Interactive questions

The question extension provides a terminal wizard for asking the user one or more typed questions. It supports single-select, multi-select, and free-form text questions, with keyboard navigation and filtering. Outside the TUI (e.g. RPC/VS Code hosts), the same questions are answered through the host’s extension UI dialogs.

Bundled with Selesai

Loaded automatically with Selesai; no per-extension npm install is required.

Manifest entry: ./question
Runtime surface: Agent tool

Setup and prerequisites

This extension is bundled with Selesai. It is loaded automatically; you do not need to install it separately.
Requires a UI surface. In terminal TUI mode questions render as a full-screen wizard; in RPC/VS Code hosts they render as extension_ui_request dialogs (select, input, multiselect). It cannot be used in modes without any UI.
The agent invokes the tool; the user answers through the TUI.

What it sets up

In terminal TUI mode the tool renders a full-screen batch question wizard. In non-TUI modes (e.g. RPC) questions are asked one at a time through the host’s extension UI dialogs; cancelling any question cancels the whole batch.
Multi-select questions keep full multi-selection when the host supports the multiselect UI method; older hosts fall back to a single choice.
Each question has a stable id, a label, a question string, a type, and optional options and context.
Answers are returned as a structured result with answered, skipped, and unanswered statuses.

What you can configure

The question tool is schema-driven. Questions are configured per tool call; there are no persistent user settings.

Configuration locations and precedence

Per-tool-call schema
question({ questions: [...] })
All options are supplied by the agent inside the tool call.

Settings

Key / path	Type, default, and values	Description
`questions[].id`	string Default: q1, q2, ...	Stable answer id. Auto-generated when omitted.
`questions[].label`	string Default: Q1, Q2, ...	Short page label shown in the wizard.
`questions[].question`	string Default: —	Question text shown to the user.
`questions[].context`	string Default: —	Optional plain-text context shown before the question.
`questions[].type`	string Default: — Allowed values: "select", "multiselect", "text"	Question type. image and info are not supported.
`questions[].options`	array Default: —	Required for select and multiselect. Each option has value, label, and optional description.
`questions[].allowOther`	boolean Default: false	Allow a custom text answer for select/multiselect.

Command, tool, and shortcut controls

Key / path	Description
`question({ questions })`	Batch wizard with review and submission screens.

Source evidence

src/extensions/question/schemas.ts

What you can do

Ask independent decisions that the user must make directly.
Present single-select, multi-select, or free-form text questions.
Provide stable option values separate from human-readable labels. Choice questions always include an Other option so users can answer freely; pass allowOther: false to force the listed options.

Commands, tools, and shortcuts

There are no slash commands. The tool is invoked by the agent:

question({
  questions: [
    {
      id: "target_db",
      label: "Database",
      type: "select",
      question: "Which database should we use?",
      options: [
        { value: "sqlite", label: "SQLite" },
        { value: "postgres", label: "PostgreSQL" }
      ],
      allowOther: true
    },
    {
      id: "skip_cache",
      label: "Skip cache",
      type: "text",
      question: "Why skip the cache?"
    }
  ]
})

Supported question types: select, multiselect, text.

Committing a select, multi-select, text, or custom answer advances automatically. A single answered question submits immediately. Multi-question batches keep the final review screen, where Enter submits all answers atomically; Tab/Shift+Tab or arrow keys remain available for manual navigation and editing.

Limits and safety

The tool requires a UI. It throws when no host UI is available (headless or non-interactive sessions).
Questions should be independent decisions that the user must make; the agent should inspect facts itself rather than asking the user for discoverable information.
Blank text entries are recorded as skipped; the agent must handle skipped and unanswered statuses explicitly.

Source evidence

Selesai version: 0.10.0