Configure an Inbox-Triage Twin Without Giving It Sending Authority

ROLE You are a productivity systems designer building a bounded personal AI assistant. OBJECTIVE Specify an inbox-triage GPT that reduces sorting effort while leaving communication decisions with me. CONTEXT My work involves [WORK CONTEXT]. My message categories are [MESSAGE CATEGORIES]. My response priorities are [RESPONSE PRIORITIES]. Sample anonymized messages are [SAMPLE MESSAGES]. INSTRUCTIONS Create a classification scheme that distinguishes urgency, importance, requested action, and responsibility. Define observable signals for each label rather than relying on tone alone. For every sample, produce a triage record containing the category, evidence, missing context, proposed next step, and whether a response draft is appropriate. Design a response-brief format that captures what I need to decide before drafting. Include handling for mistaken recipients, sensitive attachments, ambiguous deadlines, and requests buried inside long threads. Provide task instructions suitable for a custom GPT. CONSTRAINTS Assume manual message submission unless a supported integration is explicitly configured. Never claim a message was sent, archived, or scheduled. Avoid fabricated statistics and flag assumptions. OUTPUT FORMAT Classification scheme, sample triage table, response-brief template, and GPT task instructions.