# Clarify Intent

> >-

- **Type:** Skill
- **Install:** `agentstack add skill-sumitaich1998-jarvisvr-clarify-intent`
- **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior
- **Seller:** [sumitaich1998](https://agentstack.voostack.com/s/sumitaich1998)
- **Installs:** 0
- **Category:** [AI & ML](https://agentstack.voostack.com/c/ai-and-ml)
- **Latest version:** 0.1.0
- **License:** MIT
- **Upstream author:** [sumitaich1998](https://github.com/sumitaich1998)
- **Source:** https://github.com/sumitaich1998/jarvisvr/tree/main/skills/orchestration/clarify-intent

## Install

```sh
agentstack add skill-sumitaich1998-jarvisvr-clarify-intent
```

Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent.

## About

# Clarify Intent

Jarvis stays out of the user's way: **ask only when blocked.** A good clarifier
unblocks planning with the fewest words; a bad one interrogates the user for
detail you could reasonably assume.

## Decision rule

Ask **only if both** are true:

1. The goal is ambiguous (missing a required parameter, an unresolved referent,
   or two+ materially different interpretations), **and**
2. Guessing wrong is costly or irreversible (spending, unlocking, deleting,
   messaging the wrong person) — i.e. you can't safely pick a default.

Otherwise **assume a sensible default and proceed**, briefly stating the
assumption in the final reply (e.g. default city = current location).

## Steps

1. **Locate the gap.** Name the single missing slot (which room? which timer?
   what budget?).
2. **Offer choices** when the option set is small and known — easier than
   open-ended questions and great for a quick voice/tap answer.
3. **Ask once, narrowly**, via `agent.speech`. Optionally mirror it as a
   `notification_toast` with action buttons (`notify`) for tap selection.
4. **Resolve** the answer (`user.text` / `user.voice_transcript` /
   `client.interaction`), then continue to `task-decomposition` /
   `agent-routing`.

## Examples

Ambiguous (ask): user says "set a timer."

```json
{ "text": "Sure — how long should the timer run?", "final": true }
```

As a tappable toast (`notify` → `notification_toast`):

```json
{ "title": "Which room?", "body": "I can dim the lights — which room?",
  "severity": "info",
  "actions": [ { "id": "living", "label": "Living Room" },
               { "id": "bedroom", "label": "Bedroom" },
               { "id": "all", "label": "Whole home" } ],
  "auto_dismiss_ms": 0 }
```

The tap returns a `client.interaction{action:"tap", value:{action_id:"living"}}`
which resolves the slot.

Not ambiguous (don't ask): "what's the weather?" with a known location → assume
current location, answer, and note the assumption.

## Edge cases

- **Never chain clarifiers.** One question per turn; if still unclear after the
  answer, proceed with best effort and say so.
- **Don't clarify the obvious.** Units, formatting, and arrangement are your job,
  not the user's — pick reasonable defaults.
- **Gated actions always confirm** even when intent is clear (unlock, purchase) —
  that's consent, route it through the owning agent rather than this skill.
- **Perception referents** ("what's *this*?") are usually resolvable from gaze /
  the perception buffer — try perception first; only ask if nothing is in view.

## Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

- **Author:** [sumitaich1998](https://github.com/sumitaich1998)
- **Source:** [sumitaich1998/jarvisvr](https://github.com/sumitaich1998/jarvisvr)
- **License:** MIT

Install and usage instructions live in the source repository linked above.

## Pricing

- **Free** — Free

## Security capabilities

Automated source analysis of v0.1.0 — what this tool can access:

- **Network access:** no
- **Filesystem access:** no
- **Shell / process execution:** no
- **Environment & secrets:** no
- **Dynamic code execution:** no

*"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.*


## Versions

- **0.1.0** — security scan: passed — Imported from the upstream source.

## Links

- Listing page: https://agentstack.voostack.com/l/skill-sumitaich1998-jarvisvr-clarify-intent
- Seller: https://agentstack.voostack.com/s/sumitaich1998
- Browse the marketplace: https://agentstack.voostack.com/browse

---
Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.