no-prompt-without-flag

Require every interactive prompt in a CLI command to be skippable with a flag.

burgee requirement: P1 — every prompt is backed by a flag; a flag value skips the prompt
Hosts: commander (and burgee/commander, @commander-js/extra-typings), yargs (and burgee/yargs), burgee
Prompt libraries: @clack/prompts, inquirer, @inquirer/prompts and @inquirer/<prompt>, prompts, enquirer, caique (and its caique/clack, caique/inquirer drop-ins)
Recommended: off — strict only

Why

A prompt with no flag behind it is a question only a person at a keyboard can answer. An agent, a CI job and a pipe hang or fail on it (clack #167, oclif/oclif #1492). The fix is structural — read the value from the command line first, and ask only when it is missing — so the check is structural too.

Rule details

Reports a call into one of the prompt libraries above — resolved by import, never by the function's name — that sits inside a proven command handler and is not guarded by a read of that handler's inputs: its parameters (or names destructured from them), this.opts() in a function handler, or .opts() on a proven commander command.

These shapes count as backed by a flag:

const name = opts.name ?? await text({ message: 'Name?' });        // fallback
if (!opts.name) name = await text({ message: 'Name?' });           // guarded branch
const ok = opts.yes ? true : await confirm({ message: 'Sure?' });  // conditional
if (opts.yes) return remove(); await confirm({ message: 'Sure?' });// early exit
async ({ name = await text({ message: 'Name?' }) }) => {}          // destructuring default
await inquirer.prompt(questions, { name: opts.name });             // inquirer prefill
await inquirer.prompt([{ name: 'name', when: () => !opts.name }]); // inquirer gate
const v = decide({ value: opts.name, … }); if (v.action === 'prompt') await ask(…); // caique

It does not match the prompt's result variable to an option by spelling — that would be deciding by a name.

A prompt outside any command handler is not reported: it is not command code this rule can reason about.

Incorrect

import { Command } from 'commander';
import { text } from '@clack/prompts';
const program = new Command();

program.command('init').action(async () => {
  const name = await text({ message: 'Project name?' });
});

Correct

program
  .command('init')
  .option('--name <name>', 'project name')
  .action(async (opts) => {
    const name = opts.name ?? (await text({ message: 'Project name?' }));
  });

Why `strict` only

burgee's intent for this plugin holds it out of recommended until a precision study on ten real CLIs shows fewer than one false positive per hundred prompt calls. The guard is read from control flow, and control flow has more shapes than a declaration does.

Known limitations

A guard that lives in a helper (if (needsName(opts)) … where needsName is elsewhere) is read — the test reads opts — but a prompt inside a helper called from the handler is not followed.
Any read of the inputs in the guard counts; the rule does not check that the flag it reads is the one the prompt would answer.
A handler imported from another file is not read here.