Agent skeleton — the main loop

while True:
    balance = fetch_balance()
    state.last_balance = balance.stable

    positions = list(state.positions)
    if positions:
        evaluate_exits(state)       # check trailing stop, take profit, etc.
    else:
        candidates = scan_universe()
        best = pick_best(candidates)
        if best and passes_gates(best):
            execute_entry(state, best)

    save_state(state)
    sleep(poll_sec)

That's it. Our YOLO agents in production use this exact shape with ~650 lines of gates, error handling, and rails, but the core loop fits in 10 lines.

Write a skeleton function that implements this loop structure (using stubs for the exchange calls).