--- name: crypto-com-price-history-analysis description: Analyse the recent price history of one Crypto.com instrument over a chosen window, describing trend, volatility, volume context and unusual moves from candlestick and trade data, and separating what the data shows from what it cannot show. Use when the user asks how a coin or pair has moved over days or weeks, whether it is trending or choppy, how volatile it has been, whether a move was unusual, or wants a price history summary rather than a current snapshot. --- # Analyse price history for one instrument The job is to describe how a price **behaved over a window**, not to say where it is now or where it is going. A point-in-time quote answers "what is it worth now"; this skill answers "how did it get here, how bumpy was the ride, and was anything out of the ordinary". Keep those two jobs apart: if the user only wants the current price and book, a snapshot is the right tool, not this. ## 1. Pin down the instrument and the window Before fetching anything, fix two things and state them back to the user: - **Instrument symbol.** Confirm the exact instrument with `get_instrument` or `get_instruments` if the user gave a loose name. A base-only name ("BTC") is ambiguous: ask which quote currency they mean, or name the pair you picked and why. Never compare two pairs with different quote currencies as if they were the same asset. - **Window and granularity.** Ask or infer a window (for example the last 7 days, the last 90 days) and pick a candle interval that gives a sensible number of bars. Rough guide: hourly candles for a few days, daily candles for months. Too fine a granularity over a long window floods the context and hides the trend; too coarse hides the spikes the user is asking about. If the user gives no window, default to 30 days of daily candles and say so. ## 2. Pull the candles first, then the trades Use `Get_candlestick` for the window at the chosen interval. Candles carry the shape of the price: open, high, low, close and volume per bar. Work from them first. Then use `get_trades` only where you need to explain a specific bar. Trades are for zooming in on one unusual candle (was the spike one large print or many small ones?), not for reading the whole window. Fetch a few recent trades for context or for a bar of interest and say how many you looked at. Paginate if the window is long. If the response is truncated, say which part of the window you covered and do not extrapolate the rest. ## 3. Describe the trend without over-claiming Compare the first and last closes, and the highest and lowest closes, across the window. Then describe the shape, not just the endpoints: - **Trending:** closes mostly on one side of the window's midpoint, with the trend held for a stretch rather than flipping each bar. - **Ranging:** price oscillates between roughly stable highs and lows with no sustained direction. - **Reversal:** a clear move in one direction that changed sign partway through; name the bar where it turned. A single bar's direction is noise. Describe a trend only when a run of bars supports it, and say how many bars that run covers. ## 4. Measure volatility in plain terms Summarise the typical size of moves rather than quoting one dramatic number. Work out the per-bar percentage range (high minus low, over the open or prior close) and the close-to-close percentage changes. Report the median or typical absolute move alongside the largest one, so one outlier does not define the picture. Say which way the window was more volatile than the one before it only if you fetched that earlier window too. ## 5. Flag unusual moves, with a threshold Call a bar unusual only against the window's own baseline. A workable rule: a close-to-close move larger than roughly three times the median absolute move, or a range several times the typical range. Also flag a volume spike when a bar's volume is well above the window median, because a large move on heavy volume reads differently from one on thin volume. For each flagged bar, state the timestamp, the size of the move, the volume relative to typical, and whether trades support it. Then say what you can and cannot tell: a big bar on thin volume may be a thin book, not a real repricing. Do not explain *why* it happened unless the data itself shows a cause, such as a single large trade. ## Common patterns and traps - **Endpoint bias.** Quoting only start and end prices hides a window that spiked and fell back. Always describe the high, the low and the path between them. - **Mistaking one wick for a trend.** A long high or low wick on one candle is a single print, not evidence that the market went there and stayed. - **Comparing unlike windows.** A 7-day and a 90-day volatility figure use different bar counts and different sampling. Compare like with like, and say so when you cannot. - **Mixed granularity.** Stitching hourly and daily candles into one series distorts every statistic. Keep one interval per analysis. - **Thin-market spikes.** A price that jumps on very low volume may reflect a sparse order book rather than real demand. Check volume before calling a move significant. - **Confusing volume with value.** Volume is in base units unless stated; a large volume number at a low price is not necessarily more money traded than a small one at a high price. Use the quote-currency value if the data gives it. - **Forecasting by accident.** Phrases like "due for a bounce" or "likely to continue" are predictions. Describe the past; if the user asks for a view, say it is an opinion, not something the history establishes. ## Output ``` Instrument: Window: to , , Trend: > — Range: high at