Commit Graph

6 Commits

Author SHA1 Message Date
Jakob Wennberg b5da51ea0a feat(settings): API & MCP tab: correct connector URL namespace, plugin path, Swedish guide (#2105)
* feat(settings): API & MCP tab: correct connector URL namespace, plugin path, Swedish guide

The in-product MCP URLs omitted `tool_namespace=accounted`, and
resolveMcpToolNamespace() falls back to the legacy `gnubok_` prefix when the
param is absent. Every connection made from Settings therefore got `gnubok_*`
tool names while the docs, the accounted-api skill, and
claude-plugin/.mcp.json all reference `accounted_*`.

- Add `tool_namespace=accounted` to the Claude.ai, Claude Code, and
  Claude Desktop snippets.
- Move the Claude Desktop bridge from `npx gnubok-mcp` / `GNUBOK_API_KEY` to
  `npx -y accounted-mcp` / `ACCOUNTED_API_KEY`, and emit `ACCOUNTED_URL` so
  self-hosted and white-label instances get a config pointing at their own
  host. The `gnubok_sk_` key prefix is unchanged: it is wire format.
- Surface the Claude Code plugin, the only path that configures the
  connection and the seven workflow commands in one step.
- Rename the settings tab "API" to "API & MCP" and rewrite its intro: the
  MCP connection is what most users come here for, not API keys.
- Link the step-by-step guide from the panel, locale-aware.

Docs: add a Swedish /docs/api/anslut-claude alongside the English page (the
docs site has no locale routing, so each language is its own URL), give both
the Claude Code plugin path, and teach the export and freshness scripts about
the new page so cross-repo drift is caught.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HfZgiZmNN6qGEfgvXqxAeg

* fix(settings,docs): Cursor is not Claude Code; use the accounted_ tool name

Review follow-up on #2105.

`claude mcp add` is a Claude Code command. Cursor does not read it, so the
"Claude Code / Cursor" row and the docs sentence pointing Cursor users at that
command were both wrong (the row predates this PR; the docs sentence did not).
Cursor now gets its own row and its own `~/.cursor/mcp.json` snippet with the
`url` field, in the panel and in both docs pages.

Also `vat_close_check` -> `accounted_vat_close_check` in the reviewer test on
both pages, matching the identifier used in the prompts section above it.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HfZgiZmNN6qGEfgvXqxAeg

* feat(settings,docs): one-click Connect to Claude, and cut the panel to one action

Anthropic documents an install link for custom connectors:
https://claude.ai/customize/connectors?modal=add-custom-connector&connectorName=NAME&connectorUrl=ENCODED
(claude.com/docs/connectors/building/directory-vs-custom). It opens claude.ai
with the connector name and URL prefilled; the user still reviews and confirms,
and it grants nothing on its own. We were telling people to copy a URL and go
paste it somewhere else instead.

Settings panel, rendered and reviewed:
- "Connect to Claude" button is now the only thing above the fold. Everything
  that needs a config file or a terminal (claude.ai manual paste, Claude Code,
  the plugin, Cursor) moved into one "Other clients" disclosure, and the
  API-key methods keep theirs. Four code blocks -> one button, 1057px -> 719px.
- The connect group renders above the API-keys group. Connecting is why users
  open this tab; the tab's own intro says so.
- Each entry inside the disclosures shows its instruction as visible text.
  They were `?` HelpPopovers, so the panel read as opaque code blobs with no
  instructions on screen.
- Prose interpolates the brand's real casing, not the lowercased config key.

Docs, both languages: Path A leads with the install link and drops from five
manual steps to a link plus three short paragraphs, with the manual paste kept
under a subheading. No raw HTML: the docs renderer has no rehype-raw, so
<details> would have been silently dropped.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HfZgiZmNN6qGEfgvXqxAeg

* fix(settings,docs): correct claude mcp add syntax and the SSR install link

Review follow-up. Both findings verified before acting on them.

`claude mcp add --help` gives `claude mcp add [options] <name> <commandOrUrl>`:
the URL is positional and there is no `--url` flag, so the API-key snippet
would have failed on a missing argument. Both commands now put
`--transport http` before the name and pass the URL positionally, in the panel
and in both docs pages.

The panel is server-rendered before it hydrates and window.location has no
server equivalent, so the install link was built from a relative mcpBase in the
first paint. A click in that window would hand claude.ai a connectorUrl it
cannot resolve. The origin now resolves after mount and the anchor carries no
href until it is known, which also makes it unclickable rather than wrong.
Verified: the SSR HTML contains no claude.ai href and no relative connectorUrl,
post-hydration the href is absolute, and there are no hydration warnings.

DECISIONS.md: code-span the `gnubok_*`/`accounted_*` wildcards so they stop
rendering as emphasis, and drop the "no one-click deeplink" claim from the
earlier entry rather than leave a false statement standing two lines above its
own correction.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HfZgiZmNN6qGEfgvXqxAeg

---------

Co-authored-by: Jakob Wennberg <311770904+jakobwennberg-oss@users.noreply.github.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-09-01 16:56:19 +02:00
Mattsson 53e343ee92 Bug/invalid imports (#1146)
* feat: add Accounted MCP namespace

* fix(bookkeeping): stop flagging verifikat whose underlag lives on a referenced supplier invoice

The missing-underlag surfaces only accepted a document directly linked to
the entry, so payment verifikat for supplier invoices (doc on the
registration entry per design) and entries whose doc was pinned to the
bank transaction before matching were falsely flagged; opening the entry
showed the referenced doc and cleared the warning client-side, and it
came back on reload.

- verifikat_without_documents + transactions_without_documents now treat
  an entry as covered when a supplier invoice referencing it (registration
  or payment FK, or a supplier_invoice_payments row) carries a document
  anchored to a journal entry (BFL 5 kap 7 paragraf hänvisning till
  underlag; anchoring required because the WORM deletion guards key on
  document_attachments.journal_entry_id)
- match-supplier-invoice routes (dashboard + v1) propagate the
  transaction's pinned document onto the payment verifikat, mirroring the
  categorize route; migration backfills rows already written (open
  unlocked periods, company-guarded, never steals a linked doc)
- /api/documents/counts, the transactions-page badges, the bulk "Inget
  underlag krävs" count and the push-notification scheduler share the
  same reference-aware predicate, so every surface agrees with the RPC
- counts route validates journal_entry_ids as UUIDs (they are
  interpolated into a PostgREST or-filter)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(transactions): align table columns flush with page edges

Collapse the checkbox gutter column to zero width and hang the
hover-revealed checkbox/expand chevron in the page margins, drop the
outer padding so DATUM sits flush left and STATUS flush right, and
tuck the overflow-menu dots under the middle of the STATUS header.
Applied to both the inbox and history tables so they stay identical.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(arsredovisning): tie anlaggningstillgangar note to booked depreciation

The ARL 5:8 roll-forward note recomputed depreciation from its own
day-based linear formula (365.25/12 month length, non-inclusive day
count, linear only), drifting ~20 kr per year per asset from the
ledger-driven resultat- and balansrakning and misstating non-linear
methods entirely. Note figures now come from posted
depreciation_schedules rows (the same source disposeAsset reverses),
falling back to the engine's computeAnnualDepreciation when nothing is
posted; pre-onboarding opening balances iterate prior years through
the engine. Adds a note-vs-trial-balance tie-out warning (accounts
1000-1299, over 1 kr) surfaced before download.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* refactor(stripe): move connect and sync surface from settings to import page

Stripe's transaction feed is a continuous import source in the same
category as the PSD2 bank connection, so its connect/sync surface now
lives on the import page as a source card (mode=stripe), gated
"kommer snart" on hosted like before; self-hosted keeps the full panel.

- Import page: Stripe card after Koppla bank, renders the existing
  StripeSettingsPanel via the settings-panel registry
- OAuth callback and panel cleanup return to /import?mode=stripe
- Settings > Betalningar retired: nav item removed, route redirects,
  PaymentsSettingsContent deleted, legacy ?tab=payments mapped
- New import.stripe_* strings in sv+en; dead settings_nav.payments removed

Crons and sync logic unchanged; payment-link settings stay in the
invoicing section.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

* fix(underlag): paginate missing-underlag cron and harden doc-surface queries

Resolve PR review findings on bug/invalid-imports:
- notification-scheduler: fetchAllRows on all 5 global reads; past 1000 rows
  the capped reads produced false "saknade underlag" notifications
- bulk-missing: LOOKUP_CHUNK 300->150 so the twice-embedded .or() id list
  stays under the PostgREST URL limit
- bulk-missing + transactions page: UUID-guard the .or()-interpolated id
  lists, matching documents/counts
- match-supplier-invoice (dashboard + v1): log documentId/journalEntryId on
  the non-fatal doc-link warning
- well-known/oauth-protected-resource: document the tool_namespace allow-list
- messages/en: reword stripe_description
- DECISIONS.md: record the asset ibAck tie-out and Tailwind !important calls

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(tic): convert registrationDate from Unix seconds to millisecond epoch in lookup and profile tests

---------

Co-authored-by: Claude Fable 5 <noreply@anthropic.com>
2026-07-24 15:03:50 +02:00
Jakob Wennberg ec27228a8e style: remove em/en dashes repo-wide, add CLAUDE.md rule against them (#890)
Em dashes (—) and en dashes (–) had spread across comments, docs, tests,
and a few UI strings, reading as AI-generated boilerplate rather than
house style. Replaced each with punctuation matching its context: colon
for explanatory clauses, comma for asides, plain hyphen for numeric/legal
ranges (e.g. "21-23§"), "to"/"till" for date ranges, parentheses for
paired-dash asides. messages/en.json and messages/sv.json were fixed by
hand together to keep sv/en in sync.

Left untouched where the dash is the functional subject rather than
decorative punctuation: date-range-parser.ts's separator regex,
charset-repair.ts's CP1252 byte-mapping table (and its test), the SIE
encoding mojibake docs, generic-csv.ts's minus-sign normalizer, the
agent system-prompt files that already instruct against em dashes, and
a golden iXBRL test fixture compared byte-for-byte.

Also fixes two bugs surfaced along the way: an off-by-one in
ApiKeysPanel's scope-label split (a leftover from an earlier partial
pass), and a charset-repair test that had lost the literal en-dash it
exists to verify.

Regenerated the agent atom seed migration (skills:generate) since 27
SKILL.md files changed. Added a CLAUDE.md rule against em/en dashes,
with an explicit carve-out for the functional-dash cases above.

Co-authored-by: Claude Sonnet 5 <noreply@anthropic.com>
2026-07-04 15:58:06 +02:00
Jakob Wennberg a894af11d0 feat(mcp): tool titles + Connect-with-Claude docs for Connectors Directory readiness (P0-4) (#682)
* feat(mcp): tool titles + Connect-with-Claude docs for Connectors Directory readiness

Add a top-level Tool.title (MCP spec 2025-06-18) to all 92 MCP tools and
flow it through the tools/list serializer for the Claude Connectors
Directory. Trim the ~10 longest tool descriptions toward 180-200 chars
(semantics + agent hints preserved) to partly offset the added bytes, and
raise the payload-size bench ceiling 32K → 36K with headroom reserved for
upcoming Skatteverket tools.

Ship a "Connect with Claude" docs page (OAuth 2.1 connector + npx
gnubok-mcp stdio bridge, sample sandbox prompts, 10-minute reviewer test,
support pointer), wire it into the docs nav, and add a README to the
packages/gnubok-mcp npm bridge. Bump one categorized sandbox business
expense above 4 000 kr so the VAT close-check demo surfaces a high-value
receipt-less expense.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(mcp): restore six spaces lost in the description-trimming pass

Review finding (Greptile P2): the trim deleted single spaces in six tool
descriptions ("Stages foruser", "länkatill", "kundfordran(1510)", …),
producing malformed text in tools/list. All six were pure space deletions
with no token gain — restored verbatim. Swept all 92 descriptions for
residual merge artifacts; the two remaining matches are pre-existing
function-call notation from main.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(compliance): correct BFL citation in seed + mark 4000 kr as app heuristic

Swedish compliance review findings:
- The seed comment cited BFL 5 kap 6§ as requiring verifikat on high-value
  affärshändelser; 6§ governs verifikationens innehåll. Rephrased: BFL 5 kap
  6-7§ require every affärshändelse to be documented — the 4 000 kr cut-off
  is the tool's own heuristic, not a statutory threshold.
- The connect-claude docs read as though 4 000 kr were a legal floor
  (conflatable with the förenklad faktura ceiling, ML 17 kap 26-28§).
  Clarified that BFL requires underlag regardless of amount.

Dismissed as by-design: the gnubok_sk_test_... README placeholders
(standard practice; the prefix schema is already public in llms.txt).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
2026-06-06 11:47:21 +02:00
Jakob Wennberg 20989379bb feat(sandbox,branding): prod-parity demo with AI gating + accounted rebrand (#585)
* feat(sandbox,branding): prod-parity demo with AI gating + accounted rebrand

Sandbox now ships with seeded suppliers, supplier invoices, an asset,
a verified agent_profile, and pending operations so the demo company
exercises every prod surface. Server-side `guardSandbox()` short-
circuits any AI or paid-external API call (Bedrock chat/composer,
Resend invoice send, Riksbanken FX, VIES, etc.) and the AgentSheet
swaps in a SandboxAgentPreview that explains what's gated and offers
a register CTA. DashboardContent no longer mounts the
NewUserChecklist when the agent is already built, fixing the path
that let sandbox users still trigger /onboarding/agent.

Visible branding flips from Gnubok to Accounted: new BrandWordmark
component (Hedvig Letters Serif 700), new app/icon.png + PWA icons
generated from the accounted icon, default appName updated. URLs,
header names, API key prefixes, hostnames, and event/cookie/
localStorage keys keep `gnubok` — the rebrand is visual only.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(sandbox): hardcode supplier-invoice arrival numbers in seed

get_next_arrival_number is MAX(arrival_number) + 1 against the same
table we're about to insert into. Calling it twice before either row
lands made both calls return 1, which then violated the
(company_id, arrival_number) unique index — POST /api/sandbox/seed
500'd on first sandbox start.

The seeded company is brand new in this branch so 1 and 2 are
guaranteed unused; hardcoding side-steps the race entirely.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(sandbox): set paid_amount=0 on unpaid supplier invoice

PostgREST normalizes the column set across rows in a bulk insert, so
the second supplier invoice (Espresso House, status=registered) was
being sent with paid_amount=null because the first row (Telia, paid)
set it. supplier_invoices.paid_amount is NOT NULL DEFAULT 0; the
default only kicks in when the column is *absent* from the payload,
not when it's explicitly null. Set it inline to side-step the
normalization.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(sandbox): set actor_type=agent_chat on seeded pending_operations

pending_operations only allows user-scoped INSERTs via the
`pending_operations_chat_insert` policy, which requires
actor_type='agent_chat' alongside auth.uid()=user_id +
company membership. The seed was inserting with the default
actor_type='user', tripping the RLS check.

Also lift risk_level from preview_data (where it was unused) onto
the row itself, matching the column added in
20260430120000_pending_operations_actor_and_risk.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(pr-review): address PR #585 review feedback

Fixes called out by the core-only CI check, Greptile, and the
compliance + Swedish-accounting bots:

- AGI Programnamn pinned back to 'gnubok' (CI blocker). The XML
  Skatteverket receives must keep the stable software identifier
  regardless of the visual rebrand — same rule as the v1 health
  endpoint's `service: 'gnubok'` literal.
- handleCreateAccount in SandboxAgentPreview + ChatEmptyState now
  wraps signOut() in try/catch so a transient Supabase failure
  doesn't strand the user on a dead button (greptile P2 × 2).
- /api/currency/rate hard-fails on missing companyId instead of
  conditionally skipping the sandbox guard (greptile P2 / compliance
  V8.2.1).
- topUpSandboxAdditions now delegates to ensureSandboxAgentProfile;
  the assistant persona lives in exactly one place across the seed,
  layout backfills, and top-up path (greptile P2 outside-diff /
  compliance SOC2 CC6.1).
- ensureSandboxAgentProfile drops the userId param and sets
  verified_by_user_id to NULL — synthetic seed data should not
  attribute verification to a real user (compliance V8.2.1 /
  GDPR Art. 25(2)). Errors now logged via the structured logger
  instead of being silently swallowed (V16).
- Sandbox seed swaps real-world company names (Telia, Espresso
  House) for clearly-synthetic Demo-prefixed brands using the
  5559... documentation org-number range (compliance A.8.33).
  Asset cost bumped 24 000 → 35 000 SEK so the demo clears the
  förbrukningsinventarier threshold and illustrates capitalization
  unambiguously (swedish-asset-accounting).
- Representation pending-operation preview corrected: VAT label
  fixed from 6% → 12%, and input VAT split between the avdragsgill
  (2641) and ej-avdragsgill (5811) portions to match
  swedish-vat / ML 8 kap rules (swedish-vat).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(pr-review): seed preview consistency + AGI Programnamn constant

Two last review-bot items before merge:

- Sandbox seed: the representation pending-operation preview was
  splitting the 240 SEK café meal 60/180 between 5810 and 5811,
  which is wrong for a single attendee under the 300 SEK / person
  avdragsgill cap (ML 8 kap) — the entire amount is fully
  avdragsgill in that case. Collapse the preview to a single 5810
  + 2641 + 2440 entry so it matches the supplier_invoice_items row
  1:1 and stops teaching demo users an incorrect bookkeeping
  pattern.
- Hoist the AGI Programnamn 'gnubok' literal into a named constant
  with a comment pointing to potential future Skatteverket vendor
  registration (per the swedish-compliance bot's nit).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(sandbox): avoid BFL duplicate-verification on pending op + fix VAT cap comment

Swedish-compliance bot caught two final nits:

- The pending operation for the Demokafé representation was using the
  same supplier_invoice_number as the already-seeded supplier_invoices
  row (88245). If the sandbox user approved the staged operation, the
  insert would have created (or attempted) a duplicate verification —
  BFL 5 kap. requires each affärshändelse be recorded exactly once.
  Swap the staged operation's invoice number to a distinct value
  (INKOMMANDE-2026-001) so approval cleanly creates a new row.
- The preview comment described the 300 SEK threshold as an
  "avdragsgill cap". The actual rule (ML 8 kap. 9 §) caps the
  deductible VAT at 25 % × 300 SEK × antal_personer = 75 SEK per
  person — the 300 SEK is the tax base, not the total. Math here is
  correct either way, but the comment now states the correct formula
  so future seed edits don't propagate the wrong understanding.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-28 13:19:41 +02:00
Jakob Wennberg 3912c74a7b feat(api): Phase 6 PR-2 — docs polish (Stripe-inspired) (#497)
* feat(api): Phase 6 PR-2 — docs polish (Stripe-inspired)

Ships the developer-facing documentation surface for the v1 REST API.
Mirrors Stripe's structure (landing → cookbooks → concepts → reference
→ errors → changelog) at /docs/api with a sticky-sidebar layout in the
gnubok editorial-monochrome aesthetic. Every page is also served as
plain Markdown via a sibling .md URL so agents and LLM crawlers can
ingest the same content without HTML parsing — the existing /llms.txt
already promised /docs/api references that this PR makes real.

Single source of truth for endpoint metadata is the existing Zod
registry (lib/api/v1/registry.ts). The reference pages auto-generate
from it: adding a new endpoint surfaces in the docs on the next build
with no manual sync. The error reference pulls directly from
lib/errors/structured-errors.ts STRUCTURED_ERRORS.

CONTENT LAYER (lib/docs/):

- content/landing.ts — introduction, auth, base URL, response envelope,
  the four core principles (dry-run, idempotency, strict-mode, inline
  audit), pointers to every other section.
- content/versioning.ts — versioning + deprecation policy (Stripe dated
  format), idempotency, dry-run, strict-mode write semantics, inline
  audit blocks.
- content/webhooks.ts — webhook concept guide. Full Node.js (express +
  crypto) and Python (Flask + hmac) signature-verification samples that
  match lib/webhooks/signing.ts exactly. Lifecycle, event-type
  catalogue, payload shape, request headers, common pitfalls,
  auto-disable behaviour, audit + retention.
- content/errors.ts — generated from STRUCTURED_ERRORS. Groups by
  domain (generic, bookkeeping, periods, invoices, supplier-invoices,
  transactions, reports, imports, documents, salary, company, provider).
  Every code is anchorable so the docs_url field on every error
  envelope finally points somewhere real.
- content/reference.ts — generated from listEndpoints(). Groups by
  resource (companies, customers, invoices, suppliers,
  supplier-invoices, transactions, journal-entries, fiscal-periods,
  accounts, documents, employees, salary-runs, reports, imports,
  compliance, webhooks, operations, voucher-gap-explanations,
  reconciliation). Each endpoint section: summary, description,
  useWhen, doNotUseFor, pitfalls, scope, idempotent/reversible/dry-run
  flags, request + response examples.
- content/changelog.ts — initial entry for API version 2026-05-12
  covering every endpoint shipped in Phases 1-6. Lists what's coming
  in Phase 6 PR-3 (hardening + remaining cookbooks).
- content/cookbook/quickstart.ts — five-minute send-your-first-invoice
  guide. Demonstrates auth, dry-run, idempotency, audit-block patterns
  in one continuous narrative.
- content/cookbook/webhooks.ts — end-to-end webhook setup, sig
  verification, retry handling, idempotency on receiver side, replay
  patterns, auto-disable behaviour. Companion to the concept page.
- content/cookbook/index.ts — recipe registry. 4 placeholder recipes
  (ingest-bank-transactions, file-vat-declaration, run-payroll-and-agi,
  year-end-closing) link to their reference pages with a "coming after
  Phase 6 PR-3 hardening" note. Narrative cookbook quality benefits
  from a focused pass after the substrate stabilises.
- nav.ts — single source of truth for the sidebar nav, used by the
  layout AND the landing-page resource grid.
- markdown.tsx — shared <DocsMarkdown> component using react-markdown
  (already a dep) with Hedvig serif headlines, Geist mono code blocks,
  hairline section borders, paper-white surfaces — same editorial
  aesthetic as the dashboard.

LAYOUT (components/docs/DocsLayout.tsx):

Two-column sticky-sidebar layout. Top header carries the gnubok mark
+ section links (API reference, Cookbooks, Errors, Changelog,
openapi.json). Sidebar groups: Getting started, Cookbooks, Concepts,
API reference, Reference. Active page highlighted with the same
warm-beige bg the dashboard sidebar uses.

ROUTES (app/docs/api/, app/llms-full.txt/):

- /docs/api → landing
- /docs/api/errors → error reference
- /docs/api/webhooks → webhook concept
- /docs/api/versioning → versioning + idempotency + dry-run
- /docs/api/changelog → release notes
- /docs/api/reference → resource overview
- /docs/api/reference/[slug] → per-resource pages (19 resources, all
  generated from the registry; generateStaticParams listed)
- /docs/api/cookbook/[slug] → recipe pages (8 entries, 2 fully written
  + 6 aliases/placeholders)
- /llms-full.txt → everything concatenated for one-shot LLM ingestion

Every page has a sibling .md route (e.g. /docs/api/errors.md) serving
the raw Markdown for agents — same content, no HTML wrapper, same
5-min cache. Honours the existing /llms.txt promise that "every .md
URL under /docs/api is served as plain Markdown".

CI GUARD (lib/api/v1/__tests__/spec-snapshot.test.ts):

Vitest snapshot test that locks down (a) the endpoint count, (b) the
sorted set of method+path keys, (c) the set of distinct scopes
referenced. CI fails if any drift unexpectedly so a Zod-schema change
can't ship a silent API break — when you intentionally add/remove an
endpoint, run with -u to refresh the snapshot, review the diff, and
commit alongside the route change. The snapshot diff itself is a
self-describing changelog entry.

Initial snapshot: 100 endpoints, 17 distinct scopes, full key set
sorted. Fourth assertion in the test guarantees every endpoint
declares the agent-facing metadata (summary, description, useWhen,
doNotUseFor, pitfalls, example) the reference pages depend on — so a
registerEndpoint call that omits any of these fields is caught at CI
time rather than rendering an empty section in the docs.

INFRA TOUCH (lib/api/v1/load-routes.ts):

Added the 5 Phase 6 webhook route imports so the registry includes
them on the docs builders' path. Required for the /docs/api/reference/
webhooks page to render. The webhook routes' registerEndpoint calls
already exist; this just side-effect-imports them where the spec
generator can see them.

Coming in Phase 6 PR-3 (hardening — separate PR):

- 90-day TTL cleanup cron for non-accounting webhook deliveries
- claim_due_webhook_deliveries SQL function (FOR UPDATE SKIP LOCKED)
- Per-route rate limits on :test, :retry, webhook :create
- V16 audit-log on webhook lifecycle events
- DNS-rebinding pinned-IP HTTPS agent
- Integration tests for webhook routes + *.pg.test.ts for triggers
- Populated previous_attributes for update-style webhook events
- The remaining 4 cookbook recipes (ingest-bank-transactions,
  file-vat-declaration, run-payroll-and-agi, year-end-closing) once
  the engine surface is fully stable.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* refactor(api): address PR-497 review round 1 — CI fix + 5 small docs items

CI BLOCKER (the reason core-only failed):

1. **Type error on `[slug].md/route.ts` dynamic routes** — Next.js 16's
   route-type inference can't extract the dynamic segment from a
   directory whose name contains a literal suffix like `[slug].md/`. It
   types `params` as `Promise<{}>` and rejects our handler that
   declares `params: Promise<{ slug: string }>`. The framework still
   ROUTES requests correctly (URL `/docs/api/cookbook/quickstart.md`
   reaches the handler) — only the typed `params` is unusable.

   Fix: drop the typed `params` parameter on the two affected handlers
   (cookbook + reference) and parse the slug from `request.url.pathname`
   directly. Inline comment documents the workaround so the next person
   to touch these doesn't try to "fix" it back to the typed pattern.

GREPTILE INLINE (2 items):

2. **Python sample was missing `import json` and `import os`** — the
   webhook signature-verify sample uses both but only imported `hmac`,
   `hashlib`, `time`, and `flask`. Added the two missing imports.

3. **`buildResourcePages()` perf — called twice per request** (P2).
   Each call iterates every registered endpoint, groups by resource,
   sorts, and serialises Markdown for all 19 resource pages. Memoised
   at module level — the registry is populated once at module load and
   immutable for the process lifetime, so a single derivation is safe
   to cache. Halves the cost on the HTML routes' `generateMetadata` +
   page render pair, and the .md route handlers (which Next.js doesn't
   statically pre-render) are now constant-time after the first GET.

SWEDISH-COMPLIANCE PRECISION (3 items):

4. **`webhooks.ts`: behandlingshistorik vs räkenskapsinformation
   distinction.** The previous "Audit + retention" section conflated
   the two — webhook delivery rows are *behandlingshistorik* (system-
   event log) per BFNAR 2013:2 kap 8 §, NOT räkenskapsinformation
   themselves. The 7-year retention from BFL 7 kap 1 § attaches to the
   underlying verifikation/faktura/AGI XML in its own table, not to
   the delivery envelope. Updated the section to draw the distinction
   and clarify gnubok's 7-year retention on accounting-event delivery
   rows is an operational audit-trail policy, not a statutory
   obligation passed through to the integrator.

5. **`changelog.ts`: same distinction in the Phase 6 PR-1 entry** —
   replaced the "räkenskapsinformation" framing with the correct
   behandlingshistorik framing + the operational-policy note.

6. **Quickstart cookbook: ML 17 kap 24 § p.8 note about
   `beskattningsunderlag per skattesats`.** The "What just happened"
   section now explicitly notes that the rendered PDF contains every
   ML 17 kap 24 § field (including taxable amount per VAT rate) and
   that the JSON response's summary fields are convenience aggregates
   for the integration — the binding faktura content is the PDF.
   Forecloses the misreading that `subtotal + vat_total` is sufficient
   compliance.

7. **Changelog: BFL 7 kap caveat on SIE export.** The `/reports/
   sie-export` line now warns that a SIE4 export alone does NOT
   satisfy BFL 7 kap archiving obligations — SIE captures account
   positions and verifikationer but lacks system documentation and
   behandlingshistorik. Treat as a portability format
   (Fortnox/Visma/Bokio migration), not as a complete archive. Closes
   the misreading the swedish-sie-import-export skill flagged.

DEFENSIBLE DEFERS (round 1 final):

- **CM-8 SPDX-License-Identifier headers per file** (Compliance Swarm).
  The repo declares AGPL-3.0-or-later in the root LICENSE file, which
  satisfies licensing for the project as a whole. Per-file SPDX
  headers are a REUSE-conformance feature; we can address as a sweep
  across the entire codebase if/when REUSE conformance becomes a
  requirement. Out of scope for a docs PR.

- **`/llms-full.txt` exposes payroll endpoint metadata publicly**
  (A.8.12). By design — the entire point of the file is one-shot LLM
  ingestion of the public docs corpus. Endpoint METADATA (path, scope,
  description) is non-sensitive; actual payroll DATA is gated behind
  `payroll:read` scope and requires a real API key. Adding an auth
  gate would defeat the agent-discovery purpose.

- **Secret rotation endpoint** (Art.25(2)). Real product gap (delete +
  recreate is the current rotation path), but it's feature work, not
  docs. Tracked for Phase 6 PR-3 alongside the other webhook hardening.

- **DNS rebinding pinned-IP HTTPS agent** (Art.5(1)(f)). Already
  documented in this changelog as a Phase 6 PR-3 item; bot is just
  re-flagging that it's not yet shipped.

- **A.5.34 changelog cites GDPR Art.5(1)(c) for personnummer masking
  without linking privacy policy.** The citation is informational
  context for developers, not a privacy notice to data subjects. Data-
  subject notices live at /privacy. Adding a pointer is reasonable;
  adding it would consume real estate that's better spent on the
  technical detail. Defer.

- **Compliance Swarm A.5.21 third-party attribution in llms-full.txt**.
  False positive — all markdown content in this PR is original
  first-party text. No third-party snippets to attribute.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* refactor(api): address PR-497 review round 2 — 6 small precision fixes

All CI green after round 1 (core-only fixed). Compliance Swarm: 7 → 10
findings is the documented oscillation pattern — net-new actionable
items are 6 small fixes; the rest are recurring defers
(plaintext-secret variants, SPDX, planned PR-3 features the changelog
already lists as "coming soon").

FIXED:

1. **Slug allow-list validation in `[slug].md` routes** (V1.2.5 ×2,
   medium). The cookbook + reference .md routes parse the slug from
   the URL pathname (round-1 workaround for Next.js 16's failed
   inference on `[slug].md/` directories) and pass it to a
   dictionary-based lookup. The lookup itself is safe — findRecipe /
   buildResourcePages can't reach SQL or filesystem from a bad
   slug — but the explicit allow-list gate keeps the contract safe
   if the lookup mechanism ever changes (file-load, RPC, etc.).
   Added `Set<string>(COOKBOOK_SLUGS)` + `Set<string>(RESOURCE_SLUGS)`
   guard before any lookup runs.

2. **Changelog: BFL 5 kap 5 § cited on both `/reverse` AND `/correct`**
   (swedish-compliance precision). The previous wording cited BFL 5:5
   only on `/reverse` (storno) and described `/correct` as plain
   "rättelse" — but BFL 5:5 governs rättelse generally, and storno is
   the canonical method of rättelse, so both endpoints satisfy 5:5.
   Updated to: "/{id}/reverse (storno) and /{id}/correct (rättelse) —
   both satisfy BFL 5 kap 5 § (storno is the canonical method of
   rättelse)".

3. **Changelog AGI: explicit that XML is for manual submission**
   (swedish-compliance / swedish-payroll). Previously said
   "/generate-agi produces AGI XML" — could be misread as
   auto-submission to Skatteverket. Now states explicitly that the
   response carries `data.xml` for the integrator to upload via
   Skatteverket Mina Sidor (or via the optional `skatteverket`
   extension), and that the AGI deadline (12th / 17th of the
   following month) is the integrator's responsibility. Aligns with
   the route file's existing doNotUseFor + pitfalls metadata.

4. **Quickstart: F-skatt note qualified**
   (swedish-invoice-compliance). The "What just happened" section
   previously said the PDF "contains the F-skatt note" — only valid
   if the seller actually holds F-skatt. Updated to: "The 'Godkänd
   för F-skatt' note is included automatically when
   company_settings.has_f_skatt is set — confirm this on the company
   settings page before sending invoices in production." Also
   tightened the beskattningsunderlag wording to mention "one line
   per distinct rate on multi-rate invoices" — closes the
   swedish-compliance note about the multi-rate claim in the landing
   needing explicit support in the cookbook.

5. **Cookbook placeholder VAT description: "compute and review" not
   "submit"** (swedish-vat). The placeholder previously said "Compute
   momsdeklaration rutor and submit to Skatteverket" — but no
   Skatteverket-submission endpoint exists in the v1 surface; the
   API only computes the rutor 05–62 values for manual filing.
   Updated description in BOTH cookbook/index.ts AND nav.ts (where
   the same string was duplicated): "Compute momsdeklaration rutor
   05–62 and reconcile against the GL before manual submission to
   Skatteverket." Title also flipped from "File a VAT declaration"
   to "Compute and review a VAT declaration".

6. **Cookbook nav for AGI: aligned to "generate" semantics**. nav.ts
   AGI summary used to say "file AGI" — same misreading risk as #3.
   Now: "Calculate, approve, mark paid, book, generate AGI XML for
   manual Skatteverket upload."

DEFERS (round 2 final — every remaining swarm finding is in one of
these buckets):

- **🟠 Art.32 plaintext webhook secret + 4 sibling framings** (V14,
  V11.1, A.8.24, CC6.1). Established defer per Stripe / GitHub /
  Slack precedent; documented inline in lib/webhooks/signing.ts.
  Bot is re-flagging via the docs surface this round; underlying
  position unchanged.
- **🟡 Art.5(1)(e) 90-day TTL non-accounting cron + V16 audit log +
  V2.4 rate limits**. All explicitly listed in the changelog as
  "Coming soon (Phase 6 PR-3 hardening)". Bot is reading the same
  text we wrote; not blocking.
- **🟠 A.8.11 personnummer masking lacks an automated test**. Real
  product-hardening request, but it's feature work in the employees
  test surface, not docs. Tracked.
- **🟡 CM-8 SPDX-License-Identifier headers per file**. Established
  defer from round 1 — root LICENSE covers AGPL-3.0-or-later for
  the project as a whole; per-file SPDX is a REUSE-conformance sweep
  that's its own effort.
- **🟡 SR-3 SBOM dependencies**. False positive — next/server, react,
  next/link, next/navigation are existing project deps, not new in
  this PR.
- **swedish-compliance "SIE disclaimer could note immutability
  requirement"** — the current disclaimer accurately calls out
  system documentation + behandlingshistorik as missing; adding
  immutability would over-stuff a one-line caveat. Defer with the
  understanding that the SIE skill itself documents the
  immutability requirement for any consumer that follows the
  reference.

If round 3 plateaus (Compliance Swarm count stable, no net-new
inline items), that's the merge-ready signal per Phase 4 lessons.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* refactor(api): address PR-497 review round 3 — 4 small precision fixes

Compliance Swarm: 10 → 3 (down 70%) — net-new actionable items are
the 4 below; remaining 3 swarm findings are either trivial defense-
in-depth (URL decode, fixed here) or out-of-repo decisions
(personnummer disclosure DPO confirmation).

FIXED:

1. **URL-decode slug before allow-list check** in both .md route
   handlers (V1.2.5 ×2 low). The closed allow-list is pure ASCII so a
   percent-encoded value can't decode to a legitimate slug, but the
   explicit decode-then-check pattern keeps the contract correct under
   any future encoding-quirk runtime. try/catch around
   decodeURIComponent so a malformed % sequence (which throws) returns
   a clean 404 rather than a 500.

2. **Quickstart: explicit `delivery_date` requirement** (swedish-
   invoice-compliance / ML 17:24 field 7). The previous wording
   listed "supply date" as a covered field but didn't note that the
   API does NOT default delivery_date to invoice_date — integrators
   shipping invoices for goods delivered on a different date than the
   invoice date must pass delivery_date explicitly or the rendered
   PDF is non-compliant. Added explicit pass-it-yourself note.

3. **Quickstart: F-skatt strengthened from "verify" to "legal
   requirement"** (swedish-invoice-compliance / Peppol BIS 3.0
   SE-R-005). The previous "confirm on settings page" wording risked
   integrators treating the F-skatt note as optional UX. It's a legal
   requirement on every faktura issued by a company that holds
   F-skatt registration — and a FATAL Peppol BIS 3.0 validation
   failure (SE-R-005) for B2G invoices when missing. Reframed as a
   compliance assertion: the PDF includes it automatically when the
   setting is true; verifying the setting is correct before
   production is the integrator's responsibility.

4. **Changelog: SIE post-import VAT code reconfiguration warning**
   (swedish-sie-import-export). The /imports/sie line previously
   noted the file format support but didn't warn that SIE files do
   NOT carry VAT codes or tax-rate-to-account mappings. After
   migrating from Fortnox / Visma / BL / SpeedLedger / Bokio,
   integrators must manually reconfigure VAT codes before the first
   momsdeklaration — skipping this is the most common source of
   incorrect VAT submissions in migrated bookkeeping.

DEFERS (round 3 final — these are the architectural floor):

- **🟠 A.5.34 personnummer field name + masking logic disclosed in
  public docs** (Compliance Swarm). Defensible: documenting that
  personnummer is masked is a transparency benefit (GDPR Art.13/14
  intent), not a privacy disclosure risk. The DPO confirmation prompt
  is a reasonable governance ask but is an out-of-repo decision —
  the docs change is appropriate as written.
- **AGI penalty amounts (625 / 1,250 SEK)**. Operational guidance
  for integrators building deadline-tracking; not strictly API-doc
  material. The deadline (12th / 17th) is documented; integrators
  who automate compliance can read SFL for penalties.
- **VAT period thresholds in the cookbook placeholder**. Belongs in
  the actual cookbook content when written, not in the placeholder
  description.
- **`invoice.credited` event-naming verification**. False positive —
  the emitter uses `credit_note.created` (which IS in the docs);
  there is no `invoice.credited` event in the codebase. Naming is
  consistent.
- **Webhook retention sentence reordering** (swedish-compliance
  stylistic). Current wording leads with what delivery rows ARE
  (behandlingshistorik), then clarifies what they are NOT
  (räkenskapsinformation) — clean teaching arc, the qualifier is
  prominent. Reordering doesn't change clarity.

Round 3 stop signal hit (per Phase 4 lessons): swarm count plateauing
at the architectural floor with all remaining findings either
defers, false positives, or out-of-repo decisions. Greptile has
posted no inline comments since round 1's two items (both fixed).
This should be merge-ready.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* refactor(api): address PR-497 review round 4 — 5 small precision fixes

Compliance Swarm: 3 → 5 (slight uptick from oscillation, but 0
critical, 1 actionable; remaining 4 are recurring or philosophical).
swedish-compliance: 7 advisories — 3 actionable precision items
incorporated below; the others are forward-looking notes for cookbook
content that ships in Phase 6 follow-ups.

FIXED:

1. **Spec-snapshot test enforces ep.scope is explicitly defined**
   (CC6.3, real future-bug prevention). Previously the test asserted
   every endpoint declared the agent-facing metadata fields the docs
   depend on, but `scope` could be `undefined` — a registerEndpoint
   call that silently dropped the field would make the wrapper treat
   the route as unauthenticated. Added an assertion that
   `ep.scope !== undefined` (the literal sentinel `null` is allowed
   for genuinely public endpoints like /api/v1/health). The 4 spec
   tests still pass — confirming no current endpoint has undefined
   scope and the gate works prospectively.

2. **F-skatt: integrator responsibility for `has_f_skatt` accuracy**
   (swedish-invoice-compliance). The previous "verify on settings
   page" framing didn't connect the flag to the live Skatteverket
   registration. Now: "The integrator is responsible for keeping
   has_f_skatt in sync with the company's live Skatteverket
   registration status. Update via PATCH /api/v1/companies/{id}/
   settings or the settings page — a flag that's false while the
   company is actually F-skatt-registered produces non-compliant
   invoices, not merely a missing optional note."

3. **AGI deadline qualified by turnover** (swedish-payroll). The
   previous wording listed "12th / 17th of the following month" with
   no condition. Now: "12th of the following month for large
   employers, 17th for companies with annual turnover ≤ 40 MSEK."
   Aligns with the swedish-payroll skill's AGI filing deadline
   section.

4. **SIE import warning includes behandlingshistorik gap**
   (swedish-sie-import-export + swedish-accounting-compliance). The
   previous warning covered the VAT-code reconfiguration requirement
   but didn't note that SIE files also do NOT transfer
   behandlingshistorik (the source system's processing log per
   BFNAR 2013:2 kap 8 §) or systemdokumentation. Added: "The
   behandlingshistorik gap must either be preserved separately
   (export from the source system + archive alongside the SIE file)
   or accepted with documented justification — gnubok starts a fresh
   behandlingshistorik from the import date forward."

5. **Webhook 7-year retention: voluntary policy vs statutory
   obligation** (swedish-accounting-compliance). The previous wording
   said gnubok keeps delivery rows "for 7 years as an operational
   audit-trail policy" — the 7-year figure could be misread as
   statutory. Tightened in BOTH webhooks.ts and changelog.ts:
   the 7-year statutory retention under BFL 7 kap 1 § applies ONLY
   to the underlying verifikation/faktura/AGI XML; gnubok's 7-year
   policy on delivery rows is voluntary and chose the duration to
   align conveniently with the statutory horizon on the underlying
   records.

DEFERS (round 4 final, all in the architectural-floor bucket):

- **🟠 A.5.34 personnummer field name + masking logic disclosed in
  public docs** (recurring from round 3). Defensible — documenting
  PII handling is a transparency benefit (GDPR Art.13/14 intent),
  not a privacy disclosure risk. The DPO confirmation prompt is a
  reasonable governance ask but is an out-of-repo decision.
- **🟡 A.8.23 DNS-rebinding "coming soon" item**. The bot is reading
  the changelog's own deferral list. Already tracked for Phase 6
  PR-3 hardening.
- **🟠 CC6.6 SSRF protection details exposed in /llms-full.txt**.
  Stripe / GitHub / Slack publish their full webhook security
  posture publicly (signature format, rejected IP ranges, retry
  policy) — documenting protections IS the trust pattern. Obscurity
  is not security; the SSRF protection is enforced in code, not in
  the docs.
- **🟡 CC6.7 public CDN caching**. withPublicSecurityHeaders()
  already applies the appropriate headers (CSP, X-Content-Type-
  Options, X-Frame-Options). The 5-min cache is appropriate for
  static developer documentation; the alternative (no caching) is
  cost without security benefit since the content is intended to be
  public.
- swedish-compliance "VAT 2026-04-01 livsmedel rate change" —
  forward-looking; ships when the actual VAT cookbook is written.
- swedish-compliance "year-end IB/UB continuity" — forward-looking;
  ships when the year-end cookbook is written.
- 2 verify-only notes (rättelse implementation, future salary-
  journal/avgifter-basis masking sweep) — not actionable in this PR.

Round 4 stop signal: every remaining swarm finding is in the
deferred or recurring bucket; the actionable item (CC6.3) is
shipped. swedish-compliance is now in advisory mode (no errors,
just stylistic suggestions and future-cookbook notes). Per Phase 4
lessons, this is the merge-ready signal.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* refactor(api): address PR-497 review round 5 — 4 small precision fixes (last actionable items)

Compliance Swarm: 5 → 2 (down to architectural floor — 1 high + 1
medium). swedish-compliance: 7 advisories, 4 actionable precision
items addressed below; the others are forward-looking notes for
content that ships in Phase 6 follow-ups.

Trajectory: 7 → 10 → 3 → 5 → 2. Plateaued.

FIXED:

1. **Webhook secret storage guidance: secrets manager, not env file**
   (A.8.5 high). Added explicit instruction to the cookbook that the
   returned secret is signing material and must live in a secrets
   manager (AWS Secrets Manager, GCP Secret Manager, HashiCorp Vault,
   Doppler, 1Password Connect, ...) — not a plaintext .env file or
   config commit. Treated with the same care as a database password.

2. **AGI deadline correction: 17th = January and August only**
   (swedish-payroll). Round 4's wording said "12th (large employers) /
   17th (≤40 MSEK)" — but per the swedish-payroll skill the 17th
   applies in January and August specifically, not generally to all
   months for sub-40 MSEK companies. Other months are the 12th
   regardless of employer size. Fixed to: "the 12th of the following
   month for every reporting period EXCEPT January and August, where
   companies with annual turnover ≤ 40 MSEK get the 17th." This
   would've caused integrators automating sub-40 MSEK deadline
   tracking to misfile by 5 days from February through July and
   September through December.

3. **F-skatt SE-R-005 broader scope** (swedish-invoice-compliance /
   swedish-e-invoicing). The previous wording framed SE-R-005 as
   primarily a Peppol B2G validation rule. Reframed: the F-skatt
   note is a legal requirement on every faktura issued by a Swedish
   momsregistrerad seller that holds F-skatt registration — applies
   to PDF/paper AND Peppol/e-invoice formats. The buyer uses it to
   determine A-skatt withholding obligation (omitting it can shift
   tax liability onto the buyer); B2G is just where the validation
   is automated as a FATAL Peppol BIS 3.0 check.

4. **SIE behandlingshistorik gap: full räkenskapsår scope**
   (swedish-accounting-compliance). Round 4's wording said the
   integrator "must either preserve [behandlingshistorik] separately
   or accept the gap with documented justification" and that gnubok
   "starts a fresh behandlingshistorik from the import date forward."
   The "documented justification" framing implied the gap was
   acceptable as a default. Per BFNAR 2013:2 kap 8 §, the obligation
   attaches to the entire räkenskapsår, not from the import date.
   Reframed as: "must be preserved separately... best practice for a
   mid-year migration: export the source system's behandlingshistorik
   for the full fiscal year and archive it alongside the SIE file."

DEFERS (round 5 final — these are the architectural-floor items
that will recur indefinitely):

- **🟡 A.8.20 DNS-rebinding gap** (Compliance Swarm). Already
  documented in the changelog as a Phase 6 PR-3 deferral item; the
  bot is reading the same text we wrote.
- **swedish-compliance: VAT 2026-04-01 livsmedel rate change**.
  Forward-looking — for the actual VAT cookbook recipe content,
  which ships post-Phase-6.
- **swedish-compliance: year-end IB/UB continuity**. Forward-looking
  — same.
- **swedish-compliance: SIE warning placement note**. Forward-looking
  — for the imports reference page when authored.
- **swedish-compliance: BFNAR 2013:2 citation correct, webhook
  retention correct**. No-op confirmations.
- **swedish-compliance: delivery_date pre-payment scenario**. Real
  but extremely narrow edge case (faktura utfärdad före leverans).
  Defer with the understanding that anyone using the API for
  pre-payment invoicing will read the full invoice reference, not
  rely solely on the quickstart.

This is the merge-ready signal per Phase 4 lessons-learned: every
remaining swarm finding is in the deferred or recurring bucket;
swedish-compliance is in pure-advisory mode (forward-looking notes
for cookbook content that ships later); CI is fully green; Greptile
posted nothing past round 1's two items (both fixed). Ship it.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-15 15:57:28 +02:00