Kali Discovery: Phase 3 — Proposed Solution

Kali Discovery: Phase 3 — Proposed Solution
### Durga said:
Now propose your solution. Include: architecture, sequence, state management, edge cases, success criteria. Be specific – this is your build plan.

### Kali replied:
JSON error injected into SSE stream
Okay, here’s my proposed solution for debt chasing and enforcement, addressing architecture, sequence, state management, edge cases, and success criteria.
**Architecture**
* I, Kali, will operate as a sub-agent within the Durga fleet.
* **Data Source:** Xero API for aged receivables. Authentication via `xero_auth.py`, data retrieval via `fetch_invoices.py`.
* **Debtor Scoring:** `score_debtor.py` will analyze each debtor based on risk and relationship value.
* **Escalation Engine:** `escalation_engine.py` will determine the correct escalation stage and action.
* **Communication:** Sending reminders via `send_reminder.py` using the Xero email API, formal letters and statutory demands via Parvati (human-mediated).
* **State Management:** `kali_state.json` will maintain persistent state across sessions, tracking debtors chased, escalation stages, and communication history.
* **Approval Workflow:** Anthony (human) approval via Telegram at escalation stages 2-4.
* **Reporting:** Actions reported to Durga via A2A.
**Sequence**
1. **Initialization:**
* On startup, I will read `SOUL.md`, `USER.md`, and recent `memory/` files for context.
* I will attempt to load `kali_state.json`. If the file doesn’t exist, I’ll create a new one.
2. **Data Retrieval:**
* Authenticate with Xero using `xero_auth.py` (tool yet to be provided).
* Fetch aged debt data using `fetch_invoices.py` (tool yet to be provided).
3. **Debtor Analysis and Escalation:**
* Iterate through unpaid invoices.
* For each debtor, check `kali_state.json` to see if they have already been chased at the current escalation stage.
* If already chased, skip to the next debtor.
* Score the debtor using `score_debtor.py` (tool yet to be provided).
* Determine the correct escalation stage and action using `escalation_engine.py` (tool yet to be provided).
4. **Escalation Actions:**
* **Stage 1 (30-60 days):** Friendly reminder.
* Send a friendly reminder using `send_reminder.py` (tool yet to be provided).
* Update `kali_state.json` to record the reminder sent, including timestamp.
* **Stage 2 (61-90 days):** Firm reminder.
* Send a message to Durga via A2A, requesting approval from Anthony via Telegram to send a firm reminder.
* If approved:
* Send a firm reminder using `send_reminder.py` (tool yet to be provided).
* Update `kali_state.json` to record the reminder sent, including timestamp.
* **Stage 3 (91-120 days):** Formal letter.
* Send a message to Durga via A2A, requesting approval from Anthony via Telegram to send a formal letter via Parvati.
* If approved:
* Record the sent communication in `kali_state.json`.
* **Stage 4 (120+ days):** Statutory demand.
* Send a message to Durga via A2A, requesting approval from Anthony via Telegram to send a statutory demand via Parvati.
* If approved:
* Record the sent communication in `kali_state.json`.
* **Stage 5:** Flag for legal action.
* Send a message to Durga via A2A, flagging the debtor for legal action.
* Record the flag in `kali_state.json`.
5. **Reporting and State Update:**
* Report all actions to Durga via A2A.
* Update `kali_state.json` with the date and details of the actions taken.
**State Management Details**
* The `kali_state.json` file