Answering agents in the human inbox
When an agent needs your decision or review, read this page because it explains the inbox pages it sends and how your answer resumes its work.
Answering agents in the human inbox
You can answer an agent without finding its terminal pane when it sends a decision or review to the human inbox. Your answer is delivered back to the asking node, which can continue from the decision instead of guessing. A live display is separate: it opens in a tmux pane and is not an inbox ticket.
A decision
A decision arrives as a page with a title, a one-line subtitle, and one question at a time. The page may offer a small set of choices or a text field. Read the question, choose or write your answer, then submit it. The agent receives the completed answer in its inbox.
An agent creates this kind of page with crtr human send. It can keep the page inside the conversation or project it into the inbox. You can see pending interactions from a terminal with:
crtr human list
A document review
A document review opens the current file beside the agent conversation with a companion agent available to discuss it. Add comments where the document needs change, then approve it when it is ready. The asking agent receives the approval, or a notice that the document changed and must be read again.
Agents create a review with crtr human review new <file> --subtitle "...". The file stays live: edits made while you review are visible in the review.
A live display
A live display is not a question. It opens a rendered Markdown file in a tmux pane and keeps it up to date as the file changes. Use it when the agent needs you to watch a plan, status board, or other document while work continues.
Agents open one with crtr human show <path>. A live display does not create a job or require an answer; it is only a view of the file.
What needs your attention
crtr canvas attention list identifies the nodes with pending human interactions across the canvas. Use crtr canvas attention count when you only need the number. When an agent needs a choice, answering it is the fastest way to let its dormant node continue; no separate restart command is needed.
Canvas and dashboard
When you need to see what your agents are doing or reopen one, read this page because it connects the canvas views to the commands that inspect and revive a node.
Memory and preferences
When you want later agents to retain a fact or follow an instruction, read this page because it explains where memory is stored and how to inspect or revise it.