The Agent Skills Directory

[DYNAMIC_EXECUTION]: The function op_z3solve in scripts/run.py utilizes the eval() function to parse and execute constraint equations provided in the input payload. Although it attempts to sandbox the execution by setting __builtins__ to an empty dictionary and providing a restricted environment (env containing only Z3 symbols), eval() remains a high-risk function that could potentially be bypassed.
[DYNAMIC_EXECUTION]: The skill extensively uses sp.sympify() (via the _sym helper) to convert user-provided strings into SymPy expressions. While sympify is designed for mathematical parsing, it involves dynamic translation of strings into executable objects and can be targeted with malicious inputs to cause resource exhaustion or unexpected behavior.
[INDIRECT_PROMPT_INJECTION]: The skill exposes a large attack surface via the auto and calc operations, which ingest raw strings for processing. There is no evidence of rigorous sanitization for complex injection patterns in the mathematical expressions. If the agent processes data from untrusted external sources (like a web page or user-provided file) and passes it to this skill, it could lead to the execution of unintended logic within the SymPy/Z3 engines.
[EXTERNAL_DOWNLOADS]: The skill relies on external libraries including sympy, z3-solver, and pyswisseph. While these are well-known mathematical packages, they represent a dependency chain that must be managed, though their use here aligns with the skill's stated purpose.

math