version: "1.0.0" name: tres-invoice-bill-matching description: > Match TRES ledger transactions to ERP invoices/bills (AP/AR) and optionally sync them to the connected ERP (Xero, QuickBooks Online, NetSuite). Trigger this skill whenever the user wants to match, link, close, reconcile, or sync an invoice or bill against a blockchain transaction — even if they don't say "skill" or use those exact words. Trigger phrases include: "match this invoice to a transaction", "match a bill to tx", "close invoice INV-123", "close bill 9988", "link this tx to a bill", "pay this invoice from this transaction", "set up this tx as AP", "set up this tx as AR", "sync this transaction as AP/AR", "match AP/AR", "find the invoice/bill for this hash", "what bill does this tx pay". Trigger ONLY for explicit AP/AR matching/closing intent — do NOT trigger for general transaction explanations (use tres-tx-story), for ingesting an explorer link into the ledger (use tres-explorer-tx-to-ledger), or for ERP connection setup itself (use tres-settings-management).

TRES — Invoice/Bill Matching & ERP Sync

End-to-end workflow that lets the user close an open ERP invoice or bill against a blockchain transaction in the TRES ledger, then optionally push the matched entry to the connected ERP.

The flow is the same regardless of which side the user starts from (a transaction hash or an invoice/bill ID). The skill walks through seven conversational steps (verify ERP → identify input → fetch & suggest → user picks → configure payment account & fiat → confirm & apply → loop). Stay terse — show numbered options, capture the user's pick, move on. Never run a mutation without explicit "yes" from the user.

Ground rules

1. Identify the org first. Begin with get_viewer and tell the user "You're connected to {orgName}."

This makes mistakes recoverable when someone has the wrong token.

1. Read before write. Always fetch the current state of the transaction, invoice/bill, and payment account

before showing a change summary. Surprises are worse than slow.

1. Mutations require explicit approval. Show a summary table (Transaction · Invoice/Bill · Payment Account ·

Fiat alignment · Sync) and ask "Apply these changes?" before any mutation. Only proceed on a clear yes.

1. Use schema introspection when in doubt. Field names, enum values, and argument shapes can drift. If a

query/mutation errors with "unknown field" or "invalid enum", call introspect(<TypeName>) or build_query(<operationName>) and adjust — don't guess. Operations specifically called out as "verify at runtime" below are the ones most likely to need this.

1. The skill is a loop. After a successful match, ask "Match another?" and restart from Step 2. Don't

re-check the ERP — that only happens once per session.

Step 1 — Verify ERP is connected

Run:

query { integration(first: 50) { results { id integratedApp isErp connectionStatus companyName } } }

Filter the results where isErp == true and connectionStatus == "ACTIVE". The supported ERPs you should recognize are Xero, QuickBooks Online (QBO), and NetSuite — integratedApp values are XERO, QUICKBOOKS, and NETSUITE. (If you encounter unknown values, introspect("IntegrationsQueryNode") will confirm the enum.) Ignore rows where integratedApp is empty — the API occasionally returns a null row.

• No connected ERP: Tell the user they need to connect one before matching can happen, point them at

https://app.tres.finance/settings/integrations, and stop.

• One connected ERP: Use it implicitly and just mention "Matching against {companyName} ({integratedApp})."
• Multiple connected ERPs: Ask which one to use — different ERPs have separate invoice/bill stores.

Cache the chosen ERP's id, integratedApp, and companyName for later steps and the loop.

Step 2 — Identify what the user has

Ask whether they have:

• a transaction hash (e.g. 0x…) — best case, pins the match immediately,
• an invoice/bill ID or number (numeric internal ID, or the human-facing invoiceNumber/billNumber),
• or both.

If they have neither, require at least one. If they're not sure whether their identifier is an invoice or a bill, accept it and try both lookups in Step 3.

Also accept loose forms — "INV-123", "Bill 9988", "the bill for Acme last week". The freeText filter on erpInvoices / erpBills handles these.

Before moving on, always ask for the transaction date (or an approximate date range) if they haven't given a tx hash. Ledger volume is high — date is the single most useful filter for narrowing candidates. Also offer: "If you happen to have the tx hash, paste it now — it pins the match exactly." Date matters both directions (tx→bill and bill→tx).

Step 3 — Fetch the known object and produce ranked match suggestions

There are two branches. Pick by what the user provided. If they provided both a tx hash and an invoice/bill ID, skip ahead to Step 5 (match is already determined).

Branch A — User has a transaction hash

1. Fetch the transaction with its sub-transactions:

``graphql query GetTx($hash: String!) { transaction(identifier: $hash, currency: "usd", limit: 1) { results { id identifier timestamp platform children { id amount balanceFactor isInternalTransfer fiatValue sender { identifier displayName isInternal } recipient { identifier displayName isInternal } asset { symbol identifier } } } } } ``

1. Pick the relevant sub-transaction. Skip gas, skip internal transfers, prefer the one with the user's

wallet on one side and an external counterparty on the other. Determine direction:

• balanceFactor negative → outflow → look for a bill to close.
• balanceFactor positive → inflow → look for an invoice to close.

If multiple sub-txs qualify (e.g., a swap with multiple legs), present them and let the user pick one.

1. For invoices (inflow), try backend match suggestions first — they are pre-computed and ranked:

``graphql query Suggest($subTxIds: [String]!) { subTransactionToInvoiceMatchSuggestions( subTransactionId_In: $subTxIds, minScore: 0.3, ordering: "-score", first: 10 ) { results { id score confidenceTier scoreBreakdown { txHashMatch primaryMatchFactors } invoice { id invoiceId invoiceNumber customerName origAmount balance dueDate billingStatus integration { integratedApp companyName } } } } } `` Important: this endpoint is invoice-only. For bills (outflow), skip straight to the fallback below — there is no backend bill-suggestion query exposed in the schema.

1. Fallback (zero invoice suggestions, or always for bills): query erpInvoices (inflow) or erpBills

(outflow), filtered by:

• amount window: ±20% of the sub-tx fiat value,
• date window: dateCreated_Range = [tx.timestamp − 60d, tx.timestamp] — bills/invoices are issued at or

before the payment, never after. Don't bother with future-dated invoices/bills.

• contact: if the user has named a vendor/customer, pass that as freeText. Do NOT try to match the

on-chain recipient address to an ERP contact — ERP contact identifiers (from QBO/Xero/NetSuite) are internal IDs, never wallet addresses. Some contact labels happen to resemble the vendor name in text, but the schema has no "resembles" filter, so rely on freeText over vendor/customer fields. Rank the fallback set client-side by: amount proximity → date proximity (prefer the most recent bill/invoice before the tx) → name hits in freeText output.

1. Present the top ≤5 as a numbered list (never more — too many choices is worse than too few). For each row

show: confidence tier, $ amount, customer/vendor name, invoiceNumber/billNumber, due date, and the primaryMatchFactors (e.g. amount_exact, contact_match). Best-fit first.

Branch B — User has an invoice/bill ID or number

1. Resolve the entity. Try erpInvoices first, then erpBills if it's not an invoice (or vice versa if the

user said "bill"). For invoices, include the embedded suggestions: ``graphql query GetInvoice($id: Float, $text: String) { erpInvoices(id: $id, freeText: $text, first: 1) { results { id invoiceNumber customerName origAmount balance dateCreated dueDate billingStatus integration { integratedApp companyName } suggestedSubTransactionMatches { id score confidenceTier subTransaction { id amount fiatValue asset { symbol identifier } tx { identifier timestamp platform } } } } } } ` For **bills**, query erpBills — note that bills do **not** expose suggestedSubTransactionMatches, so just fetch the bill metadata (id billNumber vendorName vendorId origAmount balance dateCreated dueDate billingStatus integration { integratedApp companyName }) and go straight to the fallback step. If id lookup returns nothing, retry with freeText: $userInput`.

1. For invoices, use the embedded suggestedSubTransactionMatches as the ranked list.
2. Fallback (zero invoice suggestions, or always for bills):

• First, ask the user for an approximate date of payment (and remind them: "if you have the tx hash,

that's fastest"). Payments happen on or after the bill/invoice date, usually within days — not before. Default to a window of [dateCreated, dateCreated + 60 days] if they can't be specific.

• Query transaction with server-side filters so we don't pull the entire ledger. Use:
• platform: <configuredNetwork or user's network> (e.g. "ethereum"),
• timestamp_Gte / timestamp_Lte → the user-supplied date window,
• children_Asset_Identifier_In: [<assetIdentifier>] → the bill's configuredAsset.identifier (e.g. the

USDC contract), or the invoice's expected asset,

• children_FiatValue_Between: "<low>,<high>" → fiat window around origAmount (e.g. ±20%). This is

a string of two comma-separated numbers. Example: "0.4,0.6" for a $0.50 bill.

• Optionally children_BalanceFactor: -1 (outflow, for bills) or 1 (inflow, for invoices).
• Use children_Amount_Between as an additional filter only when the fiat price can't be trusted

(pre-priced historical assets, etc.).

• If the resulting set is still large, ask the user for a tighter date or an amount refinement before

presenting.

1. Present up to 5 candidate transactions as a numbered list — show for each:

tx hash (truncated `0xabcd…1234`), sender → recipient addresses (same shortened form), token amount + symbol, fiat value, and timestamp. The tx hash is the user-facing identifier they recognize; the raw addresses and token amounts disambiguate when fiat alone is too ambiguous (many txs cluster near the same amount). Never show internal IDs like `subTxId` to the user — they're meaningless to them. Hold the subTxId internally for the mutation; reference the tx by its hash in all user-visible output. If no good matches, say so and ask for a tx hash — don't guess-dump a big list.

Step 4 — User picks a match (or asks for more)

Prompt: "Pick a number, or type `more` to widen the search (give me a contact name, date, or amount to focus on)."

• On a number: capture the chosen (subTxId, invoiceOrBillId, entityType) and continue to Step 5.
• On more: re-query with looser windows or with the user's added hints (contact filter, expanded date range,

amount range), present a fresh list, repeat.

• On "none of these" or similar: stop gracefully — "OK, no match made. Tell me when you have more info."

Step 5 — Configure payment account and (optional) fiat alignment

Once a match pair is chosen:

1. Deposit account (for `matchApAr.depositAccountId`). This is an ERP integration account

(the GL account in Xero/QBO/NetSuite that the bill will post against — e.g. "Crypto Wallet – ETH", "Cash and cash equivalents"). It is NOT a tres internal/wallet account. The schema FK is against schema_integrationaccount; the value lives on ErpBill.depositAccount / ErpInvoice.depositAccount as IntegrationAccountQuery { id name type value }.

Resolution order:

1. Use the pre-set one on the bill/invoice. When the entity was created in the ERP, the user

typically already picked a deposit account. Read erpBills.results[].depositAccount.id (or erpInvoices.results[].depositAccount.id). If it's non-null, use that id as depositAccountId and just mention it in the summary — no prompt.

1. Otherwise ask. Query integrationAccount(integration: <erpId>, first: 100) and show the

user the ASSET-type accounts (cash/wallet/crypto GL accounts) as a numbered list; capture the chosen integrationAccount.id.

1. paymentToInternalAccounts maps wallet → asset on the tres side and is useful for payout

workflows — it is not the source of depositAccountId for matchApAr. Don't reach for it here.

1. Fiat alignment. Compare the sub-tx fiatValue to the invoice/bill origAmount. Three cases:

• fiatValue is null or 0: ask *"This transaction has no fiat value set. Align it to ${origAmount}

so the {bill/invoice} closes fully? (y/n)"* — no default, require an explicit yes.

• fiatValue differs from origAmount: ask *"Align tx fiat value from $X to $Y so the AP/AR

closes fully? (y/n)"* — no default, require an explicit yes.

• Values match: skip.

Never run `setManualFiatValue` without a clear yes — even if the user's original request implied "close the invoice", the fiat alignment is a separate write and needs its own consent. If they say no, the match still proceeds but the entity may stay partially paid.

Step 6 — Confirm and apply

Show this summary table — use human-readable values only: tx hash (truncated), deposit account name (never its id), invoice/bill number or customer/vendor name. Internal IDs (subTxId, depositAccountId, DB entityId) are for the mutation payload, never for the user.

Ask: *"Apply these changes to {orgName}? (yes/no)"* — only proceed on yes. Treat this as approval for the match only. Fiat alignment and ERP sync each need their own explicit yes captured earlier (Step 5 for fiat, the prompt below for sync) — even if the user's opening request said "sync to QBO", still confirm here, because the request was issued before they saw the actual match details.

Then run mutations in this order, reporting the result of each:

1. Match (always):

``graphql mutation Match($entityType: String!, $matches: [SubTxMatchInput]!) { matchApAr(entityType: $entityType, matches: $matches) { results { subTxId entityId depositAccountId } } } `` Variables:

• entityType: lowercase — "invoice" or "bill". Uppercase values are rejected with

Invalid entity_type.

• matches[].subTxId: string (e.g. "4548595830").
• matches[].entityId: integer — the DB id of the invoice/bill (from erpInvoices.results[].id

/ erpBills.results[].id), NOT the human-facing invoiceNumber/billNumber. Passing a string errors with "Int cannot represent non-integer value"; passing a bill number looks up nothing and errors with "One or more Bill IDs not found".

• matches[].depositAccountId: integer — the integration account id (see Step 5).

1. Fiat alignment (only if user opted in):

``graphql mutation AlignFiat($id: String!, $newFiatValue: String!, $currency: String) { setManualFiatValue(id: $id, newFiatValue: $newFiatValue, currency: $currency) { success } } ` Pass the sub-transaction ID as id, the target value as a string in newFiatValue (e.g. "123.45", the invoice/bill origAmount), and currency: "usd"`. If the response returns an error about a "locked period", the accounting period containing the tx is closed — surface that to the user; don't try to force.

1. ERP sync (ask first, every time): "Sync this transaction to {erpName} now? (y/n)" — **never sync

without an explicit yes at this point**, even if the user's original ask included "and sync to QBO". The sync call is visible in the ERP and harder to undo than the match itself, so it gets its own gate. If yes: ``graphql mutation Sync($txIds: [String]!, $entityType: String!) { syncSpecificTransactions(transactionIds: $txIds, entitySourceType: $entityType) { status } } ` entitySourceType mirrors entityType` from the match call.

When reporting results back to the user, refer to the transaction by its tx hash (truncated) and the deposit account by name — never by subTxId or depositAccountId. Those internal IDs are for debugging, not user output.

If any mutation errors, surface the error message verbatim and offer to retry. The match is the one that actually links the records — if it fails, the rest is moot.

Step 7 — Loop

Ask: "Match another transaction or invoice/bill? (y/n)"

• yes → restart from Step 2 (skip Step 1 — keep the cached ERP from this session).
• no → wrap up: "Done. Closed {n} item(s) this session."

Verified TRES MCP operations used

Out of scope (politely redirect if asked)

• Bulk matching (many ↔ many) — not yet supported by this skill.
• Sending payments (sendBillPayment, sendInvoicePayment) — separate workflow.
• Connecting or revoking the ERP itself — use the tres-settings-management skill.
• Explaining a transaction in narrative form — use tres-tx-story.
• Importing an explorer link into the ledger — use tres-explorer-tx-to-ledger.