By the end of this lesson you will have:
A connected assistant with no saved prompts makes you re-explain every job, every time. This lesson fixes that permanently. Each prompt below is a complete job description: it names the task, the voice, the rules, and the approval gate, so the assistant behaves the same way every run.
Write a prompt once and it works every day after. "Help me with my inbox" gets you a vague answer. The inbox prompt below gets you a reviewed draft queue, because it carries your never list, your price band, and the approval rule inside it. Your standing brief from lesson a-1 (Set Up Your AI Assistant) rides along automatically, so every prompt inherits her persona without you pasting it.
The rules inside these prompts are not optional padding, and the fastest way to lose the account is to edit them out to make the agent "faster." An unsupervised agent selling paid content in her DMs is how operators lose accounts and fans at the same time. Keep the rules in every copy you ever make.
In your AI assistant: save each prompt below as a saved prompt, command, or note, under the bolded name it carries here. Keep the approval line in every fan-facing prompt exactly as written.
When they are saved, the Arsenal is installed where you will actually find it tomorrow evening.
In your AI assistant:
Evening ops run, in order: (1) Go through my unread chats β who's waiting,
what does each fan want, draft a reply to every one in her voice, respecting
my never list and price band. (2) Flag any fan whose lifetime spend makes
them a whale β their drafts get marked for my personal rewrite. (3) List
pay-to-view messages from the last 7 days still unpurchased and draft one
warm follow-up each. (4) Show me everything before anything sends.
In your AI assistant:
Plan next week's content calendar: 3 reels + 1 carousel per day on
Instagram, drawn from my banked content list [paste or attach]. Give me a
day-by-day grid with caption drafts in her voice β captions follow my
language fence: no subscribers/exclusive/spicy/18+/link-in-bio phrases.
In your AI assistant:
Draft three teaser posts for this week's drop, schedule them across the
week on Fanvue, and pin the strongest one to my profile. Show me the
schedule and captions before anything publishes.
In your AI assistant:
Compare my earnings this week with last week. Name my top five spenders,
rank my posts by tips and unlocks, tell me what the winners had in common,
and give me one specific thing to do next week to grow.
In your AI assistant:
Look in my Fanvue Vault and tell me which photos and videos I have never posted
or sent, with a caption idea for each β flag anything that sits above my
content ceiling so I don't ship it by accident.
In your AI assistant:
Using my standing brief, write next month's storyline arc: 4 weekly
mini-themes in her life (places, moods, running jokes), each feeding reel
hooks, carousel ideas, and two tease moments that stay inside the language
fence. Output as a one-page brief I can hand to my production block.
In your AI assistant: run the daily driver plus any four others against her real account.
When five jobs have run to approval quality, and at least one prompt got improved and re-saved, the Arsenal is proven.
In your AI assistant: whenever you notice a job you still do by hand, run this:
Watch what I did by hand this week [describe]. Write me a reusable prompt that
does it to the same standard, with my approval gate built in, and name it.
Save what it writes under the name it gives. The Arsenal should grow a prompt or two every month.
~20 minutes.
The evening inbox run and the weekly plan both come from saved Arsenal prompts, not from scratch. Mark the lesson complete.
Your standing brief is thin. Re-open it in your assistant, add her details from lesson 1-1 (Pick Her Niche & Persona) and lesson 1-2 (Create Your AI Model) again, add three real messages she has sent that you loved, and tell the assistant "match this register." Voice is trainable; train it once.
Re-read your prompt. The approval line was missing or softened. Restore "show me everything before anything sends" verbatim. If the assistant sent despite the line, revoke the connection, reconnect, and report it in #course-help so the team can flag the client.
Quick single-tab jobs, such as scheduling one post or asking one number, go to the built-in Creator Agent from lesson a-2 (Fanvue's Built-In Creator Agent). Compound jobs that need her voice, your rules, or more than one tool go to your connected assistant. When in doubt, use whichever is already open.
Fanvue is agent-run. lesson a-5 (Agents That Run Bopgpt) is the last lesson of the section: pointing an agent at Bopgpt itself, so content production gets the same treatment the inbox just got.