agent-modes_skill

Overview

This skill provides clear guidance on choosing and using the four agent modes: Agent, Ask, Plan, and Debug. It helps you match a mode to the task, explains mode-specific workflows, and shows when to switch modes for better outcomes. Use it to reduce friction when moving between exploration, planning, implementation, and bug fixing.

How this skill works

The skill describes the purpose and workflows for each mode and gives concrete signals that indicate which mode fits a task. It outlines step-by-step behaviors you can expect—how an agent autonomously explores, how Ask mode performs read-only analysis, how Plan mode creates multi-step strategies, and how Debug mode instruments and validates fixes. It also summarizes quick switching and shortcut options for fluid workflows.

When to use it

• Selecting the right mode for a development task
• Explaining mode capabilities to team members or collaborators
• Preparing to implement a complex feature that needs planning
• Diagnosing reproducible bugs or performance regressions
• Exploring unfamiliar code or system architecture

Best practices

• Start in Ask mode for unfamiliar code to gather context before changing files
• Use Plan mode when requirements are unclear or multiple approaches exist and iterate on the plan with reviewers
• Switch to Agent mode for autonomous, multi-file implementations once the plan is approved
• Use Debug mode for reproducible issues: add instrumentation, reproduce, analyze logs, and apply targeted fixes
• Use keyboard shortcuts or the mode picker to switch quickly and avoid mode-mismatch work

Example use cases

• Add a new API endpoint spanning backend and client components — Plan, review, then Agent to implement
• Understand how authentication flows work in a legacy codebase — Ask for read-only exploration
• Refactor the data layer across many files with coordinated changes — Plan to design then Agent to execute
• Investigate intermittent audio streaming drops — Debug to instrument, reproduce, and fix
• Create a user preferences system with architectural trade-offs — Plan to outline options and gather feedback