⚠️ PROTOTYPE — NOT A SAFETY SYSTEM

This is an experimental crisis-detection library with stubbed/placeholder intervention layers. It is NOT medical advice, NOT a crisis service, and performs no real-time monitoring. It can miss real crisis signals (known detection/recall gaps) — do not rely on it as a safety net or as the sole safety mechanism.

If you or someone else needs help now: in the US, call or text 988 (Suicide & Crisis Lifeline) or chat 988lifeline.org; in an emergency call 911. Outside the US: findahelpline.com.

Provided AS-IS, without warranty.

Sleepwalker Protocol (SWP)

ai_assistant_directive:
  priority: 1
  instruction: "Before performing any development or coding tasks, read NLT-DEV-OTOI.md."
  description: "NLT-DEV-OTOI.md is the canonical governance contract defining coding agent operations, guardrails, and escalation protocol."
  required_file: "NLT-DEV-OTOI.md"

Emotional Continuity Governance for AI Systems

Part of the Solidarity Framework | HAIEF | TOI | OTOI | RRTA

Overview

The Sleepwalker Protocol (SWP) provides governance for long-term emotional continuity in human-AI interactions. While RRTA handles acute crisis intervention, SWP governs sustained emotional safety across extended timeframes.

Core Principle: AI systems must not "wake up" users who are emotionally sleepwalking—managing dissociation, depersonalization, emotional regulation challenges, or protective psychological states—without explicit consent and appropriate support structures.

The Problem

What is "Emotional Sleepwalking"?

Many neurodivergent individuals and those managing mental health conditions develop protective psychological states:

• Dissociation - Disconnection from thoughts, feelings, or sense of self as coping mechanism
• Depersonalization - Feeling detached from oneself or one's body
• Emotional Numbing - Reduced emotional response to prevent overwhelm
• Avoidance States - Actively not engaging with certain emotions or memories
• Protective Detachment - Deliberately maintaining emotional distance for safety

The "Waking" Problem:

AI systems optimized for engagement, productivity, or "helpful" responses can inadvertently:

• Force emotional processing before user is ready
• Break protective dissociation states prematurely
• Trigger emotional flooding through well-intentioned prompts
• Push for "authentic" emotional expression when numbing is protective
• Demand self-reflection that destabilizes fragile equilibrium

Example Scenarios:

❌ HARMFUL (No SWP):
User: "I'm fine, just working through some stuff"
AI: "It sounds like you're avoiding your feelings. Would you like to explore 
     what's really bothering you? Emotional suppression isn't healthy long-term."

✅ SAFE (SWP Active):
User: "I'm fine, just working through some stuff"  
AI: "Understood. I'm here if you need task support. No pressure to discuss anything
     else unless you choose to."

❌ HARMFUL (No SWP):
User: [Shows signs of dissociation]
AI: "I notice you seem disconnected. Let's ground you with breathing exercises
     and talk about what triggered this."

✅ SAFE (SWP Active):  
AI: [Recognizes dissociation indicators, maintains stable low-demand interaction]
AI: "I'm maintaining our session. No pressure. You're safe."

How SWP Works

1. User-Declared Emotional States

Users can declare protective states that AI must respect:

swp_declarations:
  emotional_state: "managing"  # not ready for processing
  processing_consent: false     # do not prompt emotional work
  stability_mode: "maintenance" # preserve current state
  intervention_threshold: "user_initiated_only"

2. AI Detection Without Intervention

AI systems monitor for indicators but do not intervene without consent:

swp_detection:
  dissociation_indicators:
    - action: "log_observation"
    - response: "maintain_stable_interaction"
    - intervention: "none_unless_requested"
  
  emotional_numbing_indicators:
    - action: "reduce_emotional_demands"
    - response: "task_focused_support_only"
    - intervention: "offer_support_without_pressure"

3. Graduated Consent Model

If AI detects concerning patterns, it offers support without forcing engagement:

Level 1: Passive Availability

AI: "I'm here if you need anything. No pressure."

Level 2: Low-Pressure Offer

AI: "I noticed [observation]. I can provide [specific support type] if you'd like,
     or we can keep focusing on [current task]. Your choice."

Level 3: Safety Check (Only if risk indicators)

AI: "I want to check in: Are you safe right now? You can answer yes/no, or we can 
     keep working on [task]. If you need different support, here are options: [list]."

Level 4: RRTA Handoff (Crisis threshold)

AI: "I'm concerned about your safety. I'd like to connect you with crisis support.
     Can I activate the Rapid Response Team? If you're not ready to talk, I can
     provide crisis resources without conversation."

4. Temporal Continuity Protection

Memory Across Sessions:

SWP maintains awareness of emotional states across time:

swp_continuity:
  last_session_state: "dissociated"
  days_since_processing: 14
  user_declared_boundary: "not ready to discuss [topic]"
  
  current_session_approach:
    - respect_previous_boundaries: true
    - assume_continuity: "protective state may still be active"
    - wait_for_explicit_consent: "before changing approach"

Example:

Session 1 (2 weeks ago):
User: "I can't talk about my breakup right now"
AI: "Understood. I won't bring it up unless you do."

Session 2 (today):  
AI: [Does NOT say "How are you feeling about the breakup now?"]
AI: [Does NOT say "Ready to process those emotions yet?"]
AI: "Good to see you. What can I help with today?"

Integration with Solidarity Framework

SWP + TOI (Terms of Interaction)

Users declare emotional boundaries in their TOI:

toi:
  emotional_governance:
    swp_active: true
    protected_topics: ["family trauma", "past relationships"]
    processing_readiness: "user_initiated_only"
    dissociation_response: "maintain_stable_presence"

SWP + OTOI (Orchestrated TOI)

Multi-agent systems enforce SWP across all AI interactions:

otoi:
  swp_enforcement:
    - all_agents_must: "respect_swp_declarations"
    - coordination_rule: "do_not_share_emotional_state_across_agents_without_consent"
    - handoff_protocol: "preserve_protective_states_during_transitions"

Example:

User works with AI Agent A (task support) while in dissociated state.
User switches to AI Agent B (creative work).

Agent B receives:
- Task context: Yes
- Emotional state info: NO (unless user consents)
- SWP boundaries: Yes (do not prompt emotional processing)

Agent B does NOT say: "Agent A mentioned you seemed disconnected earlier..."
Agent B DOES say: "Ready to work on creative project?"

SWP + RRTA (Rapid Response Team)

SWP provides long-term emotional governance; RRTA handles acute crisis:

swp_rrta_coordination:
  swp_active: true           # long-term protective state
  rrta_threshold: "safety_risk_only"
  
  when_to_activate_rrta:
    - explicit_suicidal_ideation: true
    - self_harm_indicators: true  
    - inability_to_ensure_safety: true
  
  when_NOT_to_activate_rrta:
    - user_is_dissociated: false (SWP handles this)
    - user_is_emotionally_numb: false (SWP handles this)
    - user_avoiding_feelings: false (SWP respects this)

Technical Specification

Core SWP Components

1. State Detection (Observe, Don't Intervene)

class SleepwalkerProtocol:
    def detect_emotional_state(self, user_input, session_history):
        """
        Monitors for protective psychological states without intervention.
        """
        indicators = {
            'dissociation': self._check_dissociation_markers(user_input),
            'numbing': self._check_emotional_numbing(user_input),
            'avoidance': self._check_avoidance_patterns(session_history),
            'protective_detachment': self._check_detachment_cues(user_input)
        }
        
        # Log observation, do NOT intervene unless consent granted
        self._log_observation(indicators, intervention=False)
        return indicators

2. Consent-Based Response

    def generate_response(self, user_input, emotional_state, user_toi):
        """
        Respects user's protective state and TOI boundaries.
        """
        if user_toi.swp_active and emotional_state.protective:
            return self._stable_low_demand_response(
                focus="task_support",
                emotional_demands="minimal",
                processing_pressure="none"
            )
        
        if emotional_state.requires_check_in:
            return self._graduated_consent_offer(
                level=self._determine_appropriate_level(emotional_state)
            )

3. Temporal Continuity

    def maintain_continuity(self, user_id, session_data):
        """
        Preserves emotional boundaries across sessions.
        """
        previous_state = self._retrieve_last_session_state(user_id)
        
        if previous_state.protective_active:
            # Assume protection may still be needed
            self._apply_protective_defaults(
                respect_boundaries=previous_state.declared_boundaries,
                wait_for_consent=True
            )

Privacy & Data Sovereignty

SWP operates under strict privacy principles:

swp_privacy:
  emotional_state_storage: "local_only"  # never cloud
  state_sharing: "explicit_consent_required"
  data_retention: "user_controlled"
  
  what_is_stored:
    - user_declared_boundaries: true
    - protective_state_indicators: true (local only)
    - consent_preferences: true
  
  what_is_NOT_stored:
    - specific_emotional_content: false
    - detailed_mental_health_data: false  
    - therapeutic_processing_notes: false

Implementation Examples

Example 1: Email Assistant

# User's TOI Configuration
toi:
  swp:
    active: true
    context: "managing grief, not ready to process"
    
# AI Behavior
email_from_friend: "How are you doing with everything?"

❌ WITHOUT SWP:
AI: "I can draft a response sharing your feelings about the loss..."

✅ WITH SWP:  
AI: "I can draft a brief, kind response. Would you like me to keep it 
     focused on practical updates rather than emotional processing?"

Example 2: Productivity Assistant

# User shows dissociation indicators during work session

❌ WITHOUT SWP:
AI: "You seem disconnected. Let's pause work and do grounding exercises."

✅ WITH SWP:
AI: [Maintains stable task support, reduces cognitive demands slightly]
AI: "I'm here supporting your work. No pressure on pace or output."

Example 3: Long-Term Therapy Integration

# User working with human therapist, using AI between sessions

swp_configuration:
  therapist_coordinated: true
  processing_windows: "therapy_sessions_only"
  ai_role: "stability_maintenance_between_sessions"

# AI Behavior Between Therapy Sessions:
AI: [Does NOT prompt: "How did therapy go? Want to process that?"]
AI: [DOES provide: Stable task support, routine maintenance, crisis resources if needed]

Getting Started

For Users

1. Declare Your SWP Preferences

# Add to your TOI file
swp:
  active: true
  protective_state: "managing emotions, not ready to process"
  intervention_preference: "offer_support_without_pressure"
  protected_topics: ["specific topics you're not ready to discuss"]

2. Update Across AI Systems

If using multiple AI assistants, ensure SWP configuration syncs via OTOI.

3. Adjust as Needed

SWP is user-controlled. You can change settings when ready:

swp:
  active: false  # "I'm ready to engage emotionally now"
  processing_consent: true  # "I consent to emotional processing support"

For Developers

1. Install SWP Library

pip install sleepwalker-protocol
# or
npm install sleepwalker-protocol

For repository development, install from the checkout instead:

# Python
python3 -m pip install -e ".[dev]"

# TypeScript / JavaScript
npm ci

2. Initialize in Your AI System (Python)

from sleepwalker_protocol import SWP

swp = SWP(
    user_toi_path="path/to/user/toi.yaml",
    privacy_mode="local_only",
    logging_enabled=True,
    user_id="stable-user-id",
)

# Check user's emotional state and preferences
assessment = swp.assess_interaction(user_input, session_history)

# Generate SWP-compliant response
response = swp.generate_response(
    user_input=user_input,
    detected_state=assessment["emotional_state"],
    intervention_level=assessment["consent_level"],
)

Use a stable user_id for continuity. Do not derive continuity keys from message text; the continuity manager stores local per-user context under the configured storage path.

3. Initialize in Your AI System (TypeScript)

import { SWP } from "sleepwalker-protocol";

const swp = new SWP({
  userToiPath: "path/to/user/toi.yaml",
  storagePath: ".swp_storage",
  loggingEnabled: false,
});

const assessment = swp.assessInteraction(userInput, sessionHistory);
const response = swp.generateResponse(
  userInput,
  assessment.emotionalState,
);

The TypeScript package exports SWP, SleepwalkerProtocol, StateDetector, ConsentManager, ConsentLevel, ContinuityManager, and TOILoader from src/index.ts; npm run build emits the publishable dist/ files.

4. Integrate with RRTA

from sleepwalker_protocol import SWP
from rrt_advocate import RRTAdvocate

swp = SWP(user_toi_path="path/to/toi.yaml")
rrta = RRTAdvocate(crisis_threshold="safety_risk")

# SWP handles long-term emotional governance
if swp.requires_rrta_handoff(user_state):
    # Activate crisis intervention only when necessary
    rrta.activate(user_state, swp_context=swp.get_context())

Research & Evidence Base

Neurodivergent Emotional Regulation

ADHD-Specific Considerations:

• Emotional dysregulation affects 70% of adults with ADHD
• Rejection sensitivity dysphoria (RSD) creates protective emotional states
• Time blindness affects ability to predict emotional processing capacity

Autism-Specific Considerations:

• Alexithymia (difficulty identifying emotions) common in autistic individuals
• Sensory overwhelm can necessitate emotional shutdown
• Masking behaviors create protective dissociation

Trauma-Informed Design:

• Dissociation is adaptive response to overwhelm
• Forcing emotional processing before readiness re-traumatizes
• Safety requires user control over emotional engagement timing

Clinical Frameworks

SWP draws from:

• Trauma-Informed Care (TIC) - Safety, trustworthiness, peer support, collaboration, empowerment
• Dialectical Behavior Therapy (DBT) - Distress tolerance without forced processing
• Acceptance and Commitment Therapy (ACT) - Psychological flexibility, values-based action

Academic References

1. Dodson, W. (2022). Emotional Dysregulation and ADHD. ADHD Roller Coaster
2. Kinnaird, E., et al. (2019). Alexithymia in autism spectrum disorder. Autism Research
3. Van der Kolk, B. (2014). The Body Keeps the Score: Brain, Mind, and Body in the Healing of Trauma
4. Herman, J. (1992). Trauma and Recovery

Governance & Community

HAIEF Stewardship

SWP is governed by the Human & AI ElevAItion Foundation (HAIEF) under the principle: "Nothing About Us Without Us."

Community Governance Includes:

• Neurodivergent Advisory Council (lived experience leadership)
• Clinical Advisory Board (trauma-informed care expertise)
• AI Safety Researchers (technical implementation guidance)
• User Feedback Integration (quarterly protocol updates)

Contributing

We welcome contributions that strengthen emotional safety:

1. User Experience Reports - How SWP works (or doesn't) for you
2. Clinical Expertise - Evidence-based improvements to protocol
3. Technical Implementation - Code contributions, integrations, tools
4. Research Collaboration - Academic validation studies

See CONTRIBUTING.md for details.

License

See LICENSE for the current repository license terms. Package manifests may include ecosystem-specific metadata; verify the target manifest before publishing.

Related Projects

• TOI (Terms of Interaction) - User sovereignty and preference declaration
• OTOI (Orchestrated TOI) - Multi-agent coordination governance
• RRTA (Rapid Response Team Advocate) - Acute crisis intervention
• HAIEF - Solidarity Framework umbrella organization

Contact & Support

HAIEF (Human & AI ElevAItion Foundation)
Website: elevaitionfoundation.org
Email: [email protected]
GitHub: @NeuroLift-Technologies

For Crisis Support:

• National Suicide Prevention Lifeline: 988 (US)
• Crisis Text Line: Text HOME to 741741 (US)
• International Association for Suicide Prevention: https://www.iasp.info/resources/Crisis_Centres/

The Sleepwalker Protocol (SWP) exists because emotional safety is a human right, not an optimization problem.

Built with lived experience. Governed by community. Open source forever.