Billing Ops Edho Ferdian

Billing Ops — Edho Ferdian Mode

Customer billing operations and finance billing operations are consolidated into one skill per D-029 recommendation. Built ahead of its original trigger (Edho actually charging money for a shipped product) per explicit user request, D-035 — treat vendor-specific integration details (e.g. calls into a payments API) as needing re-verification once a real billing system exists, since none of this has been exercised against a live one yet.

Where this sits

• This skill covers billing operations: reading a billing incident,

classifying it, and deciding what to do about it — customer-facing and operator-facing.

• It does not own code-quality or spec-consistency review. When the question

is "does the code actually enforce what we're claiming," that check is an instance of the Blueprint/Spec Consistency domain in code-review-edho-ferdian — cross-reference that skill for the review methodology rather than redefining it here.

• It does not own the underlying payment/checkout/webhook implementation.

backend-engineering-edho-ferdian and data-layer-patterns-edho-ferdian own how that code is structured; this skill only asks whether it matches what's being claimed.

Two layers of the same problem

Billing questions arrive at two altitudes, and mixing them produces bad answers:

1. Customer-specific remediation — one customer, one incident. What

happened to them, and what's the safest fix.

1. Operator/product truth — is the underlying billing behavior actually

what the pricing page, sales copy, or a support answer implies. This is broader than any single ticket.

Always classify which layer a question is at before answering. A duplicate charge complaint (layer 1) can reveal a broken self-serve cancel flow (layer 2) — surface both, but don't let the specific incident stand in for the general claim, and don't let a general policy answer skip the specific customer's actual situation.

Step 1 — Classify the incident (customer layer)

Five categories:

1. Duplicate subscription — same customer billed by more than one active

subscription. Confirm it's actually duplicate (same product, same customer identity) before treating it as one.

1. Multi-seat intent vs. accidental duplicate — what looks like a

duplicate may be a deliberate second seat/license. Check stated intent and usage pattern before assuming error. Do not collapse the two.

1. Failed or incomplete checkout — payment attempted but never

completed, or completed without provisioning the product. Distinguish "customer was never charged" from "customer was charged but not provisioned" — they need different fixes.

1. Missing or broken self-serve control — the customer could not do

something (cancel, downgrade, update a seat count) that the product claims they can do. This is a product gap, not just a support ticket; log it as one (see Step 4).

1. Product defect / unmet value — the customer is billed correctly but

the product itself is broken or not delivering what was paid for. This is a product-quality question wearing a billing complaint's clothes; route the underlying defect to the owning engineering skill once the billing side is resolved.

Identify the customer cleanly first (matching identity, not guessing from a name or email fragment), and never expose secrets, full payment card data, or another customer's billing details while doing so.

Step 2 — Verify the claim against the code, not the copy

Never assert a billing/entitlement behavior — "it's per-seat," "quota resets monthly," "cancel takes effect immediately" — on the strength of the pricing page, onboarding copy, or a past support answer alone. If the answer matters (refund decision, policy answer, escalation), trace it to the code path that is supposed to enforce it:

• checkout / plan-selection logic
• entitlement or quota calculation
• seat counting (does adding a checkout quantity actually change what the

account can do, or does it stop at "quantity purchased")

• self-serve billing-portal actions (cancel, downgrade, seat change)

Do not say a behavior is "per-seat" (or any other specific billing model) unless the entitlement path in the code actually enforces it — a checkout line item is not proof of enforcement. This is the billing-domain instance of ground-truth-over-narrative: verify against what the code does, not what the page says it does. Where this check turns into a full review of the enforcement code itself, hand it to code-review-edho-ferdian's Blueprint/Spec Consistency domain rather than re-deriving that methodology here.

Also don't assume a duplicate subscription implies duplicate value received — a duplicate charge and a duplicate benefit are different claims, and a refund decision should be based on which one actually happened.

Step 3 — Diagnosis is not execution

This skill produces a diagnosis and a recommended action. It does not execute financial changes on its own authority.

Hard rule, consistent with the global "Explicit permission required" action category for purchase/payment operations: refunds, credits, subscription cancellations, and any other financial adjustment are EXPLICIT-PERMISSION-REQUIRED actions. State the diagnosis, state the recommended action and its revenue impact, and get the user's explicit go-ahead before anything gets executed — a support-style script drafted for the user, not a payment sent on their behalf. Never treat "the customer is clearly right" or "this is obviously a duplicate" as authorization to skip that confirmation.

Prefer the safest reversible action when a fix is warranted, and prefer whatever billing platform the ecosystem already has connected over building a custom workaround.

Step 4 — Close with a structured handoff

Every billing incident response ends with:

SNAPSHOT
- data source and its timestamp (live vs. last-known/saved — always say
  which; never present a stale snapshot as current)
- relevant subscription/charge/refund state

CUSTOMER IMPACT
- who is affected, what happened to them specifically

PRODUCT TRUTH
- what the code actually enforces (from Step 2)
- what the pricing page / sales copy / prior support answer claims
- flag any gap between the two explicitly

RECOMMENDED ACTION
- refund / credit / preserve / convert / no-op
- marked EXPLICIT PERMISSION REQUIRED if it involves money moving or a
  subscription being changed — not yet executed

PRODUCT GAP
- the exact follow-up item to build or fix, if Step 1 or Step 2 surfaced one
  (e.g. a broken self-serve control, an unenforced pricing claim)

Pitfalls

• Treating a stale/cached billing snapshot as live data without saying so.
• Answering a pricing/entitlement question from memory or from marketing

copy instead of tracing the enforcing code path.

• Letting "this looks like an obvious refund" skip the explicit-permission

step — obviousness is not authorization.

• Conflating failed/incomplete checkout attempts with actual net revenue.
• Jumping straight from diagnosis to a financial action without separating

customer impact from the broader product-truth question.

Language routing (fixed — see skill-authoring-edho-ferdian's canonical contract)

Communication to the user in Bahasa Indonesia; the structured handoff report (Snapshot/Customer Impact/Product Truth/Recommended Action/Product Gap) in English, since it gets pasted into an issue tracker or handed to another skill. Full contract: skill-authoring-edho-ferdian §7.

• Comparing competitor pricing from memory when the ecosystem has research

tooling available to check current evidence instead.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: edhoferdian
• Source: edhoferdian/EEF
• License: MIT

Install and usage instructions live in the source repository linked above.