tournament

Generates divergent solution candidates in parallel and judges them through a bracket to synthesize a winner.

1|Updated Jun 1, 2026
One-click install
npx skills add https://github.com/kridaydave/My_Skills --skill tournament-kridaydave
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: tournament
Source: https://github.com/kridaydave/My_Skills/tree/main/tournament
Command: npx skills add https://github.com/kridaydave/My_Skills --skill tournament-kridaydave

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Single-attempt-then-iterate workflows lock in early on mediocre ideas when the solution space is wide. This Skill explores many genuinely different approaches to one open problem in parallel, then uses structured pairwise judging to find and synthesize the best answer. ## Core Features & Use Cases - Divergent Parallel Generation: Spawns multiple blind subagents, each seeded with a distinct strategy or constraint, producing structurally identical candidates for fair comparison. - Judged Bracket Elimination: Runs head-to-head pairwise judging rounds against a fixed weighted rubric, with byes for odd fields and optional semifinal cross-checks. - Champion-plus-Grafts Synthesis: Delivers the winning candidate enhanced with the strongest single ideas from runners-up, plus a visible bracket path for auditability. - Use Case: When choosing between API schema designs, architecture options, naming candidates, or algorithm approaches, run a tournament of 4-8 seeded candidates instead of polishing the first idea that comes to mind. ## Quick Start Run a tournament on this design problem: generate four competing approaches with different strategy seeds, judge them pairwise against a weighted rubric, and give me the synthesized winner.

Frequently Asked Questions about tournament

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I generate multiple competing solutions to one design problem?▼

Spawn several blind generator subagents in a single parallel batch, each seeded with a distinct strategy such as simplest, scale-first, user-first, or unconventional. Each returns a structured candidate with its approach, solution, tradeoffs, and single best idea for later comparison.

How does bracket-style judging compare design candidates?▼

Candidates are paired head-to-head each round, with one judge subagent per pair scoring both against a fixed weighted rubric. Winners advance to the next round until one champion remains, and odd fields receive a bye rather than a fabricated opponent.

When should I use a tournament instead of iterating on one answer?▼

Use it when the solution space is wide and the first plausible answer is unlikely to be the best, such as architecture choices, API shapes, naming, or algorithm selection. Below three candidates, answering directly is more cost-effective.

Why must the judging rubric be defined before generating candidates?▼

Defining weighted criteria after seeing candidates rigs the outcome toward whichever option you already prefer. Weights must sum to exactly 100 percent so pairwise comparisons stay consistent across all judge pairs.

What are the limitations of parallel candidate generation?▼

Each generator is a full subagent, so cost scales with field size; four to eight candidates is the practical range. If seeds are not forced apart, generators produce near-identical candidates and the bracket becomes meaningless.