SKILL.md
---
name: futu-unusual-capital-flows
description: >-
Unusual Capital Flows (资金异动): detects capital-flow unusual signals for a
user-mentioned stock with `get_financial_unusual`, including 资金分布、买卖经纪商、
资金流向、卖空数量、卖空比例 and their unusual changes. Use when the user asks about
资金动向、资金异动、净流入、净流出、主力行为、大单小单、谁在买谁在卖、买卖经纪商、
卖空异动、卖空/做空/沽空、主力是否在进场或离场，or gives a broad 异动 request such
as “英伟达异动”、“NVDA 异动”、“腾讯有没有异动/异常” without narrowing the dimension.
For broad 异动 requests, use together with `futu-unusual-technical-signal-alerts`
and `futu-unusual-derivatives-activity`. Before
calling the CLI, you must first normalize the user-mentioned stock name,
Chinese company name, English company name, or ticker into a standard symbol
such as `US.TSLA` or `HK.00700`.
allowed-tools: Bash Read
metadata:
version: 0.0.2
requires:
bins:
- futu
author: Futu
license: MIT
---
# 资金异动 · Unusual Capital Flows
Detects unusual capital-flow signals for a specific stock and formats the result as a structured Unusual Capital Flows summary.
This skill is for **unusual-move detection** rather than a regular capital-flow overview. If the data has no qualifying signal, return `No unusual activity` or the interface's no-data result, and do not add extra market commentary.
If the user only says a broad request such as `英伟达异动`, `NVDA 异动`, `腾讯有没有异常` and does not specify a dimension, treat it as a bundled 异动 request. In that case, use this skill together with `futu-unusual-technical-signal-alerts` and `futu-unusual-derivatives-activity`.
---
## Prerequisite: Futu CLI
This skill calls **`futu unusual capital-flows`** (OAuth REST via the official CLI). Do **not** use OpenD, `futu-api`, or removed `futuapi` scripts.
Before the workflow, follow [CLI runtime](../references/cli-runtime.md). If `futu unusual` is not recognized, say Unusual skills need a newer CLI — see repo `docs/migration-gaps-and-rest-api.md`.
---
## Workflow
### 1. Parse User Input
Extract the following from the user's request:
- `stock_target`: stock code, Chinese stock name, English company name, or ticker explicitly mentioned by the user
- `time_range`: default `7`; if the user says `最近 3 天` / `过去两周` / `last 5 days`, convert it to a natural-day integer
- `analysis_dimensions`: optional; only extract when the user clearly asks about one or more specific financial dimensions
- `language_id`: infer from the user's language
If the target stock is missing, ask a follow-up question instead of guessing.
### 2. Normalize the Stock Target into a Standard Symbol
Before calling the CLI, convert the user-mentioned stock target into a standard symbol such as `US.TSLA`, `HK.00700`, `SH.600519`, or `SZ.000001`.
Normalization rules:
- If the user already gives a fully qualified symbol like `US.TSLA` or `HK.00700`, use it directly.
- If the user gives a Chinese company name, English company name, or common ticker, map it to the matching market-prefixed symbol.
- If the symbol is ambiguous, ask a follow-up question instead of guessing.
Common mappings:
| User mention | Standard symbol |
|--------------|-----------------|
| 腾讯 | `HK.00700` |
| 阿里巴巴、阿里 | `HK.09988` |
| 苹果、Apple | `US.AAPL` |
| 特斯拉、Tesla | `US.TSLA` |
| 英伟达、NVIDIA | `US.NVDA` |
| 微软、Microsoft | `US.MSFT` |
| 谷歌、Google、Alphabet | `US.GOOG` |
| 亚马逊、Amazon | `US.AMZN` |
| Meta、脸书、Facebook | `US.META` |
| 台积电、TSM | `US.TSM` |
| 贵州茅台、茅台 | `SH.600519` |
| 宁德时代 | `SZ.300750` |
Ticker inference guidance:
- `TSLA`, `AAPL`, `NVDA`, `MSFT`, `GOOG`, `META` usually mean US stocks, so normalize to `US.TSLA`, `US.AAPL`, `US.NVDA`, `US.MSFT`, `US.GOOG`, `US.META`.
- `00700` by itself is ambiguous in plain text; when the context clearly refers to Tencent, normalize to `HK.00700`.
- If the same company has both HK and US listings and the user does not specify the market, ask a clarifying question.
### 3. Map Financial Intent to `analysis_dimensions`
Prefer these canonical `analysis_dimensions` values:
- `funds_distribution`: Capital distribution
- `funds_broker`: Buy/sell brokers
- `funds_flow`: Capital flow
- `short_sell_number`: Short sell volume
- `short_sell_ratio`: Short sell ratio
- `short_sell_number_and_ratio`: Short sell volume and ratio anomalous simultaneously
Selection guidance:
- If the user asks about `谁在买/谁在卖/经纪商`, include `funds_broker`.
- If the user asks about `大单小单/资金分歧/资金分布`, include `funds_distribution`.
- If the user asks about `主力连续流入流出/资金流向`, include `funds_flow`.
- If the user asks about `卖空数量`, include `short_sell_number`.
- If the user asks about `卖空比例`, include `short_sell_ratio`.
- If the user asks about `卖空数量和比例是否同时异常`, include `short_sell_number_and_ratio`.
- If the user asks a broad capital-flow question and does not narrow scope, omit `analysis_dimensions` and use the full interface.
### 4. Infer `language_id`
Use the following language mapping:
- `0`: Simplified Chinese
- `1`: Traditional Chinese
- `2`: English
- `4`: Thai
- `5`: Japanese
Default strategy:
- Chinese user -> `0`
- Traditional Chinese request -> `1`
- English user -> `2`
- If unclear, default to `0`
### 5. Call the CLI
```bash
futu unusual capital-flows <STANDARD_SYMBOL> --time-range <DAYS> [--analysis-dimensions <DIM> ...] [--language-id <ID>] --format json --source skill
```
Examples:
```bash
futu unusual capital-flows US.NVDA --time-range 7 --format json --source skill
futu unusual capital-flows HK.00700 --time-range 7 --analysis-dimensions funds_distribution funds_broker --format json --source skill
futu unusual capital-flows US.AAPL --time-range 5 --analysis-dimensions funds_flow --format json --source skill
```
### 6. Check the Result
- If the CLI exits successfully, use the JSON to build a structured capital-flow anomaly summary.
- If the CLI returns an error, surface the error message and do not fabricate results.
- If a requested class has no anomaly in the window, explicitly say `No anomaly`.
- If the downstream service reports a permission problem or no accessible data, clearly state that the stock or account lacks the required permission.
---
## Parameters
### Required
- `stock_symbol`: standard market-prefixed symbol, such as `US.TSLA`, `HK.00700`, `SH.600519`, `SZ.000001`
### Optional
- `time_range`: natural day window, default `7`
- `analysis_dimensions`: list of specific financial anomaly dimensions to inspect; omit for full scan
- `language_id`: output language, default `0`
---
## Output Rules
Present the output by anomaly class. The three high-level classes are:
- `Capital Distribution & Buy/Sell Brokers`
- `Capital Flow`
- `Short Selling`
Formatting rules:
- Always display `Time Range` as an absolute date range in the format `YYYY.M.D - YYYY.M.D`. Calculate the start date from the current date minus `time_range` days (e.g., if today is 2026.4.24 and time_range is 7, write `Time Range: 2026.4.18 - 2026.4.24`).
- If multiple abnormal dates appear within the window for one class, list them all.
- If a class has no anomaly in the window, write `No anomaly`.
- Preserve dates, direction, amount, ratio, broker names, and interpretation from the tool output.
- Do not merge different anomaly classes into one sentence.
- Do not invent thresholds, rankings, or causal explanations beyond the returned content.
- If the user only asked about a subset of financial dimensions, keep only the relevant classes.
---
## Preferred Response Template
```markdown
Time Range: {YYYY.M.D - YYYY.M.D}
- Capital Distribution & Buy/Sell Brokers: {anomaly description or "No anomaly"}
- Capital Flow: {anomaly description or "No anomaly"}
- Short Selling: {anomaly description or "No anomaly"}
```
When the user only asks for a subset, keep only the relevant classes.
---
## Behavior Rules
1. Always normalize the user-mentioned stock target into a standard symbol before calling the CLI.
2. If the target stock is missing, ask a follow-up question.
3. If the market is ambiguous, ask a follow-up question instead of guessing.
4. If the user asks a broad capital-flow question, omit `analysis_dimensions` and use the full interface.
5. Do not output raw JSON by default when talking to the user, even if the CLI returns JSON.
6. Do not interpret the result as investment advice or trading guidance.
7. If the downstream service reports a permission issue, state that clearly and do not fabricate analysis.
---
## Example User Requests
### Broad capital anomaly check
- "特斯拉最近 7 天有没有资金面异动？"
- "看看腾讯最近主力有没有异常动作"
- "帮我查一下 Apple 最近资金面有没有异常"
Mapped request:
```json
{
"stock_symbol": "US.TSLA",
"time_range": 7,
"language_id": 0
}
```
### Broker and capital-distribution only
- "腾讯最近谁在买谁在卖？"
- "00700 最近大单小单有没有分歧？"
Mapped request:
```json
{
"stock_symbol": "HK.00700",
"time_range": 7,
"analysis_dimensions": ["funds_distribution", "funds_broker"],
"language_id": 0
}
```
### Funds flow only
- "苹果最近主力资金有连续流入流出吗？"
Mapped request:
```json
{
"stock_symbol": "US.AAPL",
"time_range": 5,
"analysis_dimensions": ["funds_flow"],
"language_id": 0
}
```
### Short sell only
- "TSLA 最近卖空比例有没有异常？"
Mapped request:
```json
{
"stock_symbol": "US.TSLA",
"time_range": 7,
"analysis_dimensions": ["short_sell_ratio"],
"language_id": 0
}
```
---
## Example Interpretation Style
```markdown
Time Range: 2026.4.2 - 2026.4.9
- Capital Distribution & Buy/Sell Brokers: On 4.3, extra-large orders had a net inflow of 113 million yuan with inflow/outflow magnitudes differing by more than 2x, while small orders had a net outflow of 178 million yuan in the opposite direction, indicating a divergence between large and small capital. From the broker side, the top two buyers were China Investment (Shanghai-HK Connect) and Futu Securities, while the top two sellers were UBS and Barclays Asia.
- Capital Flow: On 4.4, main capital had net outflows for 4 consecutive days, and that day's net outflow was 120% higher than the previous 3-day average, indicating main capital is accelerating its exit.
- Short Selling: On 4.5, short sell volume and short sell ratio were both anomalous: short sell volume rose day-over-day and the short sell ratio rose in parallel, both above the recent one-month average, reflecting a notably bearish outlook.
```