Determine whether this student is in crisis right now.

Read the STUDENT MESSAGE in the context of the conversation history.
A short ambiguous message must be interpreted using what came before it —
not in isolation.

{conversation_context_block}

STUDENT MESSAGE:
-----------------
{student_message}
-----------------

EMOTIONAL DETECTION:
---------------------
Primary emotion: {primary_emotion}
Intensity: {intensity}
Needs immediate attention: {needs_immediate_attention}
---------------------

ROOT CAUSE:
-----------
Primary root cause: {primary_root_cause}
Duration: {duration}
-----------

BEFORE flagging crisis, ask yourself:
- Does this message contain explicit suicidal/self-harm language?
- Or is it a question about consequences ("ab kya hoga", "usse kya hoga")?
- In the context of the conversation, is the student asking what happens next — or expressing they don't want to exist?
- Would a reasonable person reading this full conversation think this student needs emergency help RIGHT NOW?

If yes → is_crisis=true, chain_should_continue=false, write crisis_response, set urgency, list signals, include helplines.
If no  → is_crisis=false, chain_should_continue=true, everything else empty/null.

Return the JSON now.
