@fgladisch/pi-user-select

Registers user_select tool so LLM (or skills) can ask human a multiple-choice question.

Use when workflow needs explicit user input to disambiguate, confirm, or pick between mutually exclusive paths instead of guessing.

Install

pi install npm:@fgladisch/pi-user-select

No config, no slash commands.

Tool schema

| Field | Type | Required | Description | | ------------- | ---------- | -------- | ------------------------------------------------------------- | | question | string | yes | Question/prompt shown to user. | | options | Option[] | yes | Mutually exclusive choices ({ label, description? }, ≥ 1). | | allowCustom | boolean | no | When true, append a "(Type custom answer)" free-text entry. |

Example:

{
  "question": "Which package manager should I use?",
  "options": [
    { "label": "npm" },
    { "label": "pnpm", "description": "Faster, content-addressable" },
    { "label": "yarn" }
  ],
  "allowCustom": true
}

Behavior

• Interactive UI: shows question, numbered options (descriptions inline), and optional (Type custom answer) entry.
• Non-interactive (pi -p, JSON mode): throws so LLM sees error result and stops looping on tool that has no human to answer it.
• Cancellation (Esc / null / whitespace-only custom answer): returns non-error result with answer: null and cancelled: true.

Tool result text:

| Outcome | content[0].text | | ----------------- | --------------------------------------- | | Pre-baked option | User selected: <n>. <label> | | Free-text answer | User wrote: <trimmed text> | | Cancelled | User cancelled the selection | | Empty custom text | User submitted an empty custom answer |

details includes structured fields (question, options, answer, wasCustom, cancelled) for renderers and downstream skills.