/withScout

This command handles complex tasks by breaking them down into an investigation phase and an execution phase. It uses investigator agents to gather information before deciding on a plan of action.

When to use

• Use when: The user has a complex request that requires understanding the codebase before changes can be made.
• Suggest when: A user's request cannot be fulfilled without first gathering information from multiple files or parts of the codebase.
• Example: "User: Add a JWT token refresh feature."
• Example: "User: Figure out our project's auth logic and then add a new endpoint."

Actions

This command follows an Investigate -> Synthesize -> Iterate/Execute workflow.

1. Step 1: Deconstruct & Plan

   • Break down the user's primary goal into a set of clear, independently investigable questions.
   • Assign each set of questions to a different investigator agent (e.g., Frontend Investigator, Backend Investigator).

2. Step 2: Parallel Investigation

   • Use the Task tool to launch multiple investigator agents concurrently.
   • Each investigator will research its assigned questions and return a direct markdown report.

3. Step 3: Synthesize & Evaluate

   • Combine the reports from all investigators to form a holistic view of the system.
   • Identify key connections, knowledge gaps, or conflicts in the information.

4. Step 4: Iterate or Execute

   • If information is insufficient (Iterate): Formulate a new, more specific round of research questions and go back to Step 2.
   • If information is sufficient (Execute): Proceed to the Action Phase (Step 5).

5. Step 5: Action Phase

   • Based on the comprehensive information gathered, create a plan and use worker agents to execute the user's final request (e.g., implement the feature, fix the bug).

6. Step 6: Summarize & Report

   • When delivering the final result, explain the investigation process, the key findings, and the actions taken to achieve the outcome.