Overview

Orchestrates 5-10 parallel research agents for comprehensive, multi-source research. Achieves up to 90% faster results compared to sequential research through concurrent execution.

How This Skill Works

Plan the query by decomposing it into independent sub-queries, then Dispatch 5-10 agents in parallel. Synthesize findings to identify consensus and conflicts, then Validate results by cross-referencing against the codebase for accuracy. Overall confidence is a weighted average of agent scores, adjusted by validation, with below 70% triggering human review.

When to Use It

• Use the /research [query] slash command
• Before specification creation
• When investigating unfamiliar parts of the codebase
• When you need multi-source evidence from code, docs, and configurations
• When faster results are needed compared to sequential research

Quick Start

1. Step 1: Plan - decompose the query into independent sub-queries
2. Step 2: Dispatch & Synthesize - run 5-10 agents in parallel and synthesize findings
3. Step 3: Validate & Review - cross-reference against the codebase and assess confidence; trigger human review if needed

Best Practices

• Plan and decompose the query into independent sub-queries before dispatch
• Choose depth (Shallow/Medium/Deep) and agent count (5/7/10) based on task complexity
• Always source from codebase, documentation, and configuration artifacts
• Synthesize results to identify consensus and conflicts across sources
• Validate findings with the codebase and trigger human review if overall confidence < 70%

Example Use Cases

• Quick fact-finding on a simple function using Shallow depth with 5 agents
• Standard research on a module using Medium depth with 7 agents
• Deep, comprehensive analysis of a complex feature with 10 agents
• Cross-referencing code and README/docs to confirm usage patterns
• Investigating unfamiliar areas of the codebase during feature planning

Frequently Asked Questions