What problem does it solve? A single attempt at a non-trivial artifact often locks in the wrong design shape early. This Skill runs multiple independent candidates at the same task, scores them against an explicit rubric, and synthesizes the strongest result instead of betting on one try. ## Core Features & Use Cases - Parallel fan-out: Spawns N subagent candidates on isolated output paths, each producing an artifact plus a rationale of rejected alternatives. - Rubric-based cross-judging: A separate judge model scores every candidate criterion by criterion, and the parent reads all candidates end to end before picking a base. - Graft and verify: Ports one or two winning ideas from losing candidates into the base by hand, then verifies the synthesized artifact and records a synthesis note. - Use Case: When designing a new API or architecture, run four models in parallel, compare their proposals against 3-6 gradeable criteria, and ship a merged design with documented trade-offs. ## Quick Start Ask the agent to arena this design task with four candidates and a rubric covering correctness, simplicity, and extensibility.
