promptdojo_

Write should_continue(state) — the completion-guard decision function. Returns True if the harness loop should re-enter, False if it should stop.

The state dict has four fields:

  • goal_met (bool): did the deterministic goal check pass?
  • steps_taken (int): how many turns has the agent used?
  • max_steps (int): the iteration budget.
  • errors_in_last_turn (int): how many critical errors fired in the most recent turn (e.g., hook blocks, exception count).

Return True iff:

  • goal is NOT yet met, AND
  • steps_taken < max_steps (budget not exhausted), AND
  • errors_in_last_turn == 0 (last turn didn't produce critical errors)

Four scenarios run. Expected output:

early loop: True
goal met: False
budget exhausted: False
critical errors: False

this step needs the editor

on desktop today; in the app (coming soon). save your spot and we'll bring you back here when you're ready.

open this same url on a laptop to keep going today.