competitive-agents

Competitive Agents

Overview

Dispatch two independent subagents to solve the same task from different angles. The main agent then synthesizes the best elements into a superior combined result.

When to Use

• Designing systems, architectures, or APIs
• Problems with multiple valid approaches (no single obvious answer)
• User explicitly requests competitive/parallel approaches
• You want to reduce single-agent bias

When NOT to Use

• Purely mechanical tasks (rename variable, fix typo, add import)
• Tasks requiring sequential steps (not parallelizable)
• Task is unclear — clarify with user first, then decide

Red Flags - Skill Still Applies

Don't skip just because:

• Task seems "simple" (simplicity ≠ obvious solution)
• You think one approach is "best" (that's exactly the bias to avoid)

If multiple valid approaches exist, use this skill.

Workflow

1. Clarify if needed — if the task is ambiguous, ask the user before dispatching
2. Dispatch 2 subagents in parallel (single message, 2 Task tool calls)
3. Wait for both results
4. You (main agent) synthesize the best elements directly — no third subagent
5. Present synthesized result to user

Competitor Prompt Template

Each competitor gets the same task but a different constraint to force divergent solutions:

Competitor A:

You are competing against another agent to solve the same task.
The better solution will be selected. Give your absolute best effort.

YOUR CONSTRAINT: Prioritize simplicity and minimalism. Fewer moving parts wins.

## Task
{task content from user}

## Requirements
- Provide a complete, well-reasoned solution
- Explain your approach and key decisions
- Consider trade-offs and alternatives you rejected

Competitor B:

You are competing against another agent to solve the same task.
The better solution will be selected. Give your absolute best effort.

YOUR CONSTRAINT: Prioritize completeness and extensibility. Cover more cases and future needs.