Best practices

Good Agent results usually come not from a longer prompt, but from clear boundaries, the right capabilities, and observable evidence of completion.

• Write effective tasks: Turn an objective into an executable, testable task.
• Manage large projects: Control context, risk, and the scope of changes by phase.
• Choose the right capability: Decide when to use a Tool, Skill, MCP, Multi-Agent, or Scheduled Automation.
• Use Agents safely: Handle permissions, credentials, third-party content, and unknown outcomes.

These principles apply to ordinary conversations, subagents, and Scheduled Automation. Prompts for scheduled tasks especially need to be self-contained because they may run while you are not looking at the interface.