The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides instructions and examples for using the pickle module to load data from files in references/code-generation-printing.md. Deserializing untrusted pickle data can lead to arbitrary code execution.
[DYNAMIC_EXECUTION]: The skill demonstrates the use of autowrap and ufuncify in references/code-generation-printing.md, which automatically compile symbolic expressions into C or Fortran code and import them as Python modules at runtime. This involves dynamic code generation and execution through external compilers (Cython, f2py).
[DYNAMIC_EXECUTION]: The skill documents parse_expr, parse_latex, and parse_mathematica in references/code-generation-printing.md. These functions convert strings into executable SymPy objects. The documentation explicitly warns that this can lead to code injection vulnerabilities, but provides it as a core capability.
[INDIRECT_PROMPT_INJECTION]: The skill introduces a significant attack surface for indirect prompt injection by encouraging the parsing of user-supplied strings into symbolic expressions.
Ingestion points: parse_expr and parse_latex functions (found in references/code-generation-printing.md) take raw string inputs.
Boundary markers: Code examples do not demonstrate the use of delimiters or warnings to treat input as data rather than instructions.
Capability inventory: The skill provides access to lambdify, autowrap, and pickle, which can execute or compile the resulting expressions.
Sanitization: While a note in references/code-generation-printing.md mentions the need for validation and sanitization, no specific implementation or best practices for doing so are provided in the examples.

sympy