Good First Issue: Add a new AI/LLM security rule to our ruleset

[ParzivalHack]

PySpector includes a dedicated set of AI/LLM security rules. We want to expand coverage with additional patterns related to prompt injection, unsafe agent behavior, or insecure model usage.

This issue is suitable for first-time contributors.

What to do:

• Propose a new (valid) AI/LLM-related rule (check that a rule for that exact class of vulnerability isnt already present in the current AI ruleset)
• Implement it using the existing rule format
• Add a small test case or example
• Document the rule behavior briefly
  No Rust knowledge required. Python and basic static analysis understanding is enough.

Help from mantainers is available through review :)

[devesssi]

Hi! I’d like to take this. I’ll add a focused rule for create_pandas_dataframe_agent(..., allow_dangerous_code=True), with positive and negative tests. I checked the existing AI rules and don’t see this case covered.