Research, alert, and journal loop
Three roles — brief, alert, journal — with a human gate between them and no path from text to an order.
Advanced · 10 min
A bot stack is a multi-agent workflow with named roles and a written handoff. One role gathers, another tries to kill the idea, another files what you accepted. Nothing in the stack holds a signing key. The human gate is the step that cannot be skipped.
Jobs in this file: A research-alert-journal loop, a shock-to-position check, and a scout-risk-scribe desk.
Three roles — brief, alert, journal — with a human gate between them and no path from text to an order.
Advanced · 10 min
When a headline hits, an agent reads your inventory and writes what is exposed. You decide what to do.
Advanced · 8 min
A multi-agent handoff where a scout proposes, risk tries to kill it, and a scribe files only what you accepted.
Advanced · 9 min
A step where the workflow stops until a person writes an explicit accept, reject, or revise. Silence is not acceptance. A model cannot mark its own work as approved.
They can, and they get worse. Separate roles so the risk pass is not the same voice that just argued for the idea. Pass a structured handoff, not the whole chat.
Outside the agents. Read-only venue keys, if you use them at all, stay in the tool that pulls fills or balances. No seed phrase, no withdrawal permission, and no prompt that asks the model to sign.