crisis-response-protocol

Crisis detection logic assesses risk levels by analyzing message content and conversation history, classifying severity to trigger the appropriate safety response tier.

Safety guardrails define critical instructions for the AI and include response validation to prevent harmful outputs, ensuring the AI adheres to strict safety protocols during crisis situations.

The tiered response protocol implements distinct actions based on severity, from gentle check-ins for low risk to emergency escalation with contact numbers for high-risk suicidal ideation.

Response validation prevents harmful outputs by checking generated text against safety guardrails before delivery, ensuring the AI adheres to critical safety instructions during mental health crises.

No external dependencies are required; the protocol operates independently using internal TypeScript interfaces and functions to assess risk and manage safety responses within the AI coaching environment.

Limitations include the inability to replace human review; the protocol disables AI interaction during severe crises and displays emergency contacts, acting as a safety guardrail rather than a substitute for professional intervention.