step: 1
state: observed
action: read_task_cards
observation: {"ok": true, "matches": ["task-003", "task-003-copy"]}
decision: needs_input