The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates indirect prompt injection through functions like analyze_dataset that process arbitrary natural language questions.
Ingestion points: The questions parameter in the analyze_dataset function in SKILL.md.
Boundary markers: No markers are used to isolate user queries or prevent the LLM from following embedded instructions.
Capability inventory: The skill uses the pandasai library, which can execute generated Python code.
Sanitization: No sanitization or validation of the input questions or generated code is present.
[COMMAND_EXECUTION]: The pandasai library's core mechanism is the dynamic generation and execution of Python code via exec(). This presents a risk where a crafted natural language query could lead to the execution of malicious system commands within the agent's environment.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions to install the pandasai package and its optional dependencies from the Python Package Index (PyPI).

pandas-ai