How to use the A-share ST Risk Forecaster agent skill

What the skill makes the agent do

Agent skills are folders of instructions and optional scripts that an agent loads when a task matches them. This one covers the A-share rule set for ST and *ST warnings. When you give it a stock code, the agent first checks whether the company is already flagged, then tests the next fiscal year against four thresholds: revenue, net profit, net assets, and cash dividends.

The skill uses the latest data available at the time, in this order of preference: a performance forecast (forecast), a performance express report (express), the Q3 report, the interim report, and finally last year's annual report. It states which level of evidence it used for each estimate. It also adds Sina regulatory penalty records as a separate evidence source, which it fetches with its own script.

The output has two parts: a risk level for the threshold check and a confidence level for the prediction. The skill is limited to A-shares on the main board, ChiNext and STAR Market. It rejects Beijing Stock Exchange stocks and Hong Kong, US, crypto and commodity markets, because their rules differ. It does not try to detect financial fraud.

• Checks current ST status first and adjusts the analysis if the stock is already ST or *ST
• Applies board-specific thresholds based on the stock code prefix
• Predicts whether the company could be flagged in the next fiscal year
• Adds Sina penalty records as independent evidence

Install the skill from the ZIP file

Download the ZIP of the skill from the page where you found this guide. Unzip it so that the skill folder sits in one of two places. For a personal install that applies to all your projects, put it in ~/.claude/skills/. For a single project, put it in that project's .claude/skills/ folder.

If you use the Claude apps instead of a local folder, you can upload the ZIP through Settings, then Capabilities, then Skills. The skill's own files are SKILL.md and scripts/fetch_sina_penalties.py. Check that SKILL.md sits directly inside the skill folder after unzipping.

• Personal install: ~/.claude/skills/ (folder contains SKILL.md)
• Project install: .claude/skills/ inside your project
• Claude apps: Settings, then Capabilities, then Skills, then upload the ZIP

Set up the data sources it needs

The skill is built to work with the tushare skill and falls back to the akshare skill when tushare data is missing or returns nothing. Both of those skills must be available alongside this one. The source says the tushare path needs a token. The current-status check (stock_st) needs about 3000 tushare points. If you do not have those points, the skill falls back to namechange, which is free, to read the company's name history.

The akshare fallback is listed without an API key requirement. Some data is not available through akshare at all, such as the audit opinion; in that case the skill is supposed to say the data is missing rather than skip it. Python 3 is required, and the Sina penalty script uses only the standard library. That script needs internet access to vip.stock.finance.sina.com.cn.

• Python 3
• tushare skill with a token (stock_st needs about 3000 points; namechange is a free fallback)
• akshare skill as a fallback data source
• Internet access to vip.stock.finance.sina.com.cn for the penalty script

Ask for an analysis with a natural request

You do not need a special command. Write a plain request that includes a stock code. The skill is designed to start when the request is about ST or delisting risk for a specific A-share stock. If you ask about something else, it will not start on its own.

Use codes in the usual form with an exchange suffix, such as 000729.SZ or 600xxx.SH. Asking about a Beijing Stock Exchange code (8xxxxx.BJ) will get a refusal, because the skill does not apply there.

• Analyze the ST risk of 000729.SZ
• Will 600xxx be ST in its next annual report?
• Is there delisting risk for 300xxx?

Read the output with its limits in mind

The result is a rule-based estimate. The skill itself warns that forecasts can be wrong and that official announcements should be checked before any decision. Treat a high-risk result as a reason to read the company's announcements, not as a verdict.

Pay attention to the confidence level and the evidence source. A forecast-based estimate is rated higher than one extrapolated from last year's annual report. Any result that relies on akshare fallback data has its confidence lowered one step, and the output should note the data source. Rough annualisation, such as doubling the first-half figure, is rated low confidence by the skill's own rules.

• Confidence and evidence level are shown for each estimate
• Results based on akshare fallback data are marked and rated one level lower
• Official announcements should be checked before acting
• The skill does not predict fraud