The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides explicit instructions to execute shell commands using hardcoded absolute paths on the local filesystem, such as 'D:/workspace/股票工作站' and 'C:/Users/CHAOFAN/.claude/...'. It triggers the execution of Python and Node.js scripts located in these paths, which reveals local directory structures and could lead to unauthorized code execution if the environment is compromised.
[DYNAMIC_EXECUTION]: The skill utilizes dynamic code execution at runtime. It provides custom JavaScript snippets for extracting data via browser-based evaluation tools like 'evaluate_script'. Additionally, the Python logic in 'scripts/_common.py' dynamically modifies the module search path ('sys.path.insert') to load external libraries from a specific hardcoded local virtual environment.
[DATA_EXFILTRATION]: The skill communicates with a wide array of external domains for financial data and industry reports, such as 'eastmoney.com', 'cninfo.com.cn', '100ppi.com', and 'smm.cn'. The automation of these network requests, coupled with the ability to execute local commands, creates a potential channel for data exfiltration if inputs are manipulated.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. Untrusted data enters the context via web scraping of news and bulletins ('data_sources.md'). No boundary markers or 'ignore embedded instructions' warnings are present. The skill possesses capabilities such as subprocess execution (running Python/Node scripts) and file system writes. Sanitization is limited to simple string truncation ('substring') in the extraction scripts, which does not prevent the processing of malicious instructions embedded in the external content.

trend-verify