Directive anatomy

Build-along · ~12 min at your own keyboard

A directive is an SOP written for an AI employee: what we're doing, what comes in, the steps, what to do when it's weird, and how we know it's done. Five sections. Miss the edge cases and you get graceful, silent failure — one lead returned instead of 100, no error thrown.

The long version: Directives: SOPs for an AI employee ↗

Do this

0/4

Checkpoints

▸You should now see a directives/ folder with your file inside.
▸You should now see five labeled sections: objective, inputs, steps, edge cases, done.
▸You should now see a number in the done check, such as a minimum result count.
▸You should now see the fresh session following the directive without your help.

The three ways this goes wrong

The edge cases section is empty or says 'handle errors'. — Write three specific odd inputs and what the flow does with each.
The run finishes and nothing tells you it came back short. — Add a minimum count to the done check and a line that says stop and report if it is not met.
The fresh session asks you questions the directive should answer. — Each question is a missing line. Add the answer to the directive and run again.

Share what happened

Paste your five sections (trimmed) with the edge cases and threshold visible.

Strip API keys, passwords, and tokens before you paste. Swap them for [KEY].

🔒 Don't paste confidential, personal-health, or payment info.