ncode CLI Using ncode
Questions from agents
The assistant, and the lead agent of a swarm, can stop and ask you a question when a choice is yours to make: which of two designs, which files to leave alone, whether a finding matters. Worker agents cannot ask; they report to their lead.
When an agent asks#
The question opens by itself over the conversation as one note, with every question of that call in it. Each question offers a list of answers and an other field for an answer of your own. Some questions let you tick several answers.
For a moment after the note opens, your keys keep typing into the draft, so a sentence you are writing is never taken as an answer.
Answering#
| Key | Does |
|---|---|
1 … 9 | pick that answer (in a multi-select question, tick or untick it) |
| Space | tick or untick the highlighted answer (multi-select only) |
| Tab | move between the list and the other field |
| ← / → | go to the previous or next question |
| Enter | confirm this question and go to the next; on the last one, send all the answers |
Nothing is sent until the final Enter, so you can go back and change an earlier answer first.
Leaving a question for later#
Esc puts the note aside and keeps the answers you have given so far. The agent keeps waiting. Ctrl+N brings the note back, and it also opens the next approval or question when several are waiting.
Questions in headless runs#
In a headless run nobody can answer, so a question stops the run. Give a headless task the context it needs up front, or run it in the full screen, where you can answer.