Jakob WennbergandClaude Fable 5
650c7be5e1
* fix(bookkeeping): revive counterparty template learning, dead since the multi-tenant refactor (#865) The learning half of counterparty templates has written nothing since 2026-03-30 (prod: 750 SIE imports, zero new templates). Two stacked bugs: - The multi-tenant refactor re-scoped categorization_templates to company_id and the lib stopped writing user_id, but user_id kept its NOT NULL: every insert failed with a null violation that supabase-js returns rather than throws, so nothing was ever logged. Migration 20260711100000 drops the NOT NULL and the dead user_id indexes. - Four of six learning call sites (both categorize routes, categorize-core, the MCP server) passed the auth user id as companyId, so even with the column fixed the writes would fail FK/RLS and corrections could never find the template they were correcting. Hardening while in here: - insertOrUpdateTemplate now checks every write result, logs failures, and returns whether a row was written; populateTemplatesFromSieVouchers reports only templates actually persisted. - Sign-mismatched matches (an incoming refund matching an expense-learned template) previously booked backwards: debit expense / credit bank for money coming IN. They are now mirrored into the correct refund shape (VAT leg reversed for deductible input VAT), flagged requires_review, and excluded from template/rule learning so a refund can never flip a learned template. - Template amounts are computed from the SEK-resolved amount, so foreign-currency transactions no longer produce unbalanced multi-line entries (or VAT computed on foreign units). - SIE extraction no longer hardcodes 25% for 2641 (rate-agnostic in BAS): the rate is inferred from voucher amounts and snapped to 25/12/6%, and reverse-charge counterparties learn vat_treatment='reverse_charge' instead of losing the RC legs (which also no longer poison the ratio base). - New pg-real test locks the exact insert column set against the real schema, so a schema/code drift like this can't ship green again. Closes #865 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(bookkeeping): mirror fiktiv-moms legs on RC credit notes, exclude import VAT accounts from ratio base Compliance-review follow-ups on #989: - REVERSE_CHARGE_VAT_ACCOUNTS gains the import output-VAT accounts (2615/2625/2635), which pair with 2645 in import vouchers exactly like the RC pairs and must not shrink the business ratio base. - A sign-mismatched match against a reverse_charge template (an RC supplier's credit note) now mirrors both fiktiv legs (credit 2645 / debit 2614) instead of booking gross, so Ruta 30/48 net back to zero. The income line-builder nets VAT credits against debit legs to keep the mirrored pair balance-neutral (identical result for all existing credit-only output-VAT paths). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(types): CategorizationTemplate.user_id is nullable since 20260711100000 (CodeRabbit) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(bookkeeping): use roundOre for the VAT netting, keep the ore-round ratchet at baseline Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> * fix(bookkeeping): review-gate stale 12% templates across the livsmedel transition, pattern-aware direction guard Compliance-review round 2 on #989: - Livsmedel VAT dropped 12% -> 6% on 2026-04-01 (Prop. 2025/26:55) while restaurang/hotell stay at 12%. A reduced_12 template whose last_seen_date predates the transition can no longer be trusted unreviewed: its match is flagged requires_review until a post-transition approval refreshes it (re-approval keeps 12%, a correction relearns 6%). Actively-confirmed 12% counterparties flow without friction. - The opposite-direction correction guard now falls back to the line pattern's business sides when the legacy fields are both settlement-ish and cannot classify a multi-line template. - Documented the accepted import-RC mirroring limitation (2614 vs 2615 ruta attribution) and the netted-vatCredit precondition. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> --------- Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
Accounted MCP server
JSON-RPC 2.0 server exposing the Accounted bookkeeping engine to MCP clients (Claude Desktop, Claude Code, etc.). Endpoint: /api/extensions/ext/mcp-server/mcp. OAuth and stdio bridge live alongside the API surface: see app/api/mcp-oauth/ and packages/gnubok-mcp/.
Tool authoring contract
Enforced by tests in __tests__/: these are not style preferences, they're guard rails.
additionalProperties: falseon everyinputSchema. Guarded bystrict-schemas.test.ts. Forces clear rejections on hallucinated fields instead of silent ignores.- Descriptions ≤ 280 chars. Guarded by
output-schema.test.ts. NoArgs:/Returns:/Examples:prose: those belong in JSON Schema. Use agent-native hints ("Use to…", "Call X first", "HIGH risk"). - Staged-operation envelope for write tools:
outputSchema: STAGED_OPERATION_SCHEMA(server.ts). Fields:staged, risk_level, actor, message, preview, period_status?, next?. Thestaged: trueboolean is the explicit completion signal; agents must not infer completion from prose. Do NOT introduce a parallel{ success, shouldContinue, output }envelope. period_statusthreading: any tool that ties to a fiscal-period-bound date (categorize, mark paid, create voucher, correct/reverse entry, approve supplier invoice) passesdateForPeriodChecktostagePendingOperation. Response then includesperiod_status: { period_id, status: open|locked|closed, lock_date }so widgets and agents disable writes without round-trips.- Scope mapping: every new tool needs an entry in
lib/auth/api-keys.tsTOOL_SCOPE_MAP. Missing entries default to deny. - Tests for new write tools: add staging-gate coverage to
__tests__/voucher-tools.test.ts(or a sibling) plus executor coverage tolib/pending-operations/__tests__/voucher-executors.test.tsif the tool stages a newoperation_type.
Determinism / cache stability
Tool definitions (name, description, inputSchema, outputSchema, annotations) are declared as static object literals at module load: no timestamps, no UUIDs, no Date/Math.random in the definition layer. This makes the tools/list JSON payload byte-stable across requests, which lets agent-side prompt caches stay warm. Do not introduce per-request non-determinism into the definitions block. Anything time-bound or random belongs inside execute().
For internal Anthropic API usage (today only extensions/general/invoice-inbox/lib/extract-invoice-fields.ts): annotate stable prefixes with cache_control: { type: 'ephemeral' } and log usage.cache_read_input_tokens for hit-ratio observability. The 1h TTL from the agent-native API plan (item 10) requires the direct Anthropic API; Accounted's Bedrock path defaults to a shorter TTL.
Payload-size watchdog
payload-size.bench.test.ts enforces a tools/list JSON payload ceiling (currently 36,000 tokens, bumped from 32,000 when top-level Tool.title landed on all tools for Claude Connectors Directory readiness). If the test fires, the right answer is rarely "raise the ceiling". Instead, trim descriptions or leverage gnubok_search_tools (already deployed; tool definitions can defer to it for discovery rather than enumerating in tools/list).
Where things live
server.ts: the tools array + JSON-RPC dispatchertool-result.ts:withNext(),toToolError()response helpersresources/: read-onlyAccounted://URIs (active company, period, recent activity, capabilities, attention items, voucher gaps, chart of accounts, VAT treatments)widgets/: inline HTML widgets (receipt-matcher, vat-review)prompts/: slash-command-style promptsskills/: domain-knowledge skill bodies served viagnubok_load_skill__tests__/: strictness guards + per-tool coverage