Comprehensive Guardrail run sequence

When multiple security configurations are deployed concurrently, they execute in a rigid, deterministic linear hierarchy as detailed below.

1. Content Filter - Runs regex and exact keyword matching to filter malicious strings instantly at the ingestion boundary.
2. Presidio PII - Identifies and masks incoming Personally Identifiable Information (PII) to prevent external token processing leaks.
3. Custom Guardrails - Evaluates programmatic workflows mapped out in the custom pre_call function hook block.
4. Tool Result Judge - Processes on new tool run results. Designed explicitly to neutralize indirect prompt injection payloads embedded in dynamic external content.
5. Input Judge - Applies deep semantic analysis exactly once per unique incoming user request message sequence.
6. [ LLM Inference ]
7. Presidio PII (Outbound) - Applies mask/unmask criteria to the generated response buffer string prior to standard delivery serialization.
8. Custom Guardrails - Evaluates programmatic post-processing routines mapped out in the custom post_call function hook block.
9. Tool Call Judge - Triggers immediately if the model returns intent metadata parsing to an external application function or tool configuration parameter.
10. Output Judge - Runs strictly on natural language responses. This validation logic is automatically bypassed for raw tool-call outputs.