* Multi-turn jailbreak * defensive handling of conv history not existing

* Add unicode handling to keyword match * Handle non-word letter endings or startings * updated tests * Ruff formatting * Address copilot nits Thank you to @yehorkardash for identifying this in our JS version

* Run models in parallel during benchmark * Updating eval docs * Formatting * Unused context * Handling race conditions

* Multi-turn Jailbreak eval results * Delete old jailbreak ROC curve image * Update data description * Data description correction

) * Proper conversation handling with Agents * Remove duplicated code * Extract user messages for non-convo aware evals * Fix logic on which guardrails to eval * Pass kwargs to Agent with context * Extract content parts in eval * Only pass conv history to those that need it

* Enhancing URL detection: If allow list has scheme, match exactly * Validating ports * Removing redundant variable assignment * Handle trailing slash in allow lists * Allow scheme-less user input * Fix port matching * improve port matching