The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill documents and provides examples for lambdify, autowrap, and ufuncify in references/code-generation-printing.md. These functions generate executable Python code or compile C/Fortran code at runtime to optimize mathematical evaluations. While legitimate for performance, these functions can be exploited if the symbolic expressions being converted originate from untrusted sources.
[DYNAMIC_EXECUTION]: The skill provides instructions for parsing strings into executable SymPy expressions using parse_expr and parse_latex (found in references/code-generation-printing.md). This creates a risk of code injection if an attacker can influence the input strings processed by the agent.
[DYNAMIC_EXECUTION]: The documentation in references/code-generation-printing.md includes a section on pickling SymPy objects using pickle.load(). Unsafe deserialization of pickle data from untrusted sources can lead to arbitrary code execution.
[COMMAND_EXECUTION]: Functions such as autowrap and codegen (documented in references/code-generation-printing.md) may invoke system-level compilers (like GCC or gfortran) and execution shells to build and run the generated mathematical modules.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface through mathematical expression parsing (parse_expr). If the agent processes expressions from external files, web content, or user inputs without sanitization, malicious logic could be embedded within the mathematical notation to influence the agent's behavior.
[EXTERNAL_DOWNLOADS]: The skill references various official documentation sites and GitHub repositories for SymPy and its dependencies. These are well-known mathematical resources and are documented neutrally as expected for an academic skill.

alterlab-sympy