Good results come from clear requests, useful context and explicit boundaries.
Ask for an outcome
Tell the agent what you want done, who the result is for and what format you need. Include deadlines, source material and constraints when they matter.
Useful commands
You can ask for practical control actions in plain language. Common examples include starting a fresh conversation, changing the working model, showing usage, checking status, stopping a task, or listing available skills. The exact commands and capabilities depend on your deployment.
Files and structured work
Ask for a document, table, checklist, HTML file or summary explicitly. If a file is not delivered correctly, ask the agent to resend it as a Telegram attachment or provide the content in chat.
Approvals and safety
Managed agents can be configured to ask before external actions. Review messages, links, emails and changes before approving anything consequential. Keep sensitive instructions and credentials out of ordinary chat messages.
Memory and continuity
An agent may retain approved preferences and useful context between conversations. Ask it to forget or correct information when needed, and avoid sharing information that the workflow does not require.