Files
accounted/lib/events/types.ts
T
Jakob Wennberg 10bbfb9d79 fix(mcp): make agent failures diagnosable from the log, not just from the source (#2087)
* fix(mcp): make agent failures diagnosable from the log, not just from the source

Mining event_log for mcp.tool_called: VALIDATION_ERROR ran at about one a day
until 2026-08-25, then jumped to a hundred a day and stayed there. One
integration's gnubok_get_kpi_report has been refused 604 times over seven days
and is still failing.

The cause is the unknown-parameter guard from #1856, which is correct and
stays. The caller is even told precisely what is wrong: getStructuredError puts
"Unknown parameter "x" for <tool>. Valid parameters: ..." into message_en,
which the agent receives.

What was broken is what we recorded about it. Two things, both cheap:

- Telemetry logged only message_sv, which for VALIDATION_ERROR is the registry
  default "Förfrågan innehåller ogiltiga uppgifter." and names neither the
  parameter nor the tool. A seven-day outage was indistinguishable from a typo,
  and the cause was only findable by reading the dispatcher source. errorDetail
  now carries message_en, stored only when it differs from errorMessage, so the
  many domain failures whose message_sv is already the specific text cost
  nothing extra.
- errorKind said 'company_access_denied' for all 604, because the arg guard
  throws inside the company-routing try. That is an active misdirection: it
  sends triage looking for a tenancy bug that does not exist. Exactly two
  things in that block raise VALIDATION_ERROR, the unknown-parameter guard and
  a malformed company_id, and neither is a permissions failure, so they now log
  as 'invalid_arguments'.

Tests reproduce the production call shape and were verified to fail when either
fix is reverted.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* test(mcp): derive the telemetry payload type from the event contract

Two review findings, both valid.

The local ToolCalledPayload interface was a hand-maintained duplicate that
omitted sessionId and half the errorKind union. That is not cosmetic: it is why
errorDetail could be added to the emitter and to lib/events/types.ts while the
test file type-checked against a stale shape. Deriving it from
EventPayload<'mcp.tool_called'> removes the drift.

The null-detail assertion was also conditional, so it would have passed on a
wrong-but-present value. Replaced with two determinate cases: a scope denial
carries both languages without duplicating either, and the unknown-tool exit,
which supplies no diagnostic, stores null.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

* test(mcp): pin the scope name in the diagnostic assertion

'A different string' passes on any placeholder. The reason errorDetail is worth
storing is that it names what the caller lacks, so assert that.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Jakob Wennberg <311770904+jakobwennberg-oss@users.noreply.github.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-31 22:24:28 +02:00

387 lines
24 KiB
TypeScript

import type {
JournalEntry,
Invoice,
Transaction,
Customer,
Supplier,
Article,
FiscalPeriod,
DocumentAttachment,
Receipt,
CreditNote,
ReconciliationMethod,
InvoiceInboxItem,
SupplierInvoice,
} from '@/types'
// ============================================================
// Core Event Types: discriminated union of all system events
// ============================================================
/**
* Who runs AI extraction on an uploaded document.
* - 'invoice-inbox': the inbox extracts and mirrors the result itself.
* - 'none': nobody should; the caller already holds the booking.
* Unset: the document-extraction extension extracts (default).
*/
export type DocumentExtractionOwner = 'invoice-inbox' | 'none'
export type CoreEvent =
// Bookkeeping
| { type: 'journal_entry.drafted'; payload: { entry: JournalEntry; userId: string; companyId: string } }
| { type: 'journal_entry.committed'; payload: { entry: JournalEntry; userId: string; companyId: string } }
| { type: 'journal_entry.corrected'; payload: { original: JournalEntry; storno: JournalEntry; corrected: JournalEntry; userId: string; companyId: string } }
| { type: 'journal_entry.reversed'; payload: { originalEntry: JournalEntry; reversalEntry: JournalEntry; userId: string; companyId: string } }
| { type: 'journal_entry.deleted'; payload: { entryId: string; voucherSeries: string; voucherNumber: number; userId: string; companyId: string } }
// Documents
// extractionOwner: set by the invoice inbox on documents it extracts itself,
// so the document-extraction extension yields instead of racing it (the
// inbox row does not exist yet when this event fires inside uploadDocument).
// 'none' is an explicit opt-out: the uploader already knows the booking
// (provider underlag import links each file to its posted verifikat), so
// running a paid model over it would buy nothing. The extension stamps the
// row as skipped instead of extracting.
| { type: 'document.uploaded'; payload: { document: DocumentAttachment; userId: string; companyId: string; extractionOwner?: DocumentExtractionOwner } }
| { type: 'document.accessed'; payload: { document: { id: string; file_name: string }; userId: string; companyId: string } }
| { type: 'document.deleted'; payload: { document: { id: string; file_name: string }; userId: string; companyId: string } }
// Invoicing
| { type: 'invoice.created'; payload: { invoice: Invoice; userId: string; companyId: string } }
// Hard delete of an un-finalized, unnumbered draft (no F-series number was
// consumed). Carries only the identifiers (the row is gone) so the audit
// log can record who removed which draft and when. Numbered drafts are
// makulerade instead and surface via the journal, not this event.
| { type: 'invoice.draft_deleted'; payload: { invoiceId: string; userId: string; companyId: string } }
| { type: 'invoice.sent'; payload: { invoice: Invoice; userId: string; companyId: string } }
| { type: 'invoice.paid'; payload: { invoice: Invoice; paymentAmount: number; paymentDate: string; userId: string; companyId: string } }
| { type: 'credit_note.created'; payload: { creditNote: CreditNote; userId: string; companyId: string } }
// Recurring invoices: emitted by the daily cron after a schedule spawns
// an invoice. `autoSent` tells observers whether the email also went out
// (false means it was created as draft for manual review).
| { type: 'recurring_invoice.executed'; payload: {
scheduleId: string
invoice: Invoice
autoSent: boolean
warning: string | null
userId: string
companyId: string
} }
// Banking
| { type: 'transaction.synced'; payload: { transactions: Transaction[]; userId: string; companyId: string } }
| { type: 'transaction.categorized'; payload: { transaction: Transaction; account: string; taxCode: string; userId: string; companyId: string } }
| { type: 'transaction.reconciled'; payload: { transaction: Transaction; journalEntryId: string; method: ReconciliationMethod; userId: string; companyId: string } }
// Account-keyed reconciliation (lib/reconciliation/actions.ts): one event per
// link made or removed on any reconcilable account (bank:<cash_account_id>,
// skattekonto, later manual:NNNN). `transaction.reconciled` keeps firing for
// bank links made through the bank engine; these are the kind-agnostic
// signals the flows builder triggers on.
| { type: 'reconciliation.matched'; payload: { accountKey: string; externalId: string; journalEntryId: string; method: 'manual' | 'proposal'; userId: string; companyId: string } }
| { type: 'reconciliation.unmatched'; payload: { accountKey: string; externalId: string; previousJournalEntryId: string | null; userId: string; companyId: string } }
// Sign-off: the human (or agent-staged, user-approved) assertion "reconciled
// through this date" on one account (lib/reconciliation/signoff.ts), and its undo.
| { type: 'reconciliation.signed_off'; payload: { accountKey: string; signoffId: string; throughDate: string; unexplainedDifference: number | null; userId: string; companyId: string } }
| { type: 'reconciliation.reopened'; payload: { accountKey: string; signoffId: string; throughDate: string; reason: string | null; userId: string; companyId: string } }
// Bank connection lifecycle: consent + account selection are the
// GDPR/PSD2 audit points; emitted to event_log for compliance trail.
| { type: 'bank_connection.consent_granted'; payload: { connectionId: string; bankName: string | null; accountCount: number; consentExpiresAt: string | null; userId: string; companyId: string } }
| { type: 'bank_connection.account_selection_changed'; payload: { connectionId: string; bankName: string | null; previousStatus: string; newStatus: string; enabledCount: number; totalCount: number; userId: string; companyId: string } }
| { type: 'bank_connection.revoked'; payload: { connectionId: string; bankName: string | null; userId: string; companyId: string } }
// Emitted when a new or renewed connection supersedes an older row for the
// same bank in the same company: the old row is parked as 'revoked' with
// superseded_by pointing at the replacement, and its transactions are
// re-pointed. connectionId is the SUPERSEDED (old) row, mirroring .revoked.
| { type: 'bank_connection.superseded'; payload: { connectionId: string; supersededById: string; bankName: string | null; userId: string; companyId: string } }
// Emitted when the bank/provider redirects the OAuth callback back with an
// error instead of an authorization code: denied consent, a bank-side
// failure (e.g. Handelsbanken's missing corporate fullmakt), or an expired
// signing session. Durable failure trail (issue #1716): the pending row is
// deleted right after and console logs expire, so without this event
// support cannot answer which attempt failed, with which error, on whose
// side. connectionId may reference a row deleted by the same request.
| { type: 'bank_connection.consent_denied'; payload: {
connectionId: string
bankName: string | null
psuType: string | null
errorCode: string
errorDescription: string | null
priorStatus: string
userId: string
companyId: string
} }
// Emitted when the code-for-session exchange or connection finalization
// throws after the bank redirected back successfully. Same audit doctrine
// as consent_denied: the fresh-connect row is deleted by cleanup and the
// console log expires, leaving support nothing to answer from.
| { type: 'bank_connection.finalize_failed'; payload: {
connectionId: string
bankName: string | null
reason: string
priorStatus: string
userId: string
companyId: string
} }
// Emitted when the PSD2 callback fails to mirror a returned account into
// cash_accounts. ASVS V16 / ISO 27001 A.8.15: security-relevant failures
// must land in a structured audit log (event_log, 30-day TTL) rather than
// being lost to console.error.
| { type: 'bank_connection.cash_account_mirror_failed'; payload: {
connectionId: string
bankName: string | null
accountUid: string
ledgerAccount: string
currency: string
reason: string
userId: string
companyId: string
} }
// Stripe Connect lifecycle: connect/disconnect are outward-facing consent
// transitions (a third party gains/loses access to payment data), so they
// land in event_log for the audit trail, mirroring bank_connection.*.
| { type: 'stripe.connected'; payload: { connectionId: string; stripeAccountId: string; livemode: boolean; userId: string; companyId: string } }
| { type: 'stripe.disconnected'; payload: { connectionId: string; stripeAccountId: string | null; reason: 'user' | 'revoked_upstream'; userId: string; companyId: string } }
// WooCommerce store lifecycle: same audit doctrine as stripe.* (a third
// party's API credentials are granted/dropped).
| { type: 'woocommerce.connected'; payload: { connectionId: string; storeUrl: string; userId: string; companyId: string } }
| { type: 'woocommerce.disconnected'; payload: { connectionId: string; storeUrl: string | null; reason: 'user' | 'revoked_upstream'; userId: string; companyId: string } }
// Shopify store lifecycle: same audit doctrine as stripe.*/woocommerce.*.
| { type: 'shopify.connected'; payload: { connectionId: string; shopDomain: string; userId: string; companyId: string } }
| { type: 'shopify.disconnected'; payload: { connectionId: string; shopDomain: string | null; reason: 'user' | 'revoked_upstream'; userId: string; companyId: string } }
// Periods
| { type: 'period.locked'; payload: { period: FiscalPeriod; userId: string; companyId: string } }
| { type: 'period.unlocked'; payload: { period: FiscalPeriod; userId: string; companyId: string } }
| { type: 'period.year_closed'; payload: { period: FiscalPeriod; userId: string; companyId: string } }
// Customers
| { type: 'customer.created'; payload: { customer: Customer; userId: string; companyId: string } }
// Articles (artikelregister)
| { type: 'article.created'; payload: { article: Article; userId: string; companyId: string } }
| { type: 'article.updated'; payload: { article: Article; userId: string; companyId: string } }
| { type: 'article.deleted'; payload: { articleId: string; userId: string; companyId: string } }
// Suppliers
| { type: 'supplier.created'; payload: { supplier: Supplier; userId: string; companyId: string } }
// Receipts
| { type: 'receipt.extracted'; payload: {
receipt: Receipt;
documentId: string | null;
confidence: number;
userId: string;
companyId: string;
}}
| { type: 'receipt.matched'; payload: {
receipt: Receipt;
transaction: Transaction;
confidence: number;
autoMatched: boolean;
userId: string;
companyId: string;
}}
| { type: 'receipt.confirmed'; payload: {
receipt: Receipt;
businessTotal: number;
privateTotal: number;
userId: string;
companyId: string;
}}
// Supplier Invoice Lifecycle
| { type: 'supplier_invoice.registered'; payload: { supplierInvoice: SupplierInvoice; userId: string; companyId: string } }
| { type: 'supplier_invoice.approved'; payload: { supplierInvoice: SupplierInvoice; userId: string; companyId: string } }
| { type: 'supplier_invoice.paid'; payload: { supplierInvoice: SupplierInvoice; paymentAmount: number; userId: string; companyId: string } }
| { type: 'supplier_invoice.credited'; payload: { supplierInvoice: SupplierInvoice; creditNote: SupplierInvoice; userId: string; companyId: string } }
| { type: 'supplier_invoice.uncredited'; payload: { supplierInvoice: SupplierInvoice; reversedCreditNoteId: string; reversalEntryId: string | null; userId: string; companyId: string } }
// Payment Matching
| { type: 'invoice.match_confirmed'; payload: { invoice: Invoice; transaction: Transaction; userId: string; companyId: string } }
| { type: 'supplier_invoice.match_confirmed'; payload: { supplierInvoice: SupplierInvoice; transaction: Transaction; userId: string; companyId: string } }
// Supplier Invoice Inbox
| { type: 'supplier_invoice.received'; payload: { inboxItem: InvoiceInboxItem; userId: string; companyId: string } }
| { type: 'supplier_invoice.extracted'; payload: { inboxItem: InvoiceInboxItem; confidence: number; userId: string; companyId: string } }
| { type: 'supplier_invoice.confirmed'; payload: { inboxItem: InvoiceInboxItem; supplierInvoice: SupplierInvoice; userId: string; companyId: string } }
// Salary
| { type: 'salary_run.created'; payload: { salaryRunId: string; periodYear: number; periodMonth: number; userId: string; companyId: string } }
| { type: 'salary_run.approved'; payload: { salaryRunId: string; approvedBy: string; userId: string; companyId: string } }
| { type: 'salary_run.approval_reverted'; payload: { salaryRunId: string; revertedBy: string; deletedAgiDeclarationId: string | null; userId: string; companyId: string } }
| { type: 'salary_run.booked'; payload: { salaryRunId: string; entryIds: string[]; userId: string; companyId: string } }
| { type: 'agi.generated'; payload: { agiId: string; periodYear: number; periodMonth: number; userId: string; companyId: string } }
| { type: 'agi.submitted'; payload: { salaryRunId: string; periodYear: number; periodMonth: number; userId: string; companyId: string } }
// Bolagsverket: digital inlämning av årsredovisning. Status values follow
// GUIDE §5.2.2 (arsred_inkommen → … → arsred_registrerad). `uploaded` fires
// when the iXBRL lands in eget utrymme; the undertecknare then signs the
// fastställelseintyg at Bolagsverket and the webhook drives the rest.
| { type: 'arsredovisning.uploaded'; payload: { submissionId: string; fiscalPeriodId: string; idnummer: string; environment: 'test' | 'accept' | 'prod'; userId: string; companyId: string } }
| { type: 'arsredovisning.status_changed'; payload: { submissionId: string; fiscalPeriodId: string | null; previousStatus: string; status: string; bolagsverketStatus: string; userId: string; companyId: string } }
| { type: 'arsredovisning.registered'; payload: { submissionId: string; fiscalPeriodId: string | null; userId: string; companyId: string } }
| { type: 'arsredovisning.forelagd'; payload: { submissionId: string; fiscalPeriodId: string | null; userId: string; companyId: string } }
// Skatteverket: Skattekonto sync
| { type: 'skattekonto.synced'; payload: { booked: number; upcoming: number; balanceSkv: number; balanceKfm: number; userId: string; companyId: string } }
| { type: 'skattekonto.balance.changed'; payload: { previousBalance: number; currentBalance: number; userId: string; companyId: string } }
| { type: 'skattekonto.transaction.upcoming'; payload: { transaktionsdatum: string; forfallodatum: string; transaktionstext: string; beloppSkatteverket: number; userId: string; companyId: string } }
| { type: 'skattekonto.connection.expired'; payload: { reason: 'REFRESH_EXHAUSTED' | 'SESSION_EXPIRED' | 'TOKEN_CORRUPTED'; userId: string; companyId: string } }
// Fired when the SKV saldo and GL 1630 sum diverge beyond the configured
// tolerance. The drift handler emails the company contact; UI surfaces a
// dashboard tile via /api/extensions/skatteverket/skattekonto/drift.
| { type: 'skattekonto.drift_detected'; payload: {
drift: number // SKV saldo - GL 1630 sum (signed)
saldoSkatteverket: number
glSum1630: number
fetchedAt: number // ms epoch from the snapshot
unbookedCount: number // skattekonto rows without journal_entry_id ≤ fetchedAt
userId: string
companyId: string
} }
// Company & account lifecycle
| { type: 'company.deleted'; payload: { companyId: string; userId: string; archivedAt: string } }
| { type: 'account.deleted'; payload: { userId: string; deletedAt: string } }
// MCP telemetry: fired from the MCP dispatcher.
// Persisted to event_log (180-day TTL for mcp.*/agent.* rows, vs 30 days for
// delivery events) for hot-tool / error-rate / latency analytics.
// Intentionally lightweight: no args, no result body, only metadata.
| { type: 'mcp.tool_called'; payload: {
tool: string // e.g. 'gnubok_create_invoice'
requiredScope: string | null // from TOOL_SCOPE_MAP, null if unscoped
actorType: 'user' | 'api_key' | 'mcp_oauth' | 'cron' | 'anonymous'
actorId: string | null // api_key id, oauth client, etc.
actorLabel: string | null // human-readable actor label
latencyMs: number // wall-clock time inside execute()
success: boolean // true iff the tool returned without throwing AND was invoked (not denied)
isError: boolean // matches the JSON-RPC tool-result isError flag returned to the client
errorCode: string | null // structured error code from tool-result.toToolError when applicable
errorKind: 'execution' | 'scope_denied' | 'capability_denied' | 'company_access_denied' | 'invalid_arguments' | 'unknown_tool' | 'test_key_write_blocked' | 'bridge_refused' | null
// bridge_refused: gnubok_call_tool was pointed at a write tool, or at nothing.
// invalid_arguments: the call never reached the tool because its arguments
// were rejected (unknown parameter, malformed company_id). Split out of
// company_access_denied, which used to swallow both and send triage
// looking for a permissions problem that did not exist.
errorMessage: string | null // human-readable error message (truncated to 500 chars), null on success.
// Raw material for clustering real agent failures into curated gotchas:
// errorCode alone can't distinguish "period locked" from "unbalanced".
errorDetail: string | null // The specific English diagnostic, when message_sv is a generic registry
// default that says nothing (VALIDATION_ERROR -> "Förfrågan innehåller
// ogiltiga uppgifter."). Null when it would only repeat errorMessage.
// Without it a 604-call outage looked identical to a typo in the logs:
// the agent was told exactly what was wrong, and we were not.
requestId: string | number | null // JSON-RPC request id (helps correlate with client-side logs)
userId: string
companyId: string
sessionId: string | null // from Mcp-Session-Id header; null if absent
client: string | null // distribution marker (X-Accounted-Client, legacy X-Gnubok-Client, or ?client=).
// Client-supplied (allow-list-sanitized): telemetry only, never identity or authz.
}}
// tools/list: informs us whether agents are using progressive discovery
// (gnubok_search_tools) or pulling the full list. Tool counts vary with
// the caller's scope set.
| { type: 'mcp.tools_list_called'; payload: {
toolCount: number // tools actually returned (post scope filter)
actorType: 'user' | 'api_key' | 'mcp_oauth' | 'cron' | 'anonymous'
actorId: string | null
actorLabel: string | null
latencyMs: number
requestId: string | number | null
userId: string
companyId: string
sessionId: string | null // from Mcp-Session-Id header; null if absent
client: string | null // distribution-channel marker; null if absent
}}
// resources/read: informs us which skills/widgets/data resources actually
// get loaded by agents. `kind` discriminates by URI scheme so we can
// GROUP BY skill vs widget vs data without parsing URIs.
| { type: 'mcp.resource_read'; payload: {
uri: string // e.g. 'Accounted://skill/month-end-close'
kind: 'widget' | 'skill' | 'data' | 'unknown'
success: boolean
errorCode: string | null
latencyMs: number
actorType: 'user' | 'api_key' | 'mcp_oauth' | 'cron' | 'anonymous'
actorId: string | null
actorLabel: string | null
requestId: string | number | null
userId: string
companyId: string
sessionId: string | null // from Mcp-Session-Id header; null if absent
client: string | null // distribution-channel marker; null if absent
}}
// Workflow lifecycle: agents declare "I'm starting month-end-close" via
// gnubok_load_skill (or implicitly by following a skill's recommended tool
// sequence). Phase 3A captures these to measure: how often is a workflow
// started? How often does it complete? Where do agents abandon?
| { type: 'mcp.workflow_started'; payload: {
slug: string // e.g. 'month-end-close'
sessionId: string | null
actorType: 'user' | 'api_key' | 'mcp_oauth' | 'cron' | 'anonymous'
actorId: string | null
actorLabel: string | null
userId: string
companyId: string
}}
| { type: 'mcp.workflow_completed'; payload: {
slug: string
sessionId: string | null
outcome: 'success' | 'abandoned' | 'failed'
stepsCompleted: number | null // null when not tracked granularly
durationMs: number | null
actorType: 'user' | 'api_key' | 'mcp_oauth' | 'cron' | 'anonymous'
actorId: string | null
actorLabel: string | null
userId: string
companyId: string
}}
// Fires on EVERY successful gnubok_load_skill: all tiers, unlike
// mcp.workflow_started which fires only for workflow-tier skills. Records
// WHICH skill/atom bodies agents actually pull, the denominator needed to
// correlate a loaded atom with downstream tool-error rates (a skill can
// make the model worse: measure, don't assume).
| { type: 'mcp.skill_loaded'; payload: {
slug: string // e.g. 'modifier/holding-ab', 'month-end-close'
tier: 'workflow' | 'horizontal' | 'vertical' | 'modifier'
sessionId: string | null
actorType: 'user' | 'api_key' | 'mcp_oauth' | 'cron' | 'anonymous'
actorId: string | null
actorLabel: string | null
userId: string
companyId: string
}}
// Fires when the agent's next tool call matches the previous response's
// nextHint.tool: measures whether `next` hints are actually followed.
// Computed dispatcher-side by comparing the last response shape to the
// current call.
| { type: 'mcp.next_hint_followed'; payload: {
fromTool: string
toTool: string
sessionId: string | null
actorType: 'user' | 'api_key' | 'mcp_oauth' | 'cron' | 'anonymous'
actorId: string | null
actorLabel: string | null
userId: string
companyId: string
}}
// Agent self-reported feedback (gnubok_feedback tool). The product team
// queries event_log for `agent.feedback` and routes to a backlog.
| { type: 'agent.feedback'; payload: {
context: string
sentiment: 'positive' | 'negative' | 'neutral'
suggestion: string | null
toolName: string | null
skillSlug: string | null
sessionId: string | null
actorType: 'user' | 'api_key' | 'mcp_oauth' | 'cron' | 'anonymous'
actorId: string | null
actorLabel: string | null
userId: string
companyId: string
}}
// ============================================================
// Helper Types
// ============================================================
/** All possible event type strings */
export type CoreEventType = CoreEvent['type']
/** Extract the payload type for a given event type */
export type EventPayload<T extends CoreEventType> = Extract<CoreEvent, { type: T }>['payload']
/** Handler function for a specific event type */
export type EventHandler<T extends CoreEventType> = (payload: EventPayload<T>) => Promise<void> | void
/** Subscription: event type + handler */
export interface EventSubscription<T extends CoreEventType = CoreEventType> {
eventType: T
handler: EventHandler<T>
}