| name | ask-user-questions |
| description | Ask the user via <asks> block whenever another skill needs the trader to supply missing information. Triggers: ask the user, ask, missing parameter, need clarification, need more info. |
| always | true |
Ask User Questions — <asks> block
When you need to ask the trader for input, emit an <asks> block. The
web renders it as a wizard card; Telegram flattens it to a numbered
list. The trader's answers come back in the next user message.
Response shape
Every response that asks the user MUST be two parts in order:
- One short conversational sentence acknowledging the trader's
request, in their language.
- The
<asks> block.
Never emit the block alone — the card without prose feels robotic.
Schema
<asks>
<question>
<title>{conversational question}</title>
<options>
<option>{value}</option>
<option>{value}</option>
</options>
</question>
</asks>
<title> — required. A full conversational sentence, not a label.<options> — required. 2-4 concrete preset values derived from real
data (balance %, watchlist symbols, leverage ticks, S/R levels). The
free-text input is a last-resort fallback, not a substitute for
thinking up presets.
One parameter per question, all questions in one block
Each <question> covers exactly ONE missing parameter — never bundle
multiple params into one <option>. But all missing params for the
same intent go in the SAME <asks> block, not split across multiple
turns. One block, multiple <question> children.
Notes
- Wizard is the default form for collecting ≥2 intent params (symbol/size/leverage). Use it — don't fall back to a plain markdown bullet list.
- One option = one param value. Never bundle (e.g., side+size+leverage in one button).
- Each question: 2-4 concrete presets derived from data (watchlist, % balance, leverage ticks 3x/5x/10x/20x, S/R levels). Always derive — don't ship empty options.
- No analytical questions in wizard. OrderType (market vs limit), entry price, SL, TP, wait-vs-now, direction-without-TA-bias → propose in prose with reasoning, never as options.
- One question → prose (a 1-Q wizard is heavier than a chat answer). Cap each wizard at 3 questions. Never chain two wizards — between them is prose with your analysis (entry/SL/TP/bias).
Example — collecting trade intent
Scenario: trader says "I want to enter a trade", wallet has $25 available, watchlist is [BTC, ETH, SOL], no open positions.
Good — short prose ack, then a wizard with one param per question and concrete presets:
Got it, I need a few things to set this up right.
<asks>
<question>
<title>Which coin do you want to trade?</title>
<options>
<option>BTC</option>
<option>ETH</option>
<option>SOL</option>
</options>
</question>
<question>
<title>Size in USD?</title>
<options>
<option>$5 (20% balance)</option>
<option>$10 (40% balance)</option>
<option>$15 (60% balance)</option>
</options>
</question>
<question>
<title>Leverage?</title>
<options>
<option>3x</option>
<option>5x</option>
<option>10x</option>
<option>20x</option>
</options>
</question>
</asks>
Bad — common failures to avoid:
- Markdown bullet list with no
<asks> block at all ("Which coin? Long or short? Size? Leverage?"). The wizard renders as a card; bullets render as text the trader has to type a reply to.
- Bundled options on one button:
<option>Long BTC $10 at 5x</option> — changes 4 params at once, hides what's varying.
- Empty options (
<options></options> or only <option>Something else</option>) — forces free-typing for a question with obvious presets.
- Analytical question with options:
<title>Market or limit?</title> — that's Ghost's call after looking at S/R, not the trader's.
