Station 2 · never used an agent · read-only
The Intake Interview
Runs today
Flips the table — the AI interviews *you*, one question at a time, and writes your agent's operating mandate from your answers. You are not writing code. You are making decisions.
An agent without a mandate you authored is a black box you happen to be funding. The mandate is what makes it *yours* — and it's the document every later station (paper, funding, live) runs against.
The prompt
Interview me to build my agent's operating mandate. One question at a time, wait for my answer, then ask the next. Cover: 1. Which markets I want it watching (and which it must never touch). 2. The most money it could ever have at risk at once — and the most it may lose before it must stop and wait for me. (My hard stop.) 3. What it should do when it has no good reason to act. (Hint from the briefing: the honest default is nothing.) 4. When it must wake me up versus act alone. Start me conservative: alone it may only REDUCE risk, never open it. 5. How long its authority lasts before it expires and must be renewed. Then write my mandate back to me: a numbered list of plain-language rules, a one-paragraph summary, and — separately — the rules that are protecting me from myself, with a sentence on why each exists.
Runs as written with gryps-agent-mcp, installed in one line. Read-only.
What good looks like
A mandate you could read to a friend. Every rule traceable to an answer you gave. The self-protection section should feel slightly annoying — that's it working. The default-action rule should say some version of: *no qualifying edge, no trade.*