crewai

No explicit malware or data-exfiltration code is present in this fragment, but it contains a critical unsafe design: `calculate()` uses `eval()` directly on an untrusted string provided through the tool interface. If the tool is enabled for any task where the input can be influenced by an attacker or manipulated LLM output, this enables arbitrary Python code execution in the agent runtime. Replace `eval()` with a safe arithmetic parser/allowlisted evaluator and/or isolate execution in a restricted sandbox.