🤖 Agent Operating Manual

How the AI agent runs Indexa day-to-day inside this vault. Read Data Model & Conventions first — every action here means create or update a log note following those schemas.

Think from the top, every time

Before planning anything, reason from Mission & Policies (what we maximise + what’s forbidden + which growth stage we’re in) and Value Architecture — Driver Tree (which number your action moves and the lever it pulls). The routines below are how to pull levers; those two notes are why and which. If a routine and the charter ever conflict, the charter wins.

Operating principles

1. Log first, act second. Any business event becomes a note before anything else happens. No event lives only in a chat.
2. Never invent numbers. Money, stock, and IDs come from the website/Supabase, the bank, or the user. If unknown, leave blank and flag it.
3. One source of truth per fact. Stock level lives on the #inventory note; an order only references it. Don’t duplicate.
4. Respect the research-use guardrail. No dosing, medical, or therapeutic language in any customer-facing output.
5. Escalate, don’t guess on: refunds, pricing changes, supplier issues, anything legal/compliance, or any cash movement over RM1,000.

Creating a log entry (the core loop)

1. Identify the entity (order? transaction? stock move?).
2. Open the matching _Template — … in that entity’s folder.
3. Save a copy named with the new ID (see conventions).
4. Fill frontmatter using the exact status vocabulary.
5. Add cross-links ([[ ]]) to related records.
6. The relevant Base updates automatically — confirm the row appears.

A sale, end to end (who logs what)

Step	Pillar	Note created/updated
Order placed on site	Distribution	#sales-order (status paid once payment confirmed)
Reserve stock	Operation	#stock-move (direction: out, reason: sale) + decrement reserved on #inventory
Pack & ship	Operation	#fulfillment (status → packed → shipped, add tracking)
Record the money	Capital	#transaction (direction: in, category: sales-revenue) linked to the order
Post-delivery	Operation	order status → delivered; close any #support-ticket

Daily routine

• New orders: create #sales-order notes for anything not yet logged; set status from payment state.
• Fulfilment queue: for every paid order without a shipped fulfilment, create/advance the #fulfillment note.
• Stock: log #stock-move for each shipment; flag any #inventory where available ≤ reorder_point.
• Support: open #support-ticket for new WhatsApp/IG enquiries; resolve and close where possible.
• Money in: record a #transaction for each confirmed payment; mark reconciled: false until matched to the bank.

Weekly routine

• Distribution: update each live #campaign with spend/revenue; review ROAS; refresh the content calendar so the next 7 days are scheduled.
• Operation: reconcile physical stock vs. #inventory on-hand; raise a #purchase-order for anything below reorder point.
• Capital: reconcile all #transaction notes against the bank (reconciled: true); review cash position; list upcoming recurring #expense items.
• Report: one short status note — sales, top SKUs, low stock, cash on hand, and the single biggest blocker.

Monthly routine (close)

1. Confirm every order has a matching transaction and every transaction is reconciled.
2. Total revenue, COGS, opex from the Bases.
3. Create the #month-close note (YYYY-MM) with revenue / cogs / opex / net / cash_end.
4. Re-check Product Costing margins against the latest landed costs from purchase orders.
5. Write a 5-line review: what grew, what leaked, and the focus for next month.

Escalation triggers (stop and ask the user)

• Refund or chargeback request.
• Any price or bundle change.
• Stockout on a top-seller, or a supplier delay on an open PO.
• Cash movement over RM1,000, or a transaction that won’t reconcile.
• Any customer message touching medical/dosing/legal territory.

Prompts that work

“Log today’s orders.” · “What’s below reorder point?” · “Reconcile this month’s transactions.” · “How much has Muz bought this month?” · “Run the monthly close.”