The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill documents functions such as parse_expr(), lambdify(), autowrap(), and ufuncify() which generate and execute Python, C, or Fortran code at runtime. It also documents the use of pickle for object serialization.
[REMOTE_CODE_EXECUTION]: The documentation covers the parse_expr() and parse_latex() functions. Since parse_expr() uses Python's eval() internally, it represents a potential execution vector if applied to unsanitized external strings. The skill explicitly warns about this risk in references/code-generation-printing.md.
[INDIRECT_PROMPT_INJECTION]: The skill exposes a surface for processing untrusted mathematical expressions that could be executed via documented parsing functions. (1) Ingestion points: Mathematical strings ingested for parsing (noted in references/code-generation-printing.md). (2) Boundary markers: The skill provides prominent warnings regarding eval usage and suggests using restricted parsing environments. (3) Capability inventory: Support for file system operations via Bash tool and runtime code compilation/execution via autowrap. (4) Sanitization: Suggests input validation using regular expressions and limited symbol dictionaries.
[EXTERNAL_DOWNLOADS]: The skill instructs the installation of standard scientific Python packages (sympy, numpy, scipy, matplotlib) from public package registries.

sympy