Files
accounted/lib/errors/__tests__/get-structured-error.test.ts
T
Jakob Wennberg 250cc7c450 feat(mcp): always-explicit retryable on structured errors + transient inference (P1-1) (#875)
Agents could not distinguish 'keep retrying' from 'stop, this is
broken' (agent.feedback): retryable was emitted only when a registry
entry declared true — absent otherwise, including for genuinely
transient DB/network failures whose SQLSTATE is lost when tools wrap
them as Error('Database error: ...').

- StructuredError.retryable is now a required boolean. Registry
  declaration wins; otherwise isTransientFailure() infers from Postgres
  SQLSTATEs (40001/40P01/57014/08xxx/53xxx/55P03), upstream HTTP
  statuses (408/429/5xx), and message signatures that survive wrapping
  (deadlock, serialization, statement timeout, fetch/socket failures).
- Unclassified transient failures surface as stable code
  TRANSIENT_ERROR (new registry entry, retryable: true).
- categorize_transaction accepts idempotency_key — the tool agents
  blind-retry after client-side approval-elicitation drops; the key
  makes that retry replay-safe instead of double-staging.
- Contract documented in .claude/rules/mcp-server.md. The planned
  'kind' field was dropped: the code registry already encodes it;
  retryable is the agent-actionable bit.

Every tool error already flows through the single dispatch point
(toToolError -> getStructuredError), so coverage is universal without
per-tool migration. Full unit suite: 6575 tests green.

Part of dev_docs/mcp_optimization_plan.md (P1-1).

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
2026-07-03 10:29:58 +02:00

134 lines
5.6 KiB
TypeScript

import { describe, it, expect } from 'vitest'
import { getStructuredError } from '../get-structured-error'
describe('getStructuredError', () => {
it('extracts code from structured bookkeeping error', () => {
const result = getStructuredError({
error: {
code: 'JOURNAL_ENTRY_NOT_BALANCED',
message: 'Debits do not match credits',
details: { totalDebit: 100, totalCredit: 90 },
},
})
expect(result.code).toBe('JOURNAL_ENTRY_NOT_BALANCED')
expect(result.message_sv).toContain('balanserar inte')
expect(result.message_en).toContain('Debits')
expect(result.remediation?.description).toContain('Recalculate')
})
it('extracts code from typed error class with code property', () => {
class FakeBookkeepingError extends Error {
readonly code = 'ACCOUNTS_NOT_IN_CHART'
readonly accountNumbers = ['1930', '2641']
constructor() {
super('Accounts not in chart')
}
}
const result = getStructuredError(new FakeBookkeepingError())
expect(result.code).toBe('ACCOUNTS_NOT_IN_CHART')
expect(result.remediation?.resource).toBe('Accounted://chart-of-accounts')
})
it('infers PERIOD_NOT_LOCKED from message text', () => {
const result = getStructuredError(new Error('Period must be locked before closing'))
expect(result.code).toBe('PERIOD_NOT_LOCKED')
expect(result.remediation?.tool).toBe('gnubok_lock_period')
})
it('infers PERIOD_HAS_UNBOOKED_TRANSACTIONS from Swedish lock-error message', () => {
const result = getStructuredError(
new Error('Kan inte låsa period: 3 affärstransaktion(er) saknar bokföring.')
)
expect(result.code).toBe('PERIOD_HAS_UNBOOKED_TRANSACTIONS')
expect(result.remediation?.tool).toBe('gnubok_list_uncategorized_transactions')
})
it('produces INSUFFICIENT_SCOPE remediation with attempted scope', () => {
const result = getStructuredError(
new Error('Insufficient scope: this API key does not have the "bookkeeping:write" scope'),
{ attemptedScope: 'bookkeeping:write' }
)
expect(result.code).toBe('INSUFFICIENT_SCOPE')
expect(result.remediation?.description).toContain('"bookkeeping:write"')
expect(result.remediation?.resource).toBe('Accounted://capabilities')
})
it('infers TRANSACTION_ALREADY_CATEGORIZED', () => {
const result = getStructuredError(new Error('Transaction already has a journal entry'))
expect(result.code).toBe('TRANSACTION_ALREADY_CATEGORIZED')
expect(result.remediation?.tool).toBe('gnubok_uncategorize_transaction')
})
it('falls back to UNKNOWN_ERROR when no code or pattern matches', () => {
const result = getStructuredError(new Error('Something weird happened'))
expect(result.code).toBe('UNKNOWN_ERROR')
expect(result.remediation).toBeUndefined()
})
it('handles plain string errors', () => {
const result = getStructuredError('Period must be locked before closing')
expect(result.code).toBe('PERIOD_NOT_LOCKED')
expect(result.message_en).toBe('Period must be locked before closing')
})
it('handles null/undefined gracefully', () => {
const result = getStructuredError(null)
expect(result.code).toBe('UNKNOWN_ERROR')
expect(result.message_en).toBe('Unknown error')
expect(result.message_sv).toBeTruthy()
})
it('always returns Swedish message even with no match', () => {
const result = getStructuredError(new Error('Random gibberish XYZ'))
expect(result.message_sv).toBeTruthy()
expect(result.message_sv.length).toBeGreaterThan(0)
})
})
describe('retryable contract (always present, transient inference)', () => {
it('is explicitly false for a plain unclassified error', () => {
const result = getStructuredError(new Error('nope'))
expect(result.code).toBe('UNKNOWN_ERROR')
expect(result.retryable).toBe(false)
})
it('classifies a wrapped deadlock message as TRANSIENT_ERROR, retryable', () => {
// Tools wrap DB errors as plain strings, losing the SQLSTATE — the
// message pattern must survive that wrapping.
const result = getStructuredError(new Error('Database error: deadlock detected'))
expect(result.code).toBe('TRANSIENT_ERROR')
expect(result.retryable).toBe(true)
})
it('classifies a Postgres serialization-failure SQLSTATE as transient', () => {
const err = Object.assign(new Error('could not complete'), { code: '40001' })
const result = getStructuredError(err)
expect(result.code).toBe('TRANSIENT_ERROR')
expect(result.retryable).toBe(true)
})
it('classifies upstream 429/503 statuses as transient', () => {
expect(getStructuredError(Object.assign(new Error('slow down'), { status: 429 })).retryable).toBe(true)
expect(getStructuredError(Object.assign(new Error('bad gateway'), { statusCode: 502 })).retryable).toBe(true)
})
it('classifies fetch/socket failures as transient', () => {
expect(getStructuredError(new Error('fetch failed')).code).toBe('TRANSIENT_ERROR')
expect(getStructuredError(new Error('socket hang up')).retryable).toBe(true)
expect(getStructuredError(new Error('connect ECONNRESET 1.2.3.4:443')).retryable).toBe(true)
})
it('keeps a specific inferred code but still computes retryable from the failure', () => {
// NOT_FOUND is permanent even though nothing in the registry says so explicitly.
const result = getStructuredError(new Error('Transaction not found'))
expect(result.code).toBe('NOT_FOUND')
expect(result.retryable).toBe(false)
})
it('lets an explicit registry retryable:true win over inference', () => {
const tagged = Object.assign(new Error('over the cap'), { code: 'RATE_LIMITED' })
const result = getStructuredError(tagged)
expect(result.retryable).toBe(true)
})
})