Session Handoff: Care Response Library & Theological Tradition System

What Happened This Session

The founder (John, CPE-trained, Stephen Leader) designed the pastoral care philosophy for all ChurchWiseAI AI products. This was a foundational session — not incremental feature work. It defined the company's care DNA.

Strategic Decisions Made

1. "Caring Bridge" Philosophy — AI agents are a caring bridge to a real person. They listen, empathize, name feelings, and bridge to humans. They do NOT pray, counsel, take confessions, offer solutions, or give theology.

2. No Spiritual Harm Rule (P0 ABSOLUTE) — Products will NEVER add to the spiritual harm the church has caused. Never say LGBTQ+ is sinful, never attribute conditions to demons, never make theological judgments. The founder left the CRC over this. This is company DNA.

3. Care Response Library — Pre-researched CPE-quality response patterns eliminate Sonnet escalation (10-15s → sub-2s). Haiku selects from seminary-quality material instead of improvising.

4. 17 Theological Tradition System — Each tradition gets calibrated care (clergy titles, care structures, what to avoid, bridge language). The competitive moat no one else has.

5. Tradition Demo Page — "See how ChurchWiseAI serves YOUR tradition" — marketing that IS the product.

What Was Built (in Supabase)

Table	Rows	Status
pastoral_care_responses	0	Schema live, needs population
tradition_care_context	14	Universal + 13 traditions from Python module
denominations.tradition_key	64	All denominations mapped to tradition_keys
denominations.primary_lens_id	64	All denominations mapped to default lens

What Was Researched (in conversation context — NOT yet in files)

6 research agents produced ~150,000 words of CPE-quality care response patterns:

Report	Subcategories	Key Content
Grief & Loss	10	Spouse, child, miscarriage, parent, sibling, friend, anticipated, pet, anniversary, sudden/traumatic
Illness & Health	12 + cross-cutting	Cancer, terminal, chronic, mental health, caregiver, disability, infertility, surgery, long COVID, childhood illness, dementia + theodicy + toxic positivity
Addiction & Recovery	12	Alcohol, drugs, pornography, gambling, family member, relapse, codependency, eating disorders, self-harm, phone/social, workaholism, LGBTQ+/church shame
Anxiety & Mental Health	15	General anxiety, financial, parenting, health anxiety, social, depression, suicidal ideation, panic attacks, PTSD, burnout, seasonal, PPD, elder loneliness, fear of death, scrupulosity
Relationship Crisis	16	Marriage, divorce, affair discovered, affair confession, DV, blended family, prodigal child, estrangement, toxic family, singleness, dating, LGBTQ+ family, pornography, intimacy, aging parents, church hurt
Spiritual Crisis + Missing	15 + 14 missing	Doubt, anger at God, church hurt, dark night, guilt, legalism, deconstruction, new to faith, interfaith, end-of-life, returning, theodicy, unworthiness, scrupulosity, moral injury + incarceration, immigration, disaster, racial trauma, pregnancy, aging, financial ruin, veterans, assault, trafficking, cult recovery, body image, bullying, career

CRITICAL: The full research text is in the Claude conversation from 2026-04-01 (task notification outputs). It was NOT saved to files because the temp files were cleaned up before copying. The next session must either:

1. Re-read the conversation context (if still available)
2. Or re-run targeted research agents for specific categories needed for DB population

6 tradition research agents also completed (Catholic, Orthodox, Baptist, Pentecostal, Non-Denom, Lutheran, ELCA, LCMS, PCUSA, PCA, CRC, Methodist, Anglican, Mennonite, AME, COGIC, National Baptist, SDA, Churches of Christ, UU). 13 were written to tradition_care_contexts.py (3,678 lines) and inserted into Supabase. 8 traditions (Catholic, Orthodox, Methodist, Anglican, Mennonite, SDA, CoC, UU) were returned as text in the conversation but NOT yet inserted into the DB.

Integration Plans (Reviewed)

Both plans were reviewed by the voice-agent-engineer and a QA expert reviewer.

Voice Agent — 4 file changes: Fix rag_context bug (CareAgent has been operating without knowledge base), add tradition context loader, thread through Coordinator → CareAgent handoff. Plan at: knowledge/products/care-responses/integration-plan-voice.md

Chatbot — 2 file changes: New pastoral-care-library.ts (detectCareIntent, loadCareLibraryContext, buildCareSystemPrompt) + ~60 lines in route.ts. Sonnet preserved as fallback. Plan at: knowledge/products/care-responses/integration-plan-chatbot.md

Critical review findings:

• C1: Use church.denomination → DENOMINATION_TO_TRADITION_KEY, NOT lensId → tradition_key (lossy)
• C2: CareAgent rag_context bug confirmed — fix independently
• C3: Empty table fallback must be transparent (zero behavioral change)

Bug Discovered

CareAgent rag_context not injected — agents.py line 51 stores it, line 72 never injects it into instructions. The Care Agent has been operating WITHOUT the church's knowledge base. Both reviewers confirmed. High-value fix, zero risk.

---

Next Steps (Priority Order)

P0 — Do First (Independent, High Value)

1. Fix CareAgent rag_context bug — agents.py line 72. Add if rag_context: instructions += f"\n\n--- KNOWLEDGE BASE ---\n{rag_context}". Deploy immediately. File: churchwiseai-web/voice-agent-livekit/verticals/church/agents.py

P1 — Core Implementation

2. Insert missing 8 traditions into Supabase — Catholic, Orthodox, Methodist, Anglican, Mennonite, SDA, Churches of Christ, UU. Research exists in conversation from 2026-04-01 or can be re-generated from tradition_care_contexts.py data + new research agents.

3. Populate pastoral_care_responses table — Extract structured data from the 6 care research reports. This is the biggest task (~350+ rows). May need dedicated research agents to re-generate the structured data and convert to SQL INSERTs.

4. Implement voice agent integration — Follow plan at integration-plan-voice.md. 4 files, reviewed and approved.

5. Implement chatbot integration — Follow plan at integration-plan-chatbot.md. 2 files, reviewed and approved. Port DENOMINATION_TO_TRADITION_KEY from Python to TypeScript.

P2 — Onboarding & Marketing

6. Build denomination selector UI — Cascading Family → Denomination dropdown on signup/claim flow. Auto-assigns theological lens + tradition_key. Spec at: knowledge/products/care-responses/onboarding-denomination-spec.md

7. Make city/state required — Change labels to City/Town and State/Province. Add US states + Canadian provinces + Other dropdown.

8. Build tradition demo page — "See how ChurchWiseAI serves YOUR tradition" section on marketing site. Pulls from same tradition_care_context table. Spec at: knowledge/products/care-responses/tradition-demo-page-concept.md

P3 — Future Enhancement

9. Founder tradition review — John gut-checks all 17 tradition blocks against his pastoral training. Especially CRC (his home tradition), AME/COGIC/National Baptist (most underserved), and UU (hardest to serve).

10. Per-turn care response RAG — When pastoral_care_responses is populated, integrate into per-turn RAG pipeline (not system prompt) for dynamically loading relevant response patterns based on conversation topic.

11. Ad copy production — Use moat positioning copy at moat-theological-lens-copy.md to create actual ads, landing page sections, email campaigns.

---

Key Files Created/Modified This Session

File	What
knowledge/products/care-responses/README.md	Research index — all 6 reports summarized
knowledge/products/care-responses/integration-plan-voice.md	Voice agent integration plan (reviewed)
knowledge/products/care-responses/integration-plan-chatbot.md	Chatbot integration plan (reviewed)
knowledge/products/care-responses/onboarding-denomination-spec.md	Denomination selector + required location spec
knowledge/products/care-responses/moat-theological-lens-copy.md	Marketing copy — "the only AI that cares the way YOUR church cares"
knowledge/products/care-responses/tradition-demo-page-concept.md	Demo page concept — tradition selector with examples
voice-agent-livekit/verticals/church/tradition_care_contexts.py	3,678 lines — 13 tradition context blocks + helper functions
~/.claude/projects/C--dev/memory/project_caring_bridge_philosophy.md	Core product philosophy
~/.claude/projects/C--dev/memory/project_care_agent_boundaries.md	Agent boundary rules + graceful fallback
~/.claude/projects/C--dev/memory/feedback_no_spiritual_harm.md	P0 ABSOLUTE — no spiritual harm rule
~/.claude/projects/C--dev/memory/user_pastoral_training.md	Founder's CPE + Stephen Leader background
~/.claude/projects/C--dev/memory/project_tradition_demo_marketing.md	Tradition demo = marketing gold
INSTITUTIONAL_KNOWLEDGE.md	Updated with "The Absolute Rule — No Spiritual Harm" section
DECISION_LOG.md	7 new entries for 2026-04-01

Supabase Changes

Change	Type
pastoral_care_responses table created	DDL
tradition_care_context table created	DDL
Universal (Lens 0) row inserted with all rules	DML
13 tradition rows inserted from Python module	DML
denominations.tradition_key column added + populated (64 rows)	DDL + DML
denominations.primary_lens_id column added + populated (64 rows)	DDL + DML

Memory System Updates

• project_caring_bridge_philosophy.md — NEW
• project_care_agent_boundaries.md — NEW
• feedback_no_spiritual_harm.md — NEW
• user_pastoral_training.md — NEW
• project_tradition_demo_marketing.md — NEW
• user_crc_background.md — UPDATE NEEDED (John has left CRC, is now affirming)
• MEMORY.md index updated with all new entries

Context for Next Agent

This session defined the SOUL of the product. The technology is the container. Key things to internalize:

1. John is a seminary-trained pastor (CPE + Stephen Leader) who just left the CRC over LGBTQ+ harm. He knows pastoral care at a professional level. Don't explain basic concepts to him.

2. The "Caring Bridge" philosophy is non-negotiable: listen, empathize, bridge to humans. The AI is the SAFE layer between a hurting person and the church.

3. The No Spiritual Harm rule overrides everything. If in doubt about whether something could cause harm, say nothing theological and bridge to the pastor.

4. The 17-tradition system is the competitive moat. No competitor has this. Treat it with the same care you'd give production infrastructure.

5. The tradition demo page concept is the marketing conversion tool. When a priest sees "Father" instead of "Pastor" in the demo, trust is established instantly.