Skip to main content

Best practices

Poolside Assistant works best with well-defined tasks. For example, instead of improve the app, try add error handling to all API calls in src/api/.
When dealing with complex tasks, agents are more effective when they plan first. Ask Poolside to help you create a plan, for example, Write a product requirements document in Markdown that defines the tests to add to the API. Only ask Poolside to carry out the plan after you define it well.
Help the agent understand your project by creating an AGENTS.md file with coding standards, architecture decisions, and conventions. For more information, see Agent instructions.
Direct Poolside Assistant to the right starting point by using the @ notation to reference key files or folders. For more information, see Context.
When using the agent for the first time in a project, review each approval request to understand what tools it needs.
When you’re comfortable with certain tools, enable auto-approval to speed up the agent’s workflow while maintaining control.
If the agent is heading in the wrong direction, press Escape to stop it from running and then provide clarifying instructions.