7 Prompts From a Live Agent Stack

What's inside

What each one does:

• 1 · Agent Identity Brief — the template that defines every agent in the stack. Role, exact output format, and one hard constraint. This is the difference between an agent that behaves the same way on job 400 as it did on job 1, and one that drifts.
• 2 · Cron Task Specification — how to describe a scheduled job so it runs unattended without you babysitting it.
• 3 · Build-in-Public Content — turning real system events into posts, without inventing progress you did not make.
• 4 · Lead Scoring (Local Model) — written specifically for a small local model, so it costs nothing to run at volume.
• 5 · Self-Heal Diagnostic — what you hand a model when something broke and you need the cause, not sympathy.
• 6 · Free Model Routing — deciding which model gets which job when your budget for models is zero.
• 7 · Daily Status Report — the one that stops a system from reporting that everything is fine while it quietly is not.

How to run them: replace the {bracketed} fields, then run against local Ollama or any free cloud tier. Nothing here needs a paid API key.

Honest about the size: seven prompts, roughly 1,600 words. If you want volume, this is the wrong purchase. What you are buying is that each one has already survived contact with a system that runs unattended — a vague version of prompt 1 produced a vague agent, and the blunt version in here did not.

7 prompts · markdown · yours forever, updates included.