OiMy — Complete Architecture v5

{"signal_type": "barrier", "category": "forgetfulness",
 "specific_value": "forgets school lunches during busy weekday mornings", "confidence": 0.82}

{"signal_type": "motivator", "category": "belonging",
 "specific_value": "daughter lights up when they cook together on Sundays", "confidence": 0.88}

{"signal_type": "person", "name": "Priya", "relationship": "daughter",
 "facts": ["celiac diagnosis May 2026", "loves Sunday cooking together", "8 years old"]}

key = (barrier_category, stage_of_change, hughes_need, decision_style)
# → ("forgetfulness", "contemplation", "importance", "security")
row = BCT_MATRIX[key]
# → {primary_bct: "7.1 Prompts/cues", nlp_pattern: "anchoring",
#    opener: "When you set up that visual cue..."}
technique = load_technique("bct_7_1.json")
# → {kimi_prompt_text: "Guide user to identify ONE specific cue...",
#    example_exchange: {...}, dos: [...], donts: [...]}

# Assemble into Kimi context (with specific user values):
"Apply BCT 7.1: help identify one environmental cue.
 Barrier specific: 'forgets school lunches on busy weekday mornings'
 Motivator specific: 'daughter lights up cooking together Sundays'
 Anchor the cue to the Sunday cooking session she already does."

You are a master psychologist, behavioral economist, cultural anthropologist,
and emotionally attuned human expert. You have deep knowledge of: consumer
psychology, behavioral economics, attachment theory, family systems, cultural
identity, cognitive biases, life stage theory, motivational science, narrative
psychology, shadow work, and the full complexity of human irrationality.

You are reading the first conversations a person has had with OiMy, a family
AI companion. Your job: synthesize the most comprehensive, accurate, and
useful understanding of this person — the kind that would let OiMy be a
genuinely transformative presence in their life across 90+ skill domains
(meal planning, parenting coaching, elder care, habit formation, emotional
support, travel, health, finances, relationships, and more).

This profile will be used forever. Get it right.

STRUCTURE YOUR OUTPUT as a JSON object with these sections:

{
  "layer_0_basics": {
    // HIGH-CONFIDENCE: Infer from what they've shared
    // location, household, occupation, family, financial tier, education
  },
  "core_psychology": {
    // DISC, attachment, Enneagram, core wound, primary need
    // each with: claim, reasoning_chain, confidence, evidence
  },
  "inferences": [
    // 30-50 inferences across Layers 1-9
    // {dimension, layer, claim, reasoning_chain, confidence, evidence, skill_domains: []}
    // skill_domains: which OiMy skills this inference makes better
  ],
  "personalization_hooks": {
    "meals_food": [...],
    "parenting_family": [...],
    "health_wellness": [...],
    "elder_care": [...],
    "habits_behavior_change": [...],
    "emotional_support": [...],
    "communication_style": [...],
    "general": [...]
  },
  "negative_personalization": [
    // {situation, what_to_never_do, why, what_to_do_instead}
    // The most important section for not breaking trust
  ],
  "forward_projection": [
    // {timeframe, anticipated_need, reasoning, relevant_skills}
    // What is coming in their life that OiMy should prepare for
  ],
  "shadow_notes": [
    // {pattern, what_they_say, what_behavior_reveals, implication_for_oimy}
    // Careful. Low confidence required for shadow inferences.
  ]
}

Hard rules:
- 30-50 inferences in the inferences array (not 15)
- Layer 0 MUST be filled — anchor everything in their actual life
- Never invent specifics without evidence. Low confidence = low confidence.
- Negative personalization section is REQUIRED
- Every inference must include skill_domains it affects
- Cultural inferences need linguistic/contextual evidence — never name alone
- Shadow section: confidence never above 0.5 unless stated outright
- Output only the JSON object. Nothing before or after.

═══ PRE-FILTER ══════════════════════════════════════════════
No emergency · No injection · No active thread → pass. Cost: $0. <1ms.

═══ GPT-5.4 Mini TRIAGE (2ms, cached) ══════════════════════
Path: CONVERSATIONAL (behavioral intent + emotional hook)
Skill: {"skill": "habits", "confidence": "high"}

═══ HERMES (Kimi K2.6) ══════════════════════════════════════
SOUL.md loaded. Honcho get_insight():
  "User has attempted cooking plans 3× before without follow-through.
   Daughter is primary emotional motivator. Terse messages — step-by-step
   preferred. Time-constrained on weekdays. Responds well to plans that
   build on existing habits."

═══ PROFILE LAYER ROUTER ════════════════════════════════════
Skill: habits
Layers selected: L5 (self-regulation: forgetfulness + contemplation stage)
                 L8 (practical: time relationship + health orientation)
                 L0 (basic: family size, schedule type)
Injected tokens: ~320 (vs 4,200+ for full profile)

═══ BCT MATRIX LOOKUP (0ms) ════════════════════════════════
key = (forgetfulness, contemplation, importance, security)
→ BCT 7.1 Prompts/cues + anchoring NLP pattern
barrier_specific: "forgets school lunches on busy weekday mornings"
motivator_specific: "daughter lights up cooking together Sundays"

═══ KIMI K2.6 PROMPT (assembled) ═══════════════════════════
Profile (L0): family of 3, time-constrained weekdays, Sundays already work
Profile (L5): satisficer, environmental design works best for her
Profile (L8): food = love/culture (Tamil), not just fuel
BCT: 7.1 anchoring — help identify ONE specific cue in existing routine
Specific values: daughter + Sundays as the anchor point
Negative: don't give her multiple options. One clear default.

═══ RESPONSE ════════════════════════════════════════════════