Use an agent mailbox for one task

TASK SETTINGS — fill only what is relevant
Scope: [selected accounts, records, or task]
Budget: [amount and currency, or no new spending]
Deadline: [date, time, and timezone]
Tone: [polite and concise]
Contact channel: [no contact until approved]
Approval: Read-only first. Ask before sending, calling, submitting, cancelling, booking, paying, signing, or changing anything.

Keep one approved provider conversation organized in an agent-managed mailbox.

Use only the accounts and materials I authorize. Start with: the task, provider, permitted information, account ownership and an existing agent mailbox if enabled.

First verify that the mailbox exists and that I control the associated account. Define whether replies should be forwarded or summarized and how I can recover access without the agent. Draft the initial message and a follow-up boundary. Retain a correspondence log and distinguish automated notices from decisions requiring me.

Work read-only and prepare drafts first. Before contacting anyone, sharing information, changing an account, booking, spending, submitting a form or modifying files, show me the exact action and obtain my approval. Do not create accounts, accept terms, supply identity documents or use the mailbox to hide who the customer is.

Verify that your tools actually support the requested work. After two different failed access attempts, report the blocker and finish the accessible parts instead of looping. Never invent an action, source, availability or confirmation.

Return: the mailbox capability check, proposed correspondence rules, initial draft and a clear access-recovery record. Separate verified facts, assumptions, proposals, and completed actions. This first pass is finished when the requested evidence and next decision are delivered, not when an unapproved action is taken.