Select best agent approach via tournament
tournamentworkflowsetup L2★36
xz1220/open-dynamic-workflows ↗What it does
Run N agents on one task with different strategies, judge pairwise until winner
Best for
Strategy selection when solution space is open-ended and head-to-head beats rubric scoring
Inputs
- · [object Object]
- · [object Object]
Outputs
- · [object Object]
Requires
- · Claude Agent API
Preconditions
Competitors run first (barrier), then bracket pairs/judges run per round (barrier per round)
Failure modes
Odd competitor gets bye advantage; judge picks A over B inconsistently; solution quality varies wildly
Trust signals
- · Single-elimination bracket structure
- · Pairwise head-to-head judging
- · Clear winner emerges