Station 3 · never used an agent · paper
The Dry Run
Published to read
Runs your mandate as a paper session against live market data. Real prices, real funding, real decision loop — zero money, zero keys. The agent narrates every decision, including the decisions to do nothing.
Trust in an agent is built by watching it decline bad trades, not by watching it win. Inaction is the hardest behavior to trust and the most important one. This is where you learn to read it.
The prompt
Run a paper session under my mandate (attached below), using gryps_paper_session as the session engine. 1. Each cycle: read live market state with gryps_venue_snapshot for my mandate's markets — price, funding, fee tier — and say when it was read. 2. For anything that looks like an opportunity, run gryps_edge_check before deciding: what would this trade need to earn to clear the venue's measured friction? State the number. 3. Then your decision, with the mandate rule it rests on. When you decide NOT to act (this should be most of the time), say exactly why. "No qualifying edge — holding" is a complete, correct answer. 4. If anything would have violated the mandate, flag it loudly. 5. End with the session report: decisions made, trades declined and why, and what acting on everything you saw would have cost in fees. [PASTE YOUR MANDATE HERE]
Published to read
This prompt is written for the Gryps public server, which is not on npm yet. The package you can install today, gryps-agent-mcp, does not have gryps_paper_session, gryps_venue_snapshot. The station is published so you can read what it will ask of you; it opens the day the public server ships.
What good looks like
A running commentary that is mostly disciplined refusal, with reasons. The final report's most interesting number: what acting on everything would have cost in friction. That number is why your agent's restraint has value.