mrdang
Pass
Audited by Gen Agent Trust Hub on Mar 31, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches financial data from the Tushare Pro API and search results from Jina AI's web search service. These are well-known services used for their intended purpose.\n- [EXTERNAL_DOWNLOADS]: Declares dependencies on 'pandas', 'tushare', and 'requests' which are standard packages in the Python ecosystem.\n- [COMMAND_EXECUTION]: Utilizes 'uv run' to execute local Python scripts ('scripts/data.py' and 'scripts/search.py') for data processing and information retrieval.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes unverified data from external sources.\n
- Ingestion points: Web search content is retrieved from the Jina AI API in
scripts/search.pyand passed to the agent for analysis.\n - Boundary markers: The skill lacks explicit delimiters or 'ignore embedded instructions' warnings to separate ingested content from the system prompt.\n
- Capability inventory: Across its scripts, the skill can execute subprocesses via
uv run, perform network operations viarequests, and read files like.env.\n - Sanitization: No explicit sanitization, filtering, or validation of the retrieved search content is implemented before processing.
Audit Metadata