2️⃣ Copy–paste prompt for Replit (CTO work)

Here’s a single block you can send straight to the Replit agent to implement A + B:

You are a senior Python/FastAPI engineer working on the LIA backend. We’ve just verified that /api/lia/outbound works end-to-end (LIA → Twilio → WhatsApp).

Next, we need two safety features:

1) Only allow outbound for specific payment bands (18, 12, 9, 6, 3).
2) Prevent duplicate conversations for the same customer/campaign/stage/channel.

Do NOT change any LLM/URE logic or db schema. Just add validation + dedupe in the outbound flow.

────────────────────────
TASK 1 – Validate payments_remaining band

In app/lia_outbound.py (where the outbound request model and process_outbound_message live):

1. After parsing the request payload, enforce:

   - `payments_remaining` must be one of: 18, 12, 9, 6, 3.

2. If it is not, raise:

   ```python
   raise HTTPException(
       status_code=400,
       detail="payments_remaining must be one of [18, 12, 9, 6, 3] for this endpoint"
   )


Use the same band mapping you already use to derive stage:

18 → "18"

12 → "12"

9 → "9"

6 → "6"

3 → "3"

If there is an existing function that maps payments_remaining → stage, reuse it. Otherwise, implement a small helper.

────────────────────────
TASK 2 – Duplicate-conversation fail-safe

Definition of duplicate:

Same customer_id

Same campaign

Same stage (derived from payments_remaining)

Same channel

We must NOT start a new conversation if one already exists for that combination.

In app/lia_outbound.py, before creating a new dp_conversations row, add a check using the existing db connection helper:

Pseudocode:

existing = await conn.fetchrow(
    """
    SELECT conversation_id, status
    FROM dp_conversations
    WHERE customer_id = $1
      AND campaign    = $2
      AND stage       = $3
      AND channel     = $4
    LIMIT 1
    """,
    customer_id,
    campaign,
    stage,
    channel,
)


If existing is not None:

Do NOT create a new conversation.

Do NOT generate a new outbound message for Twilio.

Raise an HTTP 409 with a clear payload, for example:

raise HTTPException(
    status_code=409,
    detail={
        "error": "Conversation already exists for this customer, campaign, stage and channel",
        "conversation_id": str(existing["conversation_id"]),
        "stage": stage,
    },
)


(If your existing error handling expects detail to be a string, you can instead return a string and include conversation_id in the log – but ideally return a JSON object as shown.)

If existing is None:

Proceed with the existing logic to:

create a new dp_conversations row,

generate message_text,

return OutboundResponse as currently implemented.

Make sure that:

This duplicate check runs regardless of sandbox_mode (true/false).

Status (active/paused/dnc) does NOT bypass the check; any existing row for the same tuple should block a new conversation from being created.

────────────────────────
CONSTRAINTS

Do NOT change any table schemas.

Do NOT change how message_text, conversation_id, or to_number are built if the request is valid and not a duplicate.

Do NOT change inbound or nudge endpoints.

Only:

Enforce payments_remaining ∈ {18, 12, 9, 6, 3}.

Enforce the duplicate-conversation rule described above for /api/lia/outbound.

────────────────────────
WHEN DONE

Reply with:

The code snippet showing:

where you validate payments_remaining,

and how stage is derived.

The code snippet showing:

the SELECT on dp_conversations,

and the HTTPException you raise on duplicates.

Example 409 JSON body returned for a duplicate case (payload can be anonymised).


Once that’s in, Make will see:

- **200** → send via Twilio  
- **400** (bad band / missing phone, etc.) → log/skips  
- **409** (duplicate conversation) → log/skips  

That gives you the safe “don’t spam / don’t double start” behaviour you want.

---

## 3️⃣ Which agent for scripts / guides?

> “we could ask another agent (cant remember which) to help generate the script guides for each campaign?”

That’s your **LIA – CPO (Chief Product / Prompt Officer) agent** – the one we used when designing LIA Prompt v2 and behaviour.

- **CTO agent (me)** → backend, APIs, DB, logic, safety nets  
- **CPO agent** → behavioural design, script guides, tone, per-campaign playbooks, “what should Lia say at 18/12/9/6/3 months, PCP vs HP, new vs used, etc.”

So once you’re happy the backend safety bits above are in, you can open a fresh chat and start with:

> “You are now LIA – CPO (Chief Product / Prompt) agent…”

and we’ll build the script/guide pack for each band.

---

If you like, after Replit finishes this duplicate/band work, we can:

- design how to **loop** through a DB view of eligible customers per band,  
- and hook that into Make as a daily/weekly mini-campaign.