The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill retrieves financial data and stock information from external BaoStock servers. This represents a standard surface for indirect prompt injection if the data source were to provide malicious text intended to influence agent behavior. However, this is inherent to the skill's primary function of data retrieval.
Ingestion points: Data is fetched through the baostock library in market_data.py, financial_data.py, and stock_info.py.
Boundary markers: The skill returns data in structured JSON format, assisting the agent in distinguishing data from instructions.
Capability inventory: The skill allows network-based data retrieval and local file-writing (CSV export) via CLI arguments.
Sanitization: Scripts parse API responses into JSON/DataFrames but do not perform explicit sanitization of text fields for prompt injection patterns.
[COMMAND_EXECUTION]: The skill uses Python scripts to process data. These scripts use the standard argparse library to handle user-supplied parameters (like stock codes and dates), which are then passed to the library's API functions. This approach follows security best practices and avoids the use of unsafe execution functions like eval or exec.
[EXTERNAL_DOWNLOADS]: The skill lists baostock and pandas as dependencies. These are well-established, reputable packages in the data science and finance communities. The use of environment variables to configure server IPs is a documented feature of the library intended for proxy or mirror configurations.

baostock