Python library for CLIs that use interactive prompts, subprocess runs with Rich output, and test helpers so the same flows run in tests without a TTY.

• Install: uv add ask-shell or pip install ask-shell (Python 3.13+)
• Docs: GitHub Pages

• ask: Prompts (confirm, text, select list/dict, single and multiple choice) built on questionary, with Rich styling
• shell: Run shell commands via run, run_and_wait, and run_pool; prefixed live output, optional log dirs, retries, and interrupt handling
• console: Rich console and progress utilities; configure_logging(app) wires a Typer app to Rich logging (handler, command decorator, optional secret hiding)
• shell_events: Callbacks for run lifecycle (before/after, stdout/stderr) for custom handling
• AskShellSettings: Paths and options (e.g. run log directory); plug into your config or env

When a CLI runs without a TTY, ask-shell does not block on a prompt. It writes the question to a prompt file and exits with NonInteractivePromptError, so an agent or CI job can fill in the answer and re-run the same command.

• Prompt file: cache_root/{app}/{command}/non_interactive_prompt.yaml, one file per command. CACHE_DIR (or the platform cache dir) sets cache_root.
• Lock file: a sibling .lock held for the duration of a run. A second run over the same file fails fast with PromptSessionLockedError instead of racing.

• ASK_SHELL_NON_INTERACTIVE_PROMPT_PATH: Pin the prompt file for this run. Point it at a file with earlier answers to replay them.
• ASK_SHELL_REPLAY_PROMPT_FILE_IN_TTY: In a TTY, replay recorded answers from the file instead of asking again. Opt-in, for resuming after a crash. Non-interactive runs replay without this flag.
• ASK_SHELL_SKIP_NON_INTERACTIVE_PROMPT_FILE: Skip the prompt file and its lock. Use for a command that never prompts, so a leftover file cannot block it.
• ASK_SHELL_USE_DEFAULTS: Default true. In a non-interactive run, a prompt with a caller default returns that default instead of dumping the question. Set false to always dump and collect an answer.

When a wrapped command fails and a prompt file with questions exists, ask-shell prints a hint. The hint separates the failure from the file: the failure is unrelated to the file, and the file is left over from an earlier run. It then names the actions:

• Delete the file to start over.
• Set ASK_SHELL_REPLAY_PROMPT_FILE_IN_TTY=true in a TTY to replay recorded answers.
• Set ASK_SHELL_SKIP_NON_INTERACTIVE_PROMPT_FILE=true for a command that never prompts.

Deleting the file is safe when you are not resuming: the next run recreates it. If the failure is unrelated to prompts, fix that failure first. The hint is informational, not the cause.

The file is scoped per command, so an agent's run and a human run of the same command share one path.

• The agent leaves recorded answers. A human TTY run asks directly unless they opt in with ASK_SHELL_REPLAY_PROMPT_FILE_IN_TTY=true.
• While one run holds the lock, another run over the same file fails fast. Wait for the first run, or delete the file if it is stale.

from ask_shell import ask, shell

if ask.confirm("Run the script?", default=True):
    name = ask.text("Name:", default="world")
    shell.run_and_wait(f"echo Hello, {name}")

Use question_patcher to feed fixed responses to prompts, or raise_on_question to fail fast when a prompt is hit. Both are context managers; see the ask API docs for details.

from ask_shell import ask
from ask_shell.ask import question_patcher

with question_patcher(responses=["y", "myname"]):
    assert ask.confirm("Run?")
    assert ask.text("Name:") == "myname"

MIT. See LICENSE.