AI Skills

MCP tools are "atomic capabilities" — each fetches one kind of data in a single call. Skills sit on top of the tools and provide scenario-oriented, higher-level capabilities: they compose multiple fetches, indicator calculations and signal detection into one complete analytical intent, so an AI client can drive an entire analysis pipeline from a single natural-language request.

Skills are built by composing the read-only tools from the MCP tool table with the pure-computation capabilities of stock-sdk/indicators and stock-sdk/signals. The exact skill list and triggers follow the implementation.

Built-in skills

Technical analysis

Full technical analysis of a single symbol: fetch historical K-line → overlay indicators like MA / MACD / BOLL / KDJ / RSI (stock-sdk/indicators) → detect signals such as golden / death cross and overbought / oversold (stock-sdk/signals), then have the model summarize a readable conclusion.

Example prompt:

Analyze the recent technical setup for Kweichow Moutai (sh600519) — any golden-cross signal?

Smart screening

Filter symbols across the whole market or a given board by conditions: batch-fetch quotes / K-lines → compute indicators and signals → filter and rank by price change, turnover rate, indicator crossovers, etc., returning a candidate list. Can be combined with backtesting to validate a strategy.

Example prompt:

Find A-shares with a MACD golden cross today and a turnover rate above 5%.

Market overview

A quick snapshot of the day's market: limit-up pool, stock changes, board changes, northbound capital, industry / concept board performance — summarized into a "today's market brief".

Example prompt:

How's the A-share market overall right now? Which stocks hit limit-up, and is northbound capital flowing in or out?

Live monitoring

Watch-list-style monitoring of a set of symbols: pull live quotes and intraday timeline, large orders and fund flow, and flag symbols that cross thresholds (price change, volume surge, large-order inflow, etc.).

Example prompt:

Keep an eye on my watch-list — tell me if any of them surges on volume.

How it works

Each Skill is essentially an orchestration of "a set of tool calls + computation + model synthesis":

1. The model understands the user's intent and picks the matching skill.
2. It fetches data on demand through MCP tools (quotes / K-lines / fund flow / market events, etc.).
3. It computes indicators and signals via stock-sdk/indicators and stock-sdk/signals.
4. The model synthesizes the structured results into a natural-language conclusion.

Because everything underneath is read-only, Skills never place orders or move money — they only fetch and analyze data.

Next steps

• MCP Overview: protocol and zero-dependency implementation.
• MCP Tool Table: the atomic tools the skills build on.
• Technical Indicators: the indicator and signal computation layer.