For AI agents: The documentation index is at https://docs.digitalocean.com/llms.txt. Markdown versions of pages use the same URL with index.html.md in place of the HTML page (for example, append index.html.md to the directory path instead of opening the HTML document).

Usage

doctl harness-runtime prompt <session> <prompt>... [flags]

Aliases

ask

Description

Send one prompt to an existing session, wait for the agent to finish, and print its answer. Headless execution: no TUI, no keyboard input.

The prompt is everything after the session, so it need not be quoted. Pass - to read it from stdin instead.

Only the agent’s answer goes to stdout, so it can be captured directly (ANSWER=$(doctl harness-runtime prompt my-session 'What changed?')). Progress, tool calls, and the closing token and cost summary go to stderr, where they stay visible on a terminal without corrupting a pipe. Add --include-reasoning to also emit the model’s thinking on stderr, or --quiet to silence stderr entirely. With -o json the answer, run ID, status, and usage are emitted as one object instead.

The exit code reports the run: 0 when it completed, 1 when it failed (including a cancel sent from elsewhere, which -o json reports as status cancelled), 130 when this command’s Ctrl-C cancelled it, and 124 when --timeout expired, so this composes in && chains and if tests.

Ctrl-C cancels the run — the session stays up for the next prompt — and waits for it to stop, reporting status cancelled. Press Ctrl-C again to stop waiting at once. Agents that cannot cancel a run say so, and the run keeps going. A timeout, a second Ctrl-C, or SIGTERM only stop waiting locally and do not stop the agent — the run continues and its output stays readable with doctl harness-runtime logs.

A paused session is resumed on the way in. Because nothing here can ask a human, an approval request stops the command unless --on-hitl says how to answer.

Example

doctl harness-runtime prompt my-session What is the capital of France?;

Flags

Related Commands

Global Flags