Progress view

A one-shot run (a job that does one task and ends) has no prompt, but a person watching it still wants to see what's happening and to answer checkpoints. createProgressView() is the chat's screen without the prompt: the live reply, tool calls, spinner, approval panels and status bar.

import { buildSessionOptions, createProgressView, runQuery } from "@falkenslab/agent-kit";

const { options } = await buildSessionOptions(config, runDir, spec);
const view = createProgressView({ mode: config.mode, agentLabel: "Scout>" });

const run = runQuery("Update the knowledge base with the new files in sources/", options);
for await (const event of run.events) view.render(event);

await view.close(); // removes the live area and restores the interaction port
console.log("Done.");

Without a TTY, or with plain: true, it's createConsoleRenderer() (plain lines), with a no-op close().

Options

Option	Description
formatAction	Tool call labels. See Tool labels; withToolLabels() adds the extensions' (An extension's labels).
toolPhrase	How your tools count in a folded summary.
toolDetail	How much of the tool calls shows: "full" (default), "calls" or "summary". See Tool labels.
formatResult	Your own result line per tool, or none. See Tool labels.
agentLabel	The label before the reply in plain output.
onWrite	Called with everything the console renderer writes (plain text), e.g. to mirror a log file.
renderApproval	Your own preview in the approval panels.
mode	Shown in the status bar.
theme, language	See Themes and Languages.
plain	Plain console output even on a TTY.

The view is a renderer

ProgressView has the same methods as the console renderer, so code that prints events works with either:

Method	What it does
render(event)	Draws one AgentEvent.
write(text), writeLine(text)	Writes your own text above the live area, and to onWrite. writeLine writes a line of its own; the view draws whole lines, so write text shows once its line ends (a \n in it, a writeLine(), endLine() or the next event).
endLine()	Ends the current line if needed.
startTurn()	Marks a new turn (the view starts one when created).
close()	Ends the view. Await it before printing anything else.

Ctrl+C during a checkpoint answers Stop and raises SIGINT, so your own handler can stop the run, and tell the person with writeLine():

let interrupting = false;
process.on("SIGINT", () => {
  if (interrupting) process.exit(130);
  interrupting = true;
  view.writeLine("Interrupting… (Ctrl+C again to exit without waiting)");
  void run.interrupt();
});

The plain console renderer

createConsoleRenderer() prints events as plain lines: the agent's text as it streams, [action] <label> per tool call, notices and errors. runChatTui() uses it, and so does the session log.

import { createConsoleRenderer, runQuery } from "@falkenslab/agent-kit";

const renderer = createConsoleRenderer({ agentLabel: "Scout>", onWrite: (text) => log.write(text) });
const run = runQuery("Summarize the notes", options);
for await (const event of run.events) renderer.render(event);
renderer.endLine();

output replaces where the text goes (standard output by default). Subagent actions, tool results and prompt suggestions aren't printed.