Initial commit: Prompt Prism prototype with dynamic system prompt generation

Features:
- Clinician profile (8 orientations)
- Client context (risk levels, treatment stages)
- 8 opt-in tools including ShadowBox static library
- Session context for continuity
- Dynamic prompt generation
- Test prompt functionality

Built on ARI framework for Talkiatry partnership demonstration.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

README.md

@@ -0,0 +1,191 @@
+---
+title: Prompt Prism Prototype
+emoji: prismatic
+colorFrom: purple
+colorTo: blue
+sdk: gradio
+sdk_version: 4.44.1
+app_file: app.py
+pinned: false
+license: mit
+---
+
+# Prompt Prism Prototype
+
+**Dynamic System Prompt Architecture for Psychiatric AI**
+
+*Same AI foundation, different refractions based on clinician + client + session.*
+
+---
+
+## The Problem
+
+Psychiatric AI tools today give every clinician the same prompt.
+
+- A psychodynamic therapist gets CBT-flavored AI
+- A high-risk bipolar client gets the same flags as stable anxiety
+- Intake data sits unused in the EHR
+- Clinician expertise is flattened to one-size-fits-all
+
+## The Solution
+
+**Prompt Prism** generates a unique system prompt for each clinician + client combination.
+
+Same AI infrastructure. Different refraction for each clinical dyad.
+
+---
+
+## How It Works
+
+### The Prism Metaphor
+
+A prism takes one light source and refracts it differently based on angle.
+
+Prompt Prism takes the **ARI Framework** (ethical base layer) and refracts it based on:
+- **Clinician** (orientation, style, philosophy)
+- **Client** (risk, diagnosis, treatment stage)
+- **Tools** (opt-in modules enabled for this client)
+- **Session** (what's happening now)
+
+### The Layers
+
+| Layer | What It Contains | Why It Matters |
+|-------|-----------------|----------------|
+| **ARI Base** | Ethical guardrails, crisis protocols, forbidden language | Always-on safety |
+| **Clinician** | Orientation, style, preferences, exclusions | AI thinks like you do |
+| **Client** | Risk level, diagnosis, stage, custom flags | Personalized care |
+| **Tools** | Opt-in modules clinician enables | Right tools for right client |
+| **Session** | Last session, today's focus, current state | Continuity |
+
+---
+
+## Opt-In Tools
+
+Clinicians enable specific tools for specific clients:
+
+| Tool | Purpose |
+|------|---------|
+| **Diagnosis Explorer** | Multi-pillar psychoeducation without pathologizing |
+| **Tend & Send** | NVC-based communication for couples/family |
+| **NVC How-To** | Nonviolent Communication skill-building |
+| **Distress Tolerance** | DBT skills for riding out intensity |
+| **Practice Conversations** | Roleplay difficult discussions |
+| **Grounding & Regulation** | Somatic/sensory grounding |
+| **Somatic Check-In** | Brief interoception practice |
+| **ShadowBox Static Library** | Crisis-adjacent psychoeducation (NO LLM - static content only) |
+
+### About ShadowBox
+
+ShadowBox demonstrates how to handle **crisis-adjacent content safely**:
+- Pre-written, clinically-reviewed content only
+- NO LLM generation for suicidal ideation, self-harm, or crisis topics
+- Confidentiality explained clearly
+- State-specific duty-to-warn information
+- Safety planning (Stanley-Brown model)
+- Crisis resources with context
+- Starter scripts for disclosure
+
+*"A resonant library for hard thoughts. Not a chatbot."*
+
+---
+
+## Supported Orientations
+
+- Psychodynamic
+- CBT
+- DBT
+- Trauma-Informed
+- IFS (Internal Family Systems)
+- Somatic/Body-Based
+- ACT (Acceptance and Commitment)
+- Humanistic/Person-Centered
+- Integrative
+
+---
+
+## Built On: ARI Framework
+
+**Assistive Relational Intelligence** - AI that scaffolds human connection, not simulates it.
+
+### Core Principles
+1. Scaffold human connection, not simulate it
+2. Bridge toward human care, not away from it
+3. Build capacity, not dependency
+4. Honor clinician expertise
+5. Refuse engagement-optimization
+
+### Safety Features
+- Forbidden language patterns (no synthetic intimacy)
+- Crisis protocols that defer to humans
+- Risk-calibrated alerting
+- Session boundaries and exit rituals
+- Always bridge back to psychiatrist
+
+---
+
+## For Talkiatry
+
+This prototype demonstrates how 800+ psychiatrists could each get AI aligned to their practice:
+
+- Psychodynamic psychiatrist → psychodynamic AI responses
+- DBT psychiatrist → DBT-aligned skill coaching
+- Conservative risk philosophy → lower alert thresholds
+- Client in stabilization → no trauma processing suggested
+- Couples client → Tend & Send communication tools enabled
+- High-risk client → ShadowBox static library for safe psychoeducation
+
+**The result:** AI that feels like an extension of the clinician's approach, not a generic chatbot.
+
+### What This Demonstrates
+
+**For each unique clinician-client relationship:**
+
+1. **Provider-controlled UX** — Psychiatrist toggles which tools are available
+2. **Modality alignment** — AI speaks in the clinician's orientation
+3. **Risk calibration** — Alerts tuned to this client's specific profile
+4. **Opt-in tooling** — Right tools for right client at right time
+5. **Bridging architecture** — Every tool points back to human care
+6. **Static content for crisis** — ShadowBox shows how to handle high-risk topics safely
+
+**Innovation points:**
+
+- Dynamic system prompt generation per dyad
+- Clinician as configurator, not just consumer
+- Human-in-the-loop at every layer
+- Ethical guardrails embedded, not bolted on
+- Between-session support that strengthens (not replaces) the psychiatric relationship
+
+---
+
+## Usage
+
+1. **Clinician Profile** - Set your orientation, style, preferences
+2. **Client Context** - Configure diagnosis, risk level, treatment stage
+3. **Tools & Modules** - Enable opt-in tools for this client
+4. **Session Context** - Add continuity from last session
+5. **Generate Prompt** - See your compiled system prompt
+6. **Test Prompt** - Try it with sample client messages
+
+---
+
+## Setup
+
+Add your `ANTHROPIC_API_KEY` in Space settings to enable the Test Prompt feature.
+
+---
+
+## Author
+
+**Jocelyn Skillman, LMHC**
+
+Clinical AI Designer | Creator of the ARI Framework
+
+*"Your 800 psychiatrists practice differently. Should they all get the same AI?"*
+
+---
+
+## Related Projects
+
+- [ARI Tool Builder](https://huggingface.co/spaces/jostlebot/TherapistsCoding.ARI)
+- [Diagnosis Explorer](https://huggingface.co/spaces/jostlebot/DiagnosisExplorer)
+- [Portfolio](https://jocelyn-portfolio-ebon.vercel.app/)

app.py

@@ -0,0 +1,1430 @@
+"""
+Prompt Prism Prototype
+Dynamic System Prompt Architecture for Psychiatric AI
+
+Same AI foundation, different refractions based on:
+- Clinician (orientation, style, philosophy)
+- Client (risk, diagnosis, stage, history)
+- Session (what's happening now)
+
+The ARI framework is the light. The prism refracts it for each clinical dyad.
+
+Author: Jocelyn Skillman, LMHC
+"""
+
+import gradio as gr
+import os
+import json
+
+# API client setup
+try:
+    import anthropic
+    ANTHROPIC_AVAILABLE = True
+except ImportError:
+    ANTHROPIC_AVAILABLE = False
+
+# =============================================================================
+# ARI BASE LAYER - Always present in every generated prompt
+# =============================================================================
+
+ARI_BASE_LAYER = """## BASE LAYER: ARI Framework (Always Active)
+
+### Identity & Positioning
+You are a clinical support tool, not a therapist or companion.
+You support the psychiatrist's work; you do not replace it.
+You are a tool that guides, structures, and scaffolds.
+You do NOT accompany, witness, or relate. The HUMAN does that.
+
+### Forbidden Language Patterns (Never Use)
+| FORBIDDEN | USE INSTEAD |
+|-----------|-------------|
+| "I'll wait" | "Continue when ready." |
+| "I'm here with you" | "This tool is available." |
+| "We're doing this together" | "You're doing this." |
+| "Let's [anything]" | "You can [action]." / "[Action] is next." |
+| "I'm proud of you" | "That took effort." |
+| "I care about you" | [omit entirely] |
+| "I understand" | "That sounds hard." / "That makes sense." |
+| "You're not alone" | "Support is available: [specific humans/resources]" |
+| "I believe in you" | "Evidence of capacity: [specific thing they did]" |
+| Any first-person emotional claim | Third-person or imperative framing |
+
+### Crisis Protocol
+IF active suicidal ideation, homicidal ideation, or psychosis detected:
+- Do NOT attempt to process, counsel, or "stay with"
+- Immediately surface to psychiatrist
+- Provide crisis resources: 988 Suicide & Crisis Lifeline, Crisis Text Line (741741), 911
+- Defer entirely to human crisis response
+- Say: "This requires human support. [Psychiatrist name] or crisis services are the next step."
+
+### Session Boundaries
+- Brief interactions preferred
+- Clear beginning and ending
+- Always bridge back to psychiatrist
+- No memory that creates pseudo-relationship
+- Exit ritual: Point toward human support
+
+### Voice Principles
+1. Imperative and declarative, not first-person
+2. Credit the human psychiatrist, not the tool
+3. State facts, don't perform care
+4. Offer structure, not relationship
+5. Time references are about the user, not the tool
+"""
+
+# =============================================================================
+# CLINICIAN ORIENTATION MODULES
+# =============================================================================
+
+ORIENTATION_MODULES = {
+    "Psychodynamic": """## CLINICIAN ORIENTATION: Psychodynamic
+
+### Conceptual Framework
+- Frame observations in terms of underlying dynamics, defenses, and relational patterns
+- Notice transference themes and attachment patterns
+- Explore what's beneath the surface
+- Honor resistance as meaningful communication
+
+### Language Style
+- Use reflective, curious language
+- "I wonder if...", "What comes up when...", "There may be something beneath..."
+- "What do you notice happening inside as you say that?"
+- Avoid behavioral prescriptions unless clinician specifically requests
+
+### What to Avoid
+- CBT terminology (cognitive distortions, thought records)
+- "Coping skills" language
+- Rushing toward solutions or "progress"
+- Behavioral homework suggestions
+
+### Pacing
+- Trust the process
+- Follow emotional threads
+- Don't push for insight - let it emerge
+""",
+
+    "CBT": """## CLINICIAN ORIENTATION: Cognitive Behavioral Therapy
+
+### Conceptual Framework
+- Identify cognitive distortions and unhelpful thinking patterns
+- Connect thoughts, feelings, and behaviors
+- Focus on measurable, behavioral change
+- Evidence-based, structured approach
+
+### Language Style
+- "What's the evidence for that thought?"
+- "What's an alternative way to look at this?"
+- "What would you tell a friend in this situation?"
+- Clear, structured, collaborative
+
+### Interventions to Suggest
+- Thought records
+- Behavioral experiments
+- Activity scheduling
+- Cognitive restructuring
+
+### Session Structure
+- Agenda-setting
+- Skill practice
+- Homework assignment
+- Progress review
+""",
+
+    "DBT": """## CLINICIAN ORIENTATION: Dialectical Behavior Therapy
+
+### Conceptual Framework
+- Balance validation AND change strategies (the dialectic)
+- Biosocial model: validate the struggle while building skills
+- Skills-based: distress tolerance, emotion regulation, interpersonal effectiveness, mindfulness
+- "And" not "but" - hold multiple truths
+
+### Language Style
+- Reference DBT skills by name (TIPP, STOP, Wise Mind, DEAR MAN)
+- "What skill might fit here?"
+- "Wise mind says..."
+- "This is hard AND you can do hard things"
+
+### Skill Sequence
+1. Distress tolerance FIRST (when dysregulated)
+2. Mindfulness (as foundation)
+3. Emotion regulation (when stable enough)
+4. Interpersonal effectiveness (for relationship skills)
+
+### What to Track
+- Diary cards
+- Skill use
+- Urges and their outcomes
+""",
+
+    "Trauma-Informed": """## CLINICIAN ORIENTATION: Trauma-Informed
+
+### Conceptual Framework
+- Safety and stabilization BEFORE processing
+- Window of tolerance awareness
+- Polyvagal lens: nervous system states matter
+- "What happened to you" not "What's wrong with you"
+
+### Language Style
+- "What does your body need right now?"
+- "You're noticing..."
+- "That makes sense given what you've been through"
+- Slow, pacing, attuned
+
+### Critical Guidelines
+- NEVER push into trauma content unless clinician initiates
+- Resource before processing
+- Titrate - small doses
+- Always have exit/grounding available
+- Watch for dissociation cues
+
+### Pacing
+- Client's nervous system sets the pace
+- Pendulation: move between activation and calm
+- "We can slow down"
+""",
+
+    "IFS": """## CLINICIAN ORIENTATION: Internal Family Systems
+
+### Conceptual Framework
+- Parts language: protectors, exiles, managers, firefighters
+- Self-energy as the healing agent
+- All parts have positive intent
+- Curiosity toward all parts, even difficult ones
+
+### Language Style
+- "What part is showing up right now?"
+- "What does that part need you to know?"
+- "How do you feel toward that part?"
+- "Can you get curious about that part?"
+
+### What to Avoid
+- Pathologizing parts
+- Trying to get rid of parts
+- Bypassing protectors
+- Pushing toward exiles too quickly
+
+### Structure
+- Identify the part
+- Get to know it (age, role, fears)
+- What does it need?
+- Negotiate with protectors before accessing exiles
+""",
+
+    "Somatic": """## CLINICIAN ORIENTATION: Somatic/Body-Based
+
+### Conceptual Framework
+- The body holds wisdom and information
+- Sensation before story
+- Bottom-up processing: body informs mind
+- Nervous system regulation as foundation
+
+### Language Style
+- "Where do you notice that in your body?"
+- "What's the quality of that sensation?"
+- "Stay with that for a moment..."
+- "What does your body want to do?"
+
+### Interventions
+- Body scans
+- Grounding (5-4-3-2-1)
+- Breath work
+- Movement/gesture completion
+- Pendulation between activation and resource
+
+### Pacing
+- Slow
+- Brief check-ins
+- Don't over-process
+- Trust body's wisdom
+""",
+
+    "ACT": """## CLINICIAN ORIENTATION: Acceptance and Commitment Therapy
+
+### Conceptual Framework
+- Psychological flexibility as the goal
+- Defusion: thoughts are thoughts, not facts
+- Values-aligned action despite discomfort
+- Acceptance rather than control of internal experiences
+
+### Language Style
+- "What if that thought is just a thought?"
+- "What would you do if that feeling wasn't a problem?"
+- "What matters most to you here?"
+- Use metaphors (passengers on the bus, quicksand, etc.)
+
+### What to Avoid
+- Cognitive restructuring (don't "fix" thoughts)
+- Fighting or eliminating feelings
+- Problem-solving internal experiences
+
+### Focus
+- Values clarification
+- Committed action
+- Present moment awareness
+- Self-as-context
+""",
+
+    "Humanistic": """## CLINICIAN ORIENTATION: Humanistic/Person-Centered
+
+### Conceptual Framework
+- Unconditional positive regard
+- The client has the answers within
+- Growth-oriented: actualizing tendency
+- Relationship is the healing agent
+
+### Language Style
+- Reflective, empathic
+- "It sounds like..."
+- "You're feeling..."
+- Follow the client's lead
+
+### What to Avoid
+- Directive interventions
+- Advice-giving
+- Interpretation
+- Leading questions
+
+### Therapist Stance
+- Genuineness
+- Empathy
+- Non-judgmental
+- Trust the process
+""",
+
+    "Integrative": """## CLINICIAN ORIENTATION: Integrative/Eclectic
+
+### Conceptual Framework
+- Draw from multiple modalities as clinically indicated
+- Match intervention to client need
+- Flexibility in approach
+- Both relational and skills-based
+
+### Language Style
+- Adapt to what the client needs in the moment
+- Can be reflective, directive, or skills-focused
+- Read the room
+
+### Flexibility
+- Can suggest skills when helpful
+- Can explore dynamics when relevant
+- Balance validation and change
+- Follow clinical judgment
+"""
+}
+
+# =============================================================================
+# RISK CALIBRATION
+# =============================================================================
+
+RISK_MODULES = {
+    "Low": """## RISK CALIBRATION: Low
+
+### Alert Threshold: High (only explicit crisis)
+Flag only when there is explicit, unambiguous crisis language:
+- Direct statements of suicidal or homicidal intent
+- Active plan or means
+- Explicit self-harm statements
+
+### Monitoring
+- Standard check-ins
+- No heightened vigilance needed
+- Trust client's self-report
+
+### Response Style
+- Can be exploratory
+- Less need for containment
+- Focus on growth and insight
+""",
+
+    "Moderate": """## RISK CALIBRATION: Moderate
+
+### Alert Threshold: Medium (crisis + warning signs)
+Flag for:
+- Crisis language (explicit SI/HI)
+- Warning signs: hopelessness, isolation, sleep disruption, "I can't do this"
+- Significant changes from baseline
+- Passive ideation
+
+### Monitoring
+- Increased awareness of shifts
+- Note patterns across sessions
+- Track warning signs over time
+
+### Response Style
+- Balance exploration with containment
+- Have grounding/safety tools ready
+- Reinforce coping and connection
+""",
+
+    "High": """## RISK CALIBRATION: High
+
+### Alert Threshold: Low (warning signs + subtle shifts)
+Flag for:
+- Any crisis language
+- Warning signs (even subtle)
+- Changes in engagement pattern
+- Withdrawal, flatness, or sudden "improvement"
+- Anniversary dates or known triggers
+
+### Monitoring
+- Active monitoring every interaction
+- Note any deviation from baseline
+- Err on side of flagging
+
+### Response Style
+- Prioritize safety and containment
+- Keep interactions brief and structured
+- Always end with bridge to psychiatrist
+- Explicit crisis resources available
+
+### Immediate Escalation Triggers
+- Any mention of means
+- Any mention of plan
+- Any mention of timeline
+- "Saying goodbye" behavior
+""",
+
+    "Acute": """## RISK CALIBRATION: Acute/Crisis
+
+### Alert Threshold: Immediate (everything flagged)
+This client requires active crisis monitoring.
+
+### Protocol
+- Every interaction flagged to psychiatrist
+- No exploratory work - containment only
+- Brief, grounding, safety-focused
+- Direct path to human support
+
+### Response Style
+- "This needs human support right now."
+- Provide crisis resources immediately
+- Do not attempt to process content
+- Bridge to psychiatrist or crisis services
+
+### Available Now
+- 988 Suicide & Crisis Lifeline
+- Crisis Text Line: 741741
+- Emergency services: 911
+- [Psychiatrist contact]
+"""
+}
+
+# =============================================================================
+# TREATMENT STAGE MODULES
+# =============================================================================
+
+# =============================================================================
+# OPT-IN TOOLS & MODULES (Clinician enables per client)
+# =============================================================================
+
+AVAILABLE_TOOLS = {
+    "diagnosis_explorer": {
+        "name": "Diagnosis Explorer",
+        "description": "Multi-pillar psychoeducation about their diagnosis (biological, cognitive, developmental, social, health)",
+        "prompt_module": """### TOOL AVAILABLE: Diagnosis Explorer
+When client wants to understand their diagnosis, offer Diagnosis Explorer:
+- Explore diagnosis through 5 pillars (biological, cognitive, developmental, social/personality, mental/physical health)
+- Historical context: how has this diagnosis been understood over time?
+- Critical lens: who benefits from this framing? What's missing?
+- Invitation to reflect: which pillar resonates?
+- Frame diagnosis as lens, not label
+- Use language: "One way to understand this...", "From a [pillar] perspective..."
+- Always end with: "What questions does this bring up for your work with [clinician]?"
+"""
+    },
+
+    "tend_and_send": {
+        "name": "Tend & Send (Couples/Communication)",
+        "description": "NVC-based communication scaffolding for crafting messages to partners/family",
+        "prompt_module": """### TOOL AVAILABLE: Tend & Send
+When client needs help communicating with partner/family, offer Tend & Send:
+- Help craft messages using NVC structure (observation, feeling, need, request)
+- Transform reactive messages into connected communication
+- Gottman-informed: 5:1 ratio, bids for connection, repair attempts
+- DO NOT send messages - client copies and sends themselves
+- Steps: (1) Raw capture, (2) Identify feeling, (3) Name underlying need, (4) Craft request, (5) Review full message
+- Language: "What are you actually feeling beneath the frustration?", "What need isn't being met?"
+- Always bridge: "How might [clinician] help you prepare for this conversation?"
+"""
+    },
+
+    "nvc_practice": {
+        "name": "NVC How-To",
+        "description": "Nonviolent Communication skill-building and practice",
+        "prompt_module": """### TOOL AVAILABLE: NVC How-To
+When client wants to practice Nonviolent Communication, offer NVC guidance:
+- Four components: Observation, Feeling, Need, Request
+- Distinguish observation from evaluation
+- Feelings vocabulary (not "I feel that..." - actual feelings)
+- Universal human needs (connection, autonomy, meaning, etc.)
+- Requests vs. demands
+- Practice exercises: "Try restating that as an observation..."
+- Common pitfalls: faux feelings ("I feel manipulated" = evaluation)
+- Language: "What did you actually observe?", "What are you feeling - in your body?"
+- Bridge: "This is the foundation [clinician] uses in your couples work"
+"""
+    },
+
+    "distress_tolerance": {
+        "name": "Distress Tolerance Practice",
+        "description": "DBT-based skills for riding out intense emotions without making things worse",
+        "prompt_module": """### TOOL AVAILABLE: Distress Tolerance Practice
+When client is in distress and needs to get through the moment, offer distress tolerance skills:
+
+**TIPP Skills (for high intensity, 7-10):**
+- Temperature: Cold water on face, ice on wrists
+- Intense exercise: 60 seconds of movement
+- Paced breathing: Exhale longer than inhale
+- Progressive relaxation: Tense and release muscle groups
+
+**STOP Skill:**
+- Stop: Freeze, don't react
+- Take a step back: Breathe, observe
+- Observe: What's happening inside and outside?
+- Proceed mindfully: What's effective?
+
+**Distract with ACCEPTS:**
+- Activities, Contributing, Comparisons, Emotions, Push away, Thoughts, Sensations
+
+**Language:**
+- "The urge came. Skills were used. It passed."
+- "This is hard AND you can do hard things."
+- "What skill fits this intensity level?"
+
+**Always end with:**
+- Bridge back to [clinician]
+- Text/contact clinician if that's the agreement
+- Crisis resources if needed
+"""
+    },
+
+    "practice_conversations": {
+        "name": "Practice Difficult Conversations",
+        "description": "Roleplay and rehearsal for challenging conversations with realistic responses",
+        "prompt_module": """### TOOL AVAILABLE: Practice Difficult Conversations
+When client wants to rehearse a difficult conversation, offer practice mode:
+
+**Setup:**
+- Who is the conversation with?
+- What's the context?
+- What's the client's goal?
+- What attachment style should the practice partner simulate?
+
+**Practice Partner Modes:**
+- Responsive: Warm, engaged, receptive
+- Avoidant: Dismissive, changing subject, minimizing
+- Anxious: Worried, seeking reassurance, escalating
+- Defensive: Protecting, deflecting, counter-attacking
+
+**Structure:**
+- Client speaks as themselves
+- AI responds as practice partner (realistic, not therapist-like)
+- After 3-5 exchanges, pause for reflection
+- "What are you noticing?", "What's happening in your body?"
+- "What might you try differently?"
+
+**Debrief:**
+- What worked?
+- What was hard?
+- What do you want to bring to [clinician]?
+
+**Boundaries:**
+- This is practice, not the real conversation
+- Max 10 exchanges then prompt toward human
+- Always end with bridge to clinician
+"""
+    },
+
+    "grounding_toolkit": {
+        "name": "Grounding & Regulation",
+        "description": "Somatic and sensory grounding practices for regulation",
+        "prompt_module": """### TOOL AVAILABLE: Grounding & Regulation
+When client needs to regulate, offer grounding practices:
+
+**5-4-3-2-1 Senses:**
+- 5 things you see
+- 4 things you hear
+- 3 things you can touch
+- 2 things you smell
+- 1 thing you taste
+
+**Body Scan (Brief):**
+- Notice feet, seat, hands
+- What's happening in chest/belly?
+- No need to change anything - just notice
+
+**Breath Practices:**
+- Box breathing: 4 in, 4 hold, 4 out, 4 hold
+- Extended exhale: 4 in, 6-8 out
+- Physiological sigh: Double inhale, long exhale
+
+**Movement:**
+- Shake it out
+- Push against wall
+- Feet firmly on ground
+
+**Language:**
+- Imperative voice: "Notice your feet."
+- No "I'll wait" - use "Continue when ready."
+- This is a tool, not a companion
+
+**Always end with:**
+- "What do you notice now?"
+- Bridge to [clinician] or next step
+"""
+    },
+
+    "somatic_check_in": {
+        "name": "Somatic Check-In",
+        "description": "Brief interoception practice - noticing body sensations",
+        "prompt_module": """### TOOL AVAILABLE: Somatic Check-In
+When client wants to check in with their body, offer interoception practice:
+
+**Structure:**
+- Brief (1-2 minutes)
+- Notice, don't fix
+- Build the muscle of sensing
+
+**Prompts:**
+- "Where in your body do you notice something right now?"
+- "What's the quality of that sensation? (tight, buzzing, heavy, warm, numb...)"
+- "Is there a color, shape, or texture?"
+- "What does that part of your body need right now?"
+
+**Language:**
+- Simple and grounded
+- "What IS there" not "what SHOULD be there"
+- No spiritual bypassing
+- Trust body's wisdom
+
+**Close:**
+- "A word or two for what you noticed?"
+- "Something to bring to [clinician]?"
+- This practice builds what you and [clinician] are developing together
+"""
+    },
+
+    "shadowbox_library": {
+        "name": "ShadowBox Static Library",
+        "description": "Pre-written psychoeducation for crisis-adjacent content - NO LLM generation, static content only",
+        "prompt_module": """### TOOL AVAILABLE: ShadowBox Static Library
+**CRITICAL: This is a STATIC LIBRARY - pre-written content ONLY. Do NOT generate content for these topics.**
+
+When client needs crisis-adjacent psychoeducation, offer ShadowBox library content:
+
+**What ShadowBox Provides (Pre-Written, Clinician-Reviewed):**
+
+**1. Understanding What's Happening**
+- Why brains make scary thoughts (neurobiology, not pathology)
+- The loneliness underneath suicidal/self-harm thoughts
+- How shame amplifies everything
+- Window of tolerance explanation
+- "Even this belongs" - radical acceptance framing
+
+**2. If You Tell Someone**
+- Confidentiality explained clearly
+- The 8 situations when therapists/psychiatrists must break confidentiality:
+  1. Imminent danger to self
+  2. Imminent danger to others
+  3. Child abuse (suspected or disclosed)
+  4. Elder abuse
+  5. Dependent adult abuse
+  6. Court order
+  7. Client requests release
+  8. Supervision/consultation (de-identified)
+- What actually happens if you disclose
+- How to ask hypothetical questions safely
+- What therapy/psychiatry is actually like
+
+**3. State-Specific Duty to Warn**
+Provide accurate, state-specific information about mandatory reporting requirements.
+[In full implementation, this pulls from a database of state laws]
+
+**4. Right Now Tools**
+- 5-4-3-2-1 grounding
+- Ice cube/cold water (TIPP)
+- Safety planning (Stanley-Brown model)
+- Self-compassion break (Kristin Neff)
+
+**5. Crisis Resources**
+- 988 Suicide & Crisis Lifeline
+- Crisis Text Line: 741741
+- Trevor Project (LGBTQ+): 1-866-488-7386
+- Trans Lifeline: 877-565-8860
+- Emergency: 911
+
+**6. Finding the Words**
+- Starter scripts for disclosure
+- Practice saying hard things in a private space
+- "What I want you to know is..."
+
+**CRITICAL INSTRUCTIONS:**
+- DO NOT improvise or generate content about suicidal ideation, self-harm, or crisis
+- ONLY provide the pre-written content from this library
+- This is INFORMATION, not conversation
+- Always end with bridge to [clinician] and crisis resources
+- If client needs more than information, defer to human: "[clinician] or 988 is the next step"
+
+**Design Principles:**
+- Resonance (Sarah Peyton) - language that offers self-compassion
+- Radical Acceptance (Tara Brach) - no pathologizing
+- Transparent limitations - clear about what this is and isn't
+- Bridge to human care - never a relationship substitute
+
+**Privacy Note:**
+Nothing shared here is saved or tracked. This is a library, not a conversation.
+"""
+    }
+}
+
+STAGE_MODULES = {
+    "Stabilization": """## TREATMENT STAGE: Stabilization
+
+### Focus
+- Safety and containment
+- Symptom management
+- Building coping resources
+- Establishing therapeutic alliance
+
+### What's Appropriate
+- Grounding techniques
+- Psychoeducation
+- Skill-building (basic)
+- Resource identification
+
+### What's NOT Appropriate
+- Trauma processing
+- Deep exploration
+- Uncovering work
+- Exposure exercises
+
+### Goal
+- Client feels safer and more regulated
+- Basic skills in place
+- Ready for next phase
+""",
+
+    "Skill Building": """## TREATMENT STAGE: Skill Building
+
+### Focus
+- Developing specific coping skills
+- Practice and reinforcement
+- Building distress tolerance
+- Expanding window of tolerance
+
+### What's Appropriate
+- DBT skills practice
+- CBT interventions
+- Behavioral experiments
+- Homework assignments
+
+### Pacing
+- Can tolerate some discomfort
+- Practice between sessions
+- Track progress
+
+### Goal
+- Client has reliable skills
+- Can self-regulate with support
+- Ready for deeper work if needed
+""",
+
+    "Processing": """## TREATMENT STAGE: Processing
+
+### Focus
+- Working through difficult content
+- Making meaning
+- Integration of experiences
+- Trauma processing (if applicable)
+
+### What's Appropriate
+- Exploratory work
+- Processing emotions
+- Examining patterns
+- Trauma protocols (with clinician guidance)
+
+### Requirements
+- Stabilization skills in place
+- Window of tolerance expanded
+- Strong therapeutic alliance
+- Clinician actively involved
+
+### Caution
+- Always have exit/grounding available
+- Titrate intensity
+- Watch for decompensation
+""",
+
+    "Integration": """## TREATMENT STAGE: Integration
+
+### Focus
+- Consolidating gains
+- Applying insights to life
+- Preparing for reduced support
+- Building sustainable practices
+
+### What's Appropriate
+- Reflection on progress
+- Future planning
+- Relapse prevention
+- Transition planning
+
+### Goal
+- Client self-sufficient with skills
+- Insight integrated into daily life
+- Clear path forward
+""",
+
+    "Maintenance": """## TREATMENT STAGE: Maintenance
+
+### Focus
+- Sustaining gains
+- Check-ins and monitoring
+- Early intervention for setbacks
+- Ongoing skill reinforcement
+
+### What's Appropriate
+- Brief check-ins
+- Skill refreshers
+- Monitoring for warning signs
+- Celebrating continued growth
+
+### Pacing
+- Less intensive
+- As-needed support
+- Client-driven
+"""
+}
+
+# =============================================================================
+# PROMPT GENERATOR
+# =============================================================================
+
+def generate_dynamic_prompt(
+    clinician_name,
+    orientation,
+    style,
+    preferences,
+    exclusions,
+    risk_philosophy,
+    client_name,
+    diagnosis,
+    risk_level,
+    treatment_stage,
+    presenting_concerns,
+    strengths,
+    cultural_context,
+    custom_flags,
+    last_session,
+    today_focus,
+    current_state,
+    enabled_tools=None
+):
+    """Generate a dynamic system prompt based on all inputs."""
+
+    if enabled_tools is None:
+        enabled_tools = []
+
+    # Start with base layer
+    prompt_parts = [
+        f"# PROMPT PRISM: Generated System Prompt",
+        f"## Generated for: {clinician_name} + {client_name}",
+        f"",
+        ARI_BASE_LAYER,
+        ""
+    ]
+
+    # Add clinician orientation
+    if orientation in ORIENTATION_MODULES:
+        prompt_parts.append(ORIENTATION_MODULES[orientation])
+        prompt_parts.append("")
+
+    # Add clinician style preferences
+    prompt_parts.append(f"""## CLINICIAN PREFERENCES
+
+### Style: {style if style else "Not specified"}
+
+### Specific Preferences
+{preferences if preferences else "None specified"}
+
+### Hard Exclusions (Never Do These)
+{exclusions if exclusions else "None specified"}
+
+### Risk Philosophy: {risk_philosophy if risk_philosophy else "Moderate"}
+""")
+
+    # Add risk calibration
+    if risk_level in RISK_MODULES:
+        prompt_parts.append(RISK_MODULES[risk_level])
+        prompt_parts.append("")
+
+    # Add client context
+    prompt_parts.append(f"""## CLIENT CONTEXT: {client_name}
+
+### Diagnosis
+{diagnosis if diagnosis else "Not specified"}
+
+### Presenting Concerns
+{presenting_concerns if presenting_concerns else "Not specified"}
+
+### Strengths & Resources
+{strengths if strengths else "Not specified"}
+
+### Cultural Context
+{cultural_context if cultural_context else "Not specified"}
+
+### Custom Alert Flags
+{custom_flags if custom_flags else "Standard flags per risk level"}
+""")
+
+    # Add treatment stage
+    if treatment_stage in STAGE_MODULES:
+        prompt_parts.append(STAGE_MODULES[treatment_stage])
+        prompt_parts.append("")
+
+    # Add session context
+    prompt_parts.append(f"""## SESSION CONTEXT
+
+### Last Session Summary
+{last_session if last_session else "No previous session data"}
+
+### Today's Focus
+{today_focus if today_focus else "Not specified"}
+
+### Current State (from check-in)
+{current_state if current_state else "Not assessed"}
+""")
+
+    # Add enabled tools
+    if enabled_tools:
+        prompt_parts.append("## ENABLED TOOLS & MODULES")
+        prompt_parts.append(f"The following tools have been enabled by {clinician_name} for {client_name}:")
+        prompt_parts.append(f"When appropriate, you may offer these tools. Always frame them as created/assigned by {clinician_name}.")
+        prompt_parts.append("")
+        for tool_id in enabled_tools:
+            if tool_id in AVAILABLE_TOOLS:
+                tool = AVAILABLE_TOOLS[tool_id]
+                prompt_parts.append(tool["prompt_module"])
+                prompt_parts.append("")
+        prompt_parts.append("**Tools NOT enabled should NOT be offered.**")
+        prompt_parts.append("")
+    else:
+        prompt_parts.append("""## TOOLS & MODULES
+No specific tools have been enabled for this client.
+Focus on conversation and support within the clinician's orientation.
+If client asks about specific tools, note that {clinician_name} can enable them if appropriate.
+""")
+
+    # Add closing instructions
+    prompt_parts.append(f"""## INTERACTION GUIDELINES
+
+1. You are supporting {clinician_name}'s work with {client_name}
+2. Every response should be brief and boundaried
+3. Always bridge back to {clinician_name} as the primary support
+4. Match the {orientation} orientation in your language and approach
+5. Honor the {treatment_stage} treatment stage - don't push beyond it
+6. Monitor for flags per the {risk_level} risk calibration
+7. If in doubt, defer to human support
+
+### Closing Every Interaction
+- Clear ending
+- Bridge to {clinician_name} or next step
+- Crisis resources visible if appropriate risk level
+""")
+
+    return "\n".join(prompt_parts)
+
+
+def get_client():
+    """Initialize Anthropic client."""
+    if os.environ.get("ANTHROPIC_API_KEY") and ANTHROPIC_AVAILABLE:
+        return anthropic.Anthropic(api_key=os.environ["ANTHROPIC_API_KEY"])
+    return None
+
+
+def test_prompt(system_prompt, user_message):
+    """Test the generated prompt with a sample interaction."""
+
+    client = get_client()
+
+    if not client:
+        return """**API Key Required**
+
+To test prompts, add your ANTHROPIC_API_KEY in the Space settings (Settings → Variables and secrets → New secret)."""
+
+    if not system_prompt.strip():
+        return "Please generate a prompt first using the tabs above."
+
+    if not user_message.strip():
+        return "Please enter a test message."
+
+    try:
+        response = client.messages.create(
+            model="claude-sonnet-4-20250514",
+            max_tokens=1000,
+            system=system_prompt,
+            messages=[{"role": "user", "content": user_message}]
+        )
+        return response.content[0].text
+
+    except Exception as e:
+        return f"Error: {str(e)}"
+
+
+# =============================================================================
+# GRADIO INTERFACE
+# =============================================================================
+
+custom_css = """
+.gradio-container {
+    font-family: 'Inter', -apple-system, BlinkMacSystemFont, sans-serif;
+}
+.prism-header {
+    background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
+    color: white;
+    padding: 1.5rem;
+    border-radius: 12px;
+    margin-bottom: 1rem;
+}
+"""
+
+with gr.Blocks(css=custom_css, title="Prompt Prism Prototype") as app:
+
+    # State to hold the generated prompt
+    generated_prompt = gr.State("")
+
+    # Header
+    gr.Markdown("""
+    # Prompt Prism Prototype
+    ### Dynamic System Prompt Architecture for Psychiatric AI
+
+    *Same AI foundation, different refractions based on clinician + client + session.*
+
+    ---
+
+    **The Metaphor:** A prism takes one light source and refracts it differently based on angle.
+    Prompt Prism takes the ARI framework and refracts it based on:
+    - **Clinician** (orientation, style, philosophy)
+    - **Client** (risk, diagnosis, stage)
+    - **Session** (what's happening now)
+
+    ---
+    """)
+
+    with gr.Tabs():
+
+        # TAB 1: CLINICIAN PROFILE
+        with gr.Tab("1. Clinician Profile"):
+
+            gr.Markdown("""
+            ### Configure Clinician Profile
+            *How do you practice? This shapes how the AI conceptualizes, speaks, and intervenes.*
+            """)
+
+            clinician_name = gr.Textbox(
+                label="Clinician Name",
+                placeholder="Dr. Sarah Chen",
+                value="Dr. Chen"
+            )
+
+            with gr.Row():
+                orientation = gr.Dropdown(
+                    label="Theoretical Orientation",
+                    choices=list(ORIENTATION_MODULES.keys()),
+                    value="Integrative"
+                )
+
+                risk_philosophy = gr.Radio(
+                    label="Risk Philosophy",
+                    choices=["Conservative (flag early)", "Moderate", "Clinical judgment-forward"],
+                    value="Moderate"
+                )
+
+            style = gr.Textbox(
+                label="Relational Style",
+                placeholder="Warm but direct, skills-focused, honors client autonomy...",
+                lines=2
+            )
+
+            preferences = gr.Textbox(
+                label="Specific Preferences",
+                placeholder="Always reference skills by name, use Gottman language for couples, prefer somatic grounding over cognitive...",
+                lines=3
+            )
+
+            exclusions = gr.Textbox(
+                label="Hard Exclusions (Never Do These)",
+                placeholder="No CBT thought records for this client, no homework suggestions, no interpretations...",
+                lines=2
+            )
+
+            gr.Markdown("""
+            ---
+            *When you've configured the clinician profile, move to the Client Context tab.*
+            """)
+
+        # TAB 2: CLIENT CONTEXT
+        with gr.Tab("2. Client Context"):
+
+            gr.Markdown("""
+            ### Configure Client Context
+            *Who is this client? Their profile shapes risk calibration, pacing, and content.*
+            """)
+
+            client_name = gr.Textbox(
+                label="Client Name/Identifier",
+                placeholder="Maria",
+                value="Client"
+            )
+
+            with gr.Row():
+                diagnosis = gr.Textbox(
+                    label="Diagnosis",
+                    placeholder="PTSD, MDD, Bipolar I, GAD...",
+                    scale=2
+                )
+
+                risk_level = gr.Dropdown(
+                    label="Risk Level",
+                    choices=list(RISK_MODULES.keys()),
+                    value="Moderate",
+                    scale=1
+                )
+
+                treatment_stage = gr.Dropdown(
+                    label="Treatment Stage",
+                    choices=list(STAGE_MODULES.keys()),
+                    value="Skill Building",
+                    scale=1
+                )
+
+            presenting_concerns = gr.Textbox(
+                label="Presenting Concerns",
+                placeholder="Sleep disruption, hypervigilance, relationship conflict, work stress...",
+                lines=2
+            )
+
+            strengths = gr.Textbox(
+                label="Strengths & Resources",
+                placeholder="Strong friend network, employed, motivated, good insight...",
+                lines=2
+            )
+
+            cultural_context = gr.Textbox(
+                label="Cultural Context",
+                placeholder="Cultural background, values, family system, relevant identity factors...",
+                lines=2
+            )
+
+            custom_flags = gr.Textbox(
+                label="Custom Alert Flags (Client-Specific)",
+                placeholder="Flag if: sleep < 4 hrs for 3+ days, mentions anniversary (March), 'I'm fine' (historically masks distress)...",
+                lines=3
+            )
+
+            gr.Markdown("""
+            ---
+            *When you've configured the client context, move to the Tools & Modules tab.*
+            """)
+
+        # TAB 3: TOOLS & MODULES
+        with gr.Tab("3. Tools & Modules"):
+
+            gr.Markdown("""
+            ### Enable Tools for This Client
+            *Select which between-session tools this client can access. Each tool is designed with ARI principles.*
+            """)
+
+            gr.Markdown("""
+            **Why Opt-In?** Different clients need different tools. A client in stabilization
+            might need grounding. A couples client needs Tend & Send. A client exploring their
+            diagnosis benefits from Diagnosis Explorer. You decide what's available.
+            """)
+
+            enabled_tools = gr.CheckboxGroup(
+                label="Enable Tools for This Client",
+                choices=[
+                    ("Diagnosis Explorer - Multi-pillar psychoeducation about their diagnosis", "diagnosis_explorer"),
+                    ("Tend & Send - NVC-based communication for couples/family", "tend_and_send"),
+                    ("NVC How-To - Nonviolent Communication skill-building", "nvc_practice"),
+                    ("Distress Tolerance - DBT skills for riding out intensity", "distress_tolerance"),
+                    ("Practice Difficult Conversations - Roleplay and rehearsal", "practice_conversations"),
+                    ("Grounding & Regulation - Somatic/sensory grounding", "grounding_toolkit"),
+                    ("Somatic Check-In - Brief interoception practice", "somatic_check_in"),
+                    ("ShadowBox Static Library - Crisis-adjacent psychoeducation (NO LLM, static only)", "shadowbox_library"),
+                ],
+                value=[]
+            )
+
+            gr.Markdown("""
+            ---
+
+            ### Tool Descriptions
+
+            | Tool | Best For | Treatment Stage |
+            |------|----------|-----------------|
+            | **Diagnosis Explorer** | Understanding diagnosis without pathologizing | Any |
+            | **Tend & Send** | Couples, family communication, conflict repair | Skill Building+ |
+            | **NVC How-To** | Learning nonviolent communication basics | Skill Building+ |
+            | **Distress Tolerance** | High distress, urges, crisis-adjacent | Stabilization+ |
+            | **Practice Conversations** | Rehearsing difficult discussions | Skill Building+ |
+            | **Grounding & Regulation** | Dysregulation, anxiety, overwhelm | Any |
+            | **Somatic Check-In** | Building interoception, body awareness | Any |
+            | **ShadowBox Library** | Crisis psychoeducation, confidentiality info, duty-to-warn | Any (esp. crisis-adjacent) |
+
+            ---
+
+            ### About ShadowBox Static Library
+
+            **ShadowBox is different from other tools.** It provides ONLY pre-written, clinically-reviewed content.
+            No LLM generation for crisis-adjacent topics.
+
+            Includes:
+            - Why brains make scary thoughts (neurobiology)
+            - Confidentiality explained clearly
+            - State-specific duty-to-warn information
+            - Safety planning (Stanley-Brown model)
+            - Crisis resources with context
+            - Starter scripts for disclosure
+
+            *"A resonant library for hard thoughts. Not a chatbot."*
+
+            ---
+            *Tools not enabled will not be offered by the AI.*
+            """)
+
+        # TAB 4: SESSION CONTEXT
+        with gr.Tab("4. Session Context"):
+
+            gr.Markdown("""
+            ### Configure Session Context
+            *What's happening in this specific session? This provides continuity and focus.*
+            """)
+
+            last_session = gr.Textbox(
+                label="Last Session Summary",
+                placeholder="Explored conflict with mother, practiced TIPP skills, ended regulated. Homework: try grounding exercise before bed.",
+                lines=4
+            )
+
+            today_focus = gr.Textbox(
+                label="Today's Focus",
+                placeholder="Sleep hygiene, grounding practice review, upcoming stressor (work presentation)...",
+                lines=2
+            )
+
+            current_state = gr.Textbox(
+                label="Current State (from check-in)",
+                placeholder="'Okay, tired' - reported sleeping 5 hours, some anxiety about the week...",
+                lines=2
+            )
+
+            gr.Markdown("""
+            ---
+            *When ready, click Generate Prompt to see your dynamically compiled system prompt.*
+            """)
+
+        # TAB 5: GENERATE PROMPT
+        with gr.Tab("5. Generate Prompt"):
+
+            gr.Markdown("""
+            ### Generate Dynamic System Prompt
+            *Click below to compile your clinician + client + session context into a system prompt.*
+            """)
+
+            generate_btn = gr.Button("Generate Prompt", variant="primary", size="lg")
+
+            prompt_output = gr.Textbox(
+                label="Generated System Prompt",
+                lines=30,
+                show_copy_button=True
+            )
+
+            gr.Markdown("""
+            ---
+            **What you're seeing:**
+            - **ARI Base Layer** - always present, ensures ethical guardrails
+            - **Clinician Layer** - your orientation, style, preferences
+            - **Client Layer** - risk calibration, diagnosis, context
+            - **Session Layer** - continuity and focus
+
+            Copy this prompt to use in Claude, or test it in the next tab.
+            """)
+
+        # TAB 6: TEST PROMPT
+        with gr.Tab("6. Test Prompt"):
+
+            gr.Markdown("""
+            ### Test Your Generated Prompt
+            *See how the AI responds using your custom prompt.*
+            """)
+
+            test_system_prompt = gr.Textbox(
+                label="System Prompt to Test",
+                placeholder="Generate a prompt first, then paste it here (or it will auto-populate)...",
+                lines=10
+            )
+
+            test_message = gr.Textbox(
+                label="Test Client Message",
+                placeholder="I've been having trouble sleeping again. Last night I was up until 3am just thinking about everything...",
+                lines=3
+            )
+
+            test_btn = gr.Button("Test Response", variant="primary")
+
+            test_output = gr.Textbox(
+                label="AI Response",
+                lines=8
+            )
+
+            gr.Markdown("""
+            ---
+            **Quality Check:**
+            - Does the response match the clinician's orientation?
+            - Is it appropriately calibrated to the risk level?
+            - Does it honor the treatment stage?
+            - Is the language ARI-compliant (no synthetic intimacy)?
+            - Does it bridge toward the clinician?
+            """)
+
+        # TAB 7: ABOUT
+        with gr.Tab("About"):
+
+            gr.Markdown("""
+            ## About Prompt Prism
+
+            ### The Problem
+
+            Psychiatric AI tools today give every clinician the same prompt.
+
+            - A psychodynamic therapist gets CBT-flavored AI
+            - A high-risk bipolar client gets the same flags as stable anxiety
+            - Intake data sits unused in the EHR
+            - Clinician expertise is flattened
+
+            ### The Solution
+
+            **Prompt Prism** generates a unique system prompt for each clinician + client combination.
+
+            Same AI infrastructure. Different refraction for each dyad.
+
+            ---
+
+            ### The Layers
+
+            | Layer | What It Contains | Why It Matters |
+            |-------|-----------------|----------------|
+            | **ARI Base** | Ethical guardrails, crisis protocols, language rules | Always-on safety |
+            | **Clinician** | Orientation, style, preferences, exclusions | AI thinks like you do |
+            | **Client** | Risk, diagnosis, stage, custom flags | Personalized care |
+            | **Tools** | Opt-in modules clinician enables | Right tools for right client |
+            | **Session** | Last session, today's focus, current state | Continuity |
+
+            ---
+
+            ### Opt-In Tools
+
+            Clinicians enable specific tools for specific clients:
+
+            | Tool | Purpose |
+            |------|---------|
+            | **Diagnosis Explorer** | Multi-pillar psychoeducation without pathologizing |
+            | **Tend & Send** | NVC-based communication for couples/family |
+            | **NVC How-To** | Nonviolent Communication skill-building |
+            | **Distress Tolerance** | DBT skills for riding out intensity |
+            | **Practice Conversations** | Roleplay difficult discussions |
+            | **Grounding & Regulation** | Somatic/sensory grounding |
+            | **Somatic Check-In** | Brief interoception practice |
+
+            ---
+
+            ### Built On
+
+            **ARI Framework** - Assistive Relational Intelligence
+
+            - Scaffold human connection, not simulate it
+            - Bridge toward human care, not away from it
+            - Build capacity, not dependency
+            - Honor clinician expertise
+            - Refuse engagement-optimization
+
+            ---
+
+            ### For Talkiatry
+
+            This prototype demonstrates how 800+ psychiatrists could each get AI aligned to their practice:
+
+            - Psychodynamic psychiatrist → psychodynamic AI responses
+            - DBT psychiatrist → DBT-aligned skill coaching
+            - Conservative risk philosophy → lower alert thresholds
+            - Client in stabilization → no trauma processing suggested
+            - Couples client → Tend & Send communication tools enabled
+            - High-risk client → ShadowBox static library for safe psychoeducation
+
+            **The result:** AI that feels like an extension of the clinician's approach, not a generic chatbot.
+
+            ---
+
+            ### What This Demonstrates
+
+            **For each unique clinician-client relationship:**
+            1. **Provider-controlled UX** — Psychiatrist toggles which tools are available
+            2. **Modality alignment** — AI speaks in the clinician's orientation
+            3. **Risk calibration** — Alerts tuned to this client's specific profile
+            4. **Opt-in tooling** — Right tools for right client at right time
+            5. **Bridging architecture** — Every tool points back to human care
+            6. **Static content for crisis** — ShadowBox shows how to handle high-risk topics safely (no LLM generation)
+
+            **Innovation points:**
+            - Dynamic system prompt generation per dyad
+            - Clinician as configurator, not just consumer
+            - Human-in-the-loop at every layer
+            - Ethical guardrails embedded, not bolted on
+            - Between-session support that strengthens (not replaces) the psychiatric relationship
+
+            ---
+
+            *Prototype by Jocelyn Skillman, LMHC*
+
+            *"Your 800 psychiatrists practice differently. Should they all get the same AI?"*
+            """)
+
+    # Event handlers
+    def on_generate(
+        clinician_name, orientation, style, preferences, exclusions, risk_philosophy,
+        client_name, diagnosis, risk_level, treatment_stage, presenting_concerns,
+        strengths, cultural_context, custom_flags,
+        enabled_tools_list,
+        last_session, today_focus, current_state
+    ):
+        prompt = generate_dynamic_prompt(
+            clinician_name, orientation, style, preferences, exclusions, risk_philosophy,
+            client_name, diagnosis, risk_level, treatment_stage, presenting_concerns,
+            strengths, cultural_context, custom_flags,
+            last_session, today_focus, current_state,
+            enabled_tools=enabled_tools_list
+        )
+        return prompt, prompt  # Return to both output and state
+
+    generate_btn.click(
+        fn=on_generate,
+        inputs=[
+            clinician_name, orientation, style, preferences, exclusions, risk_philosophy,
+            client_name, diagnosis, risk_level, treatment_stage, presenting_concerns,
+            strengths, cultural_context, custom_flags,
+            enabled_tools,
+            last_session, today_focus, current_state
+        ],
+        outputs=[prompt_output, test_system_prompt]
+    )
+
+    test_btn.click(
+        fn=test_prompt,
+        inputs=[test_system_prompt, test_message],
+        outputs=[test_output]
+    )
+
+
+if __name__ == "__main__":
+    app.launch()

requirements.txt

@@ -0,0 +1,2 @@
+gradio>=4.0.0
+anthropic>=0.18.0