The Agent Skills Directory

[PROMPT_INJECTION] (HIGH): Vulnerable to Indirect Prompt Injection (Category 8) due to extensive ingestion of external content.
Ingestion points: Interfaces like irm_qa_sh (forum questions/answers), anns_d (company announcements), and ths_hot (external hot lists) fetch text generated by third parties.
Boundary markers: Absent. The documentation does not specify the use of delimiters or 'ignore embedded instructions' warnings for external data.
Capability inventory: The skill is designed for data analysis pipelines where fetched content influences the agent's logic and the execution of validation scripts (e.g., validate_data_quality in SKILL.md).
Sanitization: Absent. There is no evidence of filtering or escaping logic applied to the external text before it enters the LLM's context.
[CREDENTIALS_UNSAFE] (LOW): The documentation (README.md, SKILL.md) instructs users to provide a TUSHARE_TOKEN via environment variables or direct code injection (pro = ts.pro_api('your_token_here')). While standard for this API, it creates a risk of token exposure if the agent environment is compromised.
[COMMAND_EXECUTION] (LOW): The skill provides Python code snippets for execution. While necessary for data analysis, this requires the agent to operate within a sandbox to prevent arbitrary code execution based on manipulated data inputs.

tushare-complete