sadd:do-competitively

Competitive multi-agent generation improves task synthesis by running parallel generators to explore diverse approaches, then using a meta-judge and multiple judges to systematically evaluate and select the highest-quality solution.

Use multi-stage evaluation for complex decision-making when output quality matters more than speed, such as for design decisions, algorithm selection, and high-stakes synthesis tasks requiring robust verification.

Orchestrate parallel agents by providing a task description and target output path; the system runs four parallel agents—one meta-judge and three generators—to produce, evaluate, and synthesize competing solutions.

Yes, the meta-judge evaluation supports adaptive strategy selection through SELECT_AND_POLISH, REDESIGN, and FULL_SYNTHESIS strategies to accelerate high-quality outputs based on evaluation results.

Yes, you can use multi-agent planning for algorithm selection tasks. It is explicitly designed for complex, high-stakes scenarios where multiple competing implementations must be systematically evaluated to ensure robust outputs.