Skip to content
Back to skills

Ask User

ASecurity

Conventions for asking the user questions during a workflow via the AskUserQuestion tool. Other skills reference this to keep prompts consistent and structured.

  • 36 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added October 7, 2026
ai-agentsgobackend

Works with

  • cli

Security analysis

A100/100

Scanned October 7, 2026

npx -y skills add desplega-ai/ai-toolbox --skill ask-user --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Ask User?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Ask User
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/desplega-ai-ask-user/badge)](https://www.skillsdirectory.com/skills/desplega-ai-ask-user)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: ask-user
description: Conventions for asking the user questions during a workflow via the AskUserQuestion tool. Other skills reference this to keep prompts consistent and structured.
---

# Ask User

Thin helper that codifies how to ask the user questions during a workflow. Other skills (planning, research, brainstorming, implementing) reference this skill instead of duplicating AskUserQuestion conventions.

## Core Rule

**Use the AskUserQuestion tool — never as plain-text bullets in chat.**

Asking in chat looks like a request for free-form prose; AskUserQuestion gives the user one click per decision and the answer is structured. Always prefer the tool.

## When to Use

- Clarifying ambiguous user input
- Choosing between viable design options
- Confirming a non-trivial assumption before acting on it
- Approving a phase boundary, plan structure, or destructive action
- Capturing a workflow preference (e.g., commit-per-phase)

## When NOT to Use

- Rhetorical questions or thinking-out-loud
- Asking permission for trivial, reversible actions (just do them)
- "Is the plan ready?" / "Should I proceed?" — those are not decisions, they're checkpoints; just present the artifact and stop

## Conventions

### Question shape

- **One sentence**, ends with `?`
- Include enough context that the user doesn't need to scroll back
- Don't reference invisible artifacts ("does the plan look good?" — they may not see it yet)

### Options

- 2–4 mutually exclusive choices
- Each option: 1–5 word **label**, plus a one-sentence **description** explaining the trade-off or implication
- If you have a recommendation, put it **first** and append `(Recommended)` to the label
- Never include "Other" — the tool adds it automatically

### Header chip

- Max 12 characters, names the *axis* of the choice (e.g. `Auth method`, `Approach`, `Scope`)

### Multi-select

Use `multiSelect: true` only when choices are non-exclusive (e.g., "which features to enable?"). Default is single-select.

### Batching

If you have multiple independent questions to ask at the same checkpoint, send them in **one** AskUserQuestion call (up to 4 questions) — don't ping-pong.

## Good Example

```
question: "Which storage backend should the worker use?"
header: "Storage"
options:
  - label: "Postgres (Recommended)"
    description: "Existing infra, transactional, easy to query. Adds one new schema."
  - label: "Redis"
    description: "Faster but requires new ops. Lossy on eviction — bad fit for audit data."
  - label: "S3 + manifest"
    description: "Cheapest at scale but adds latency and a sync layer."
```

## Bad Examples

```
"What do you think?"                      → too vague, no options
"Should I continue? (yes/no)"             → checkpoint, not a decision; just stop
"Pick one: A, B, C, D, E, F, G"           → too many options; collapse or split
"Want me to also fix the typo?"           → trivial reversible action; just do it
```

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.