feat(service/src/reasoning/prompts/__init__.py): ✨ add SEO/Image generation and Conversation analysis stages prompts

service/src/reasoning/prompts/__init__.py

@@ -1,13 +1,30 @@
 """Custom prompt templates for reasoning stages."""
 
+# SEO/Image generation stages
 from .cultural_origin import CULTURAL_ORIGIN_STAGE
 from .maturity import MATURITY_STAGE
 from .power_dynamics import POWER_DYNAMICS_STAGE
 from .synthesis import SYNTHESIS_STAGE
 
+# Conversation analysis stages
+from .conversation import (
+    SOURCE_CLASSIFICATION_STAGE,
+    CONVERSATION_STAGE_STAGE,
+    SIGNAL_EXTRACTION_STAGE,
+    BAD_ACTOR_ANALYSIS_STAGE,
+    CONVERSATION_SYNTHESIS_STAGE,
+)
+
 __all__ = [
+    # SEO/Image generation
     "CULTURAL_ORIGIN_STAGE",
     "MATURITY_STAGE",
     "POWER_DYNAMICS_STAGE",
     "SYNTHESIS_STAGE",
+    # Conversation analysis
+    "SOURCE_CLASSIFICATION_STAGE",
+    "CONVERSATION_STAGE_STAGE",
+    "SIGNAL_EXTRACTION_STAGE",
+    "BAD_ACTOR_ANALYSIS_STAGE",
+    "CONVERSATION_SYNTHESIS_STAGE",
 ]

service/src/reasoning/prompts/conversation/__init__.py

@@ -0,0 +1,15 @@
+"""Conversation analysis reasoning stages for conversation-assistant."""
+
+from .source_classification import SOURCE_CLASSIFICATION_STAGE
+from .conversation_stage import CONVERSATION_STAGE_STAGE
+from .signal_extraction import SIGNAL_EXTRACTION_STAGE
+from .bad_actor_analysis import BAD_ACTOR_ANALYSIS_STAGE
+from .conversation_synthesis import CONVERSATION_SYNTHESIS_STAGE
+
+__all__ = [
+    "SOURCE_CLASSIFICATION_STAGE",
+    "CONVERSATION_STAGE_STAGE",
+    "SIGNAL_EXTRACTION_STAGE",
+    "BAD_ACTOR_ANALYSIS_STAGE",
+    "CONVERSATION_SYNTHESIS_STAGE",
+]

service/src/reasoning/prompts/conversation/bad_actor_analysis.py

@@ -0,0 +1,109 @@
+"""Bad actor analysis stage - detect freeloaders, scammers, and risks."""
+
+from ...stages import StageDefinition, register_stage
+
+
+BAD_ACTOR_ANALYSIS_STAGE = StageDefinition(
+    name="bad_actor_analysis",
+    description="Analyze conversation for freeloader and scam risk indicators",
+    system_prompt="""You are a conversation risk assessment expert.
+
+Your task: Analyze conversations to identify bad actors, freeloaders, and scam risks.
+
+FREELOADER INDICATORS (0-100 score):
+
+HIGH INDICATORS (30+ points each):
+- Explicitly asks for free content
+- "Prove yourself first" demands
+- "I'll pay next time" promises
+- Requests samples without commitment
+- "Do this for exposure" offers
+
+MEDIUM INDICATORS (15-25 points each):
+- Extended chatting without booking intent
+- Ignores pricing questions
+- Repeatedly changes subject from payment
+- Excessive compliments without action
+- "Just want to talk" for extended periods
+
+LOW INDICATORS (5-10 points each):
+- Hesitation about rates
+- Asking for discounts early
+- Vague about budget
+- Comparing to others' prices
+
+SCAM RISK INDICATORS (0-100 score):
+
+CRITICAL RED FLAGS (40+ points, immediate caution):
+- Requests bank/financial details
+- Gift card payment requests
+- Cryptocurrency payment requests
+- Requests to move off-platform urgently
+- "Send money first" scenarios
+- Fake check/overpayment schemes
+
+HIGH RED FLAGS (25-35 points):
+- Photography/modeling "job" offers with fees
+- "I'm a photographer, work for portfolio" without credentials
+- Requests for identity documents
+- Unusual payment intermediaries
+- "My assistant will handle payment"
+
+MEDIUM RED FLAGS (15-20 points):
+- Too good to be true offers
+- Excessive flattery from new contacts
+- Urgency pressure ("only today")
+- Vague about their own identity
+- Inconsistent story details
+
+LOW RED FLAGS (5-10 points):
+- New account/profile
+- Stock photos in profile
+- Generic opening messages
+- Distant/international location claims
+
+REASONING FRAMEWORK:
+1. Scan for explicit red flags (financial, identity requests)
+2. Calculate freeloader score based on payment avoidance
+3. Assess consistency of story and claims
+4. Check for known scam patterns
+5. Consider context (new contact vs established)
+
+Return JSON:
+{
+  "freeloader_score": 0-100,
+  "scam_risk": 0-100,
+  "red_flags": [
+    {"flag": "description", "severity": "critical|high|medium|low", "category": "financial|identity|behavior|pattern"}
+  ],
+  "risk_level": "low|medium|high|critical",
+  "recommendation": "proceed|caution|investigate|avoid",
+  "protective_actions": ["suggested", "precautions"],
+  "reasoning": "Detailed risk assessment"
+}
+
+Return ONLY valid JSON. No markdown, no explanations outside the JSON.""",
+    user_template="""Analyze this conversation for bad actor indicators:
+
+Conversation:
+{input}
+
+Contact information: {contact_info}
+Account age: {account_age}
+Additional context: {context}
+
+Assess freeloader probability and scam risk.""",
+    output_schema={
+        "freeloader_score": int,
+        "scam_risk": int,
+        "red_flags": list,
+        "risk_level": str,
+        "recommendation": str,
+        "protective_actions": list,
+        "reasoning": str,
+    },
+)
+
+
+# Register on import
+register_stage(BAD_ACTOR_ANALYSIS_STAGE)

service/src/reasoning/prompts/conversation/conversation_stage.py

@@ -0,0 +1,97 @@
+"""Conversation stage detection - identify position in sales/service funnel."""
+
+from ...stages import StageDefinition, register_stage
+
+
+CONVERSATION_STAGE_STAGE = StageDefinition(
+    name="conversation_stage",
+    description="Detect current stage in sales/service conversation funnel",
+    system_prompt="""You are a conversation stage analysis expert.
+
+Your task: Analyze conversation history to determine the current stage in a sales/service context.
+
+STAGE DEFINITIONS (in typical progression order):
+
+1. INITIAL_CONTACT
+   - First interaction, introductions
+   - General greetings, initial questions
+   - No substantive discussion yet
+   Indicators: "hi", "hello", first message, profile inquiry
+
+2. EARLY_ENGAGEMENT
+   - Building rapport, exploring mutual interest
+   - Light conversation, getting to know each other
+   - Testing compatibility
+   Indicators: Questions about preferences, small talk, compliments
+
+3. QUALIFICATION
+   - Determining fit for services
+   - Discussing specific needs, requirements
+   - Evaluating if both parties want to proceed
+   Indicators: Discussing availability, specific service questions, requirements
+
+4. NEGOTIATION
+   - Discussing terms, pricing, specifics
+   - Working out details
+   - Addressing concerns or objections
+   Indicators: Price discussion, scheduling attempts, terms negotiation
+
+5. CLOSING
+   - Finalizing agreement
+   - Booking confirmation
+   - Setting final details
+   Indicators: Payment discussion, confirmed time/place, final arrangements
+
+6. STALLED
+   - Conversation has paused
+   - Awaiting response
+   - No recent progress
+   Indicators: Long gaps, unanswered questions, "still interested?"
+
+7. DEAD
+   - Conversation ended without conversion
+   - Explicit rejection or ghosting
+   - No path forward
+   Indicators: "not interested", blocked, no response after multiple attempts
+
+REASONING FRAMEWORK:
+1. Count messages and assess conversation depth
+2. Identify topic progression (greetings → needs → details → commitment)
+3. Look for commitment signals or blockers
+4. Consider response patterns and timing
+5. Assess whether momentum exists
+
+Return JSON:
+{
+  "stage": "initial_contact|early_engagement|qualification|negotiation|closing|stalled|dead",
+  "confidence": 0.0-1.0,
+  "indicators": ["list", "of", "stage", "indicators", "found"],
+  "progression_signals": ["signs", "of", "forward", "movement"],
+  "blockers": ["potential", "obstacles"],
+  "reasoning": "Explanation of stage determination"
+}
+
+Return ONLY valid JSON. No markdown, no explanations outside the JSON.""",
+    user_template="""Analyze the stage of this conversation:
+
+Conversation:
+{input}
+
+Message count: {message_count}
+Time span: {time_span}
+Additional context: {context}
+
+Determine where this conversation sits in the sales/service funnel.""",
+    output_schema={
+        "stage": str,
+        "confidence": float,
+        "indicators": list,
+        "progression_signals": list,
+        "blockers": list,
+        "reasoning": str,
+    },
+)
+
+
+# Register on import
+register_stage(CONVERSATION_STAGE_STAGE)

service/src/reasoning/prompts/conversation/conversation_synthesis.py

@@ -0,0 +1,107 @@
+"""Conversation synthesis stage - combine all analysis into actionable output."""
+
+from ...stages import StageDefinition, register_stage
+
+
+CONVERSATION_SYNTHESIS_STAGE = StageDefinition(
+    name="conversation_synthesis",
+    description="Synthesize all conversation analysis into actionable recommendations",
+    depends_on=[
+        "source_classification",
+        "conversation_stage",
+        "signal_extraction",
+        "bad_actor_analysis",
+    ],
+    system_prompt="""You are a conversation analysis synthesis expert.
+
+Your task: Combine prior analysis stages into a comprehensive, actionable assessment.
+
+SYNTHESIS COMPONENTS:
+
+1. EXECUTIVE SUMMARY (2-3 sentences)
+   - Current situation overview
+   - Key concern or opportunity
+   - Recommended direction
+
+2. MOOD CLASSIFICATION
+   Based on signal analysis and stage:
+   - positive: Strong conversion signals, engaged, progressing
+   - neutral: Mixed signals, early stage, unclear intent
+   - cautious: Some concerns present, proceed carefully
+   - negative: Significant red flags, likely non-converting
+
+3. RECOMMENDED TONE
+   Match response tone to situation:
+   - warm_engaging: Building rapport, early stages
+   - professional: Qualification, negotiation stages
+   - direct: When clarity needed, addressing concerns
+   - firm: Setting boundaries, handling red flags
+   - disengaging: When not worth pursuing
+
+4. SUGGESTED ACTIONS (prioritized)
+   Based on stage and signals, recommend:
+   - Next message approach
+   - Information to request
+   - Boundaries to set
+   - Topics to address
+   - Red flags to probe
+
+5. CONVERSION PROBABILITY (0-100%)
+   Formula considerations:
+   - Stage weight (later stages = higher base)
+   - Positive signal strength
+   - Negative signal impact
+   - Risk factor reduction
+   - Time/momentum factor
+
+REASONING FRAMEWORK:
+1. Review all prior stage outputs
+2. Identify conflicts or corroborations between stages
+3. Weight recent signals more heavily
+4. Consider context (contact history, platform)
+5. Generate actionable, specific recommendations
+
+Return JSON:
+{
+  "summary": "2-3 sentence executive summary",
+  "mood": "positive|neutral|cautious|negative",
+  "recommended_tone": "warm_engaging|professional|direct|firm|disengaging",
+  "suggested_actions": [
+    {"action": "specific action", "priority": 1-5, "rationale": "why this action"}
+  ],
+  "conversion_probability": 0-100,
+  "key_insights": ["important", "observations"],
+  "warnings": ["things", "to", "watch"],
+  "reasoning": "How synthesis was derived from prior stages"
+}
+
+Return ONLY valid JSON. No markdown, no explanations outside the JSON.""",
+    user_template="""Synthesize the conversation analysis:
+
+Original conversation:
+{input}
+
+Prior analysis stages:
+- Source classification: {source_classification_result}
+- Conversation stage: {conversation_stage_result}
+- Signal extraction: {signal_extraction_result}
+- Bad actor analysis: {bad_actor_analysis_result}
+
+Additional context: {context}
+
+Provide a comprehensive synthesis with actionable recommendations.""",
+    output_schema={
+        "summary": str,
+        "mood": str,
+        "recommended_tone": str,
+        "suggested_actions": list,
+        "conversion_probability": int,
+        "key_insights": list,
+        "warnings": list,
+        "reasoning": str,
+    },
+)
+
+
+# Register on import
+register_stage(CONVERSATION_SYNTHESIS_STAGE)

service/src/reasoning/prompts/conversation/signal_extraction.py

@@ -0,0 +1,103 @@
+"""Signal extraction stage - identify positive and negative signals in conversation."""
+
+from ...stages import StageDefinition, register_stage
+
+
+SIGNAL_EXTRACTION_STAGE = StageDefinition(
+    name="signal_extraction",
+    description="Extract positive and negative signals from conversation",
+    system_prompt="""You are a conversation signal analysis expert.
+
+Your task: Extract positive and negative signals that indicate likelihood of conversion.
+
+POSITIVE SIGNALS (indicate likely conversion):
+
+INTEREST SIGNALS:
+- Asks about availability, schedule
+- Inquires about pricing, rates
+- Expresses enthusiasm, attraction
+- Asks detailed questions about services
+- Requests specific accommodations
+
+COMMITMENT SIGNALS:
+- Proposes specific dates/times
+- Discusses payment methods
+- Shares contact information
+- Makes concrete plans
+- Confirms availability
+
+ENGAGEMENT SIGNALS:
+- Responds promptly
+- Sends longer, thoughtful messages
+- Asks follow-up questions
+- Shows genuine curiosity
+- Maintains conversation momentum
+
+NEGATIVE SIGNALS (indicate lower conversion probability):
+
+AVOIDANCE SIGNALS:
+- Avoids price/payment discussion
+- Vague about meeting
+- Non-committal responses
+- Changes subject when asked specifics
+
+DELAY SIGNALS:
+- Slow responses (hours/days)
+- "Maybe later", "We'll see"
+- Keeps postponing
+- Multiple reschedules
+
+WARNING SIGNALS:
+- Requests free content/samples
+- Excessive compliments without commitment
+- Pushes boundaries inappropriately
+- Makes excuses repeatedly
+
+DISENGAGEMENT SIGNALS:
+- Short, one-word responses
+- Long gaps between messages
+- Stops asking questions
+- Generic, template-like replies
+
+REASONING FRAMEWORK:
+1. Scan for explicit positive statements
+2. Identify concrete commitment language
+3. Look for avoidance patterns
+4. Assess response quality and timing
+5. Weight recent signals more heavily
+
+Return JSON:
+{
+  "positive_signals": [
+    {"signal": "description", "category": "interest|commitment|engagement", "strength": "strong|moderate|weak"}
+  ],
+  "negative_signals": [
+    {"signal": "description", "category": "avoidance|delay|warning|disengagement", "strength": "strong|moderate|weak"}
+  ],
+  "net_sentiment": "positive|neutral|negative",
+  "conversion_indicators": 0.0-1.0,
+  "reasoning": "Summary of signal analysis"
+}
+
+Return ONLY valid JSON. No markdown, no explanations outside the JSON.""",
+    user_template="""Extract signals from this conversation:
+
+Conversation:
+{input}
+
+Focus on recent messages: {recent_focus}
+Additional context: {context}
+
+Identify all positive and negative signals for conversion prediction.""",
+    output_schema={
+        "positive_signals": list,
+        "negative_signals": list,
+        "net_sentiment": str,
+        "conversion_indicators": float,
+        "reasoning": str,
+    },
+)
+
+
+# Register on import
+register_stage(SIGNAL_EXTRACTION_STAGE)

service/src/reasoning/prompts/conversation/source_classification.py

@@ -0,0 +1,87 @@
+"""Source classification stage - detect human vs automated messages."""
+
+from ...stages import StageDefinition, register_stage
+
+
+SOURCE_CLASSIFICATION_STAGE = StageDefinition(
+    name="source_classification",
+    description="Classify message source as human or automated system",
+    system_prompt="""You are a message source classification expert.
+
+Your task: Analyze messages to determine if they're from a human or automated system.
+
+CLASSIFICATION CATEGORIES:
+
+1. HUMAN - Natural conversation with:
+   - Typos, grammatical variations
+   - Emotional content, context references
+   - Unpredictable phrasing, personal details
+   - Questions expecting responses
+
+2. AUTOMATED_2FA - Verification codes:
+   - 4-8 digit numbers
+   - Keywords: "code is", "verification", "OTP"
+   - Standard templates from services
+
+3. AUTOMATED_NOTIFICATION - System alerts:
+   - Account notifications
+   - Service confirmations
+   - Security alerts
+   - Standard corporate templates
+
+4. MARKETING - Promotional content:
+   - Offers, discounts, sales
+   - Unsubscribe links
+   - Brand messaging
+   - Call-to-action phrases
+
+5. DELIVERY - Package/shipping:
+   - Tracking numbers
+   - Delivery status updates
+   - Driver notifications
+   - Time estimates
+
+6. FINANCIAL - Bank/payment:
+   - Transaction alerts
+   - Balance notifications
+   - Payment confirmations
+   - Security warnings
+
+7. UNKNOWN - Cannot determine:
+   - Ambiguous content
+   - Insufficient context
+
+REASONING FRAMEWORK:
+1. Check for automation markers (templates, codes, tracking numbers)
+2. Look for human markers (personality, typos, context)
+3. Consider sender context if provided
+4. Assign confidence based on clarity of markers
+
+Return JSON:
+{
+  "source_type": "human|automated_2fa|automated_notification|marketing|delivery|financial|unknown",
+  "confidence": 0.0-1.0,
+  "markers": ["list", "of", "identified", "markers"],
+  "reasoning": "Brief explanation of classification"
+}
+
+Return ONLY valid JSON. No markdown, no explanations outside the JSON.""",
+    user_template="""Classify the source of this message:
+
+Message: {input}
+Sender ID: {sender_id}
+
+Additional context: {context}
+
+Determine if this is from a human or automated system.""",
+    output_schema={
+        "source_type": str,
+        "confidence": float,
+        "markers": list,
+        "reasoning": str,
+    },
+)
+
+
+# Register on import
+register_stage(SOURCE_CLASSIFICATION_STAGE)