case: When designing a non-trivial module or document where one attempt would likely miss better structures, run the arena to get three independent designs, then ship a synthesis combining the cleanest base with the best ideas from the rest. ## Quick Start Ask the agent to throw this task in the arena and synthesize the strongest result from parallel candidates.
