erp-mafia/accounted/accounted-api
Claude Code and agent skill by erp-mafia/accounted
- community
Checking that the SKILL.md file is still in the repository.
Consume the Accounted REST API (Swedish double-entry bookkeeping SaaS, https://app.gnubok.se/api/v1). Use when building an integration, app, backend job, or agent tool layer against Accounted: invoices, customers, suppliers, supplier invoices, journal entries (bokföring), bank transactions and reconciliation, payroll (lön), VAT/moms and financial reports, SIE import/export, documents, webhooks. Covers auth with gnubok_sk_ API keys, conventions (dry-run, idempotency, cursor pagination, scopes), and all 294 endpoints.
Install
npx vskill@latest install erp-mafia/accounted/accounted-apiSKILL.md
Accounted API integration
Accounted is Swedish double-entry bookkeeping (bokföring) as a service: BAS chart of accounts, verifikationer with legally immutable audit trails, VAT (moms), payroll (lön), invoicing, bank reconciliation, and statutory reports, exposed as a REST API designed for agents and integrations first.
This skill is for building software against the REST API (an app, a backend job, an agent tool layer). If the goal is to operate a ledger conversationally (book receipts, run month close), use the Accounted MCP connector and its workflow skills instead: install the accounted plugin or see https://app.gnubok.se/docs/api/connect-claude.
If you have used Stripe's API the shape will feel familiar: bearer keys, dated versions, idempotency keys, webhook signatures, cursor pagination. The domain rules are Swedish accounting law; the Gotchas section below stops the classic violations before you ship them.
Auth and base URL
Every request sends a bearer key:
- Base URL:
https://app.gnubok.se/api/v1(legacy machine host, permanent).
https://app.accounted.se/api/v1 serves the identical API.
- Keys are created in the Accounted dashboard under Settings -> API
(/settings/api). Two prefixes:
gnubok_sk_live_*commits real writes.gnubok_sk_test_*reads real company data but forces every write into
dry-run (responses carry X-Gnubok-Mode: test). Develop and run evals with a test key; switch to live last.
- Each key carries scopes (
invoices:read,invoices:write,
payroll:write, webhooks:manage, ...). Every endpoint in the index below is annotated with its required scope; a missing scope returns 403.
- Rate limit: 100 requests/minute per key. On
429, honorRetry-After. - URLs carry the company id explicitly
(/api/v1/companies/{companyId}/invoices). A key can act on any company its user is a member of; start every session with GET /api/v1/companies to discover ids. There is no implicit "current company".
First calls, in order: GET /api/v1/health (no auth, connectivity), then GET /api/v1/companies (auth works, discover companyId).
Conventions
These rules hold across the whole surface; endpoint entries below do not repeat them.
Response envelope. Success: { "data": ..., "meta": { "request_id", "api_version", "next_cursor"?, "audit"?, "partial_expansions"? } }. Errors replace data with error (no meta; request_id moves inside error).
Errors. Stable machine codes with agent-oriented remediation:
React to code, read message_en and recovery_hint, follow valid_alternatives when present (e.g. next_open_period). Standard codes on every endpoint: 400 validation, 401 bad key, 403 missing scope, 404, 429 rate limited (honor Retry-After), 500. Full catalogue: https://app.gnubok.se/docs/api/errors. Only endpoint-specific codes are mentioned per endpoint below.
Cursor pagination. List endpoints take ?cursor= and return meta.next_cursor; loop until it is absent/null. A stale or tampered cursor is NOT an error: the first page is returned again, so terminate on next_cursor, never on "page looks familiar".
Dry-run on every write that supports it (dry-run badge in the index). Send ?dry_run=true (or X-Dry-Run: true): the response is always 200 with data.dry_run: true plus a preview (would-be record, journal lines, voucher number) and the X-Dry-Run: true response header; nothing is committed. Commit by re-issuing without the flag and with the SAME Idempotency-Key (the dry run is not cached against the key). Preview first on any financial write; it is free.
Idempotency-Key. Send a fresh UUID header on every POST/PATCH/DELETE; several endpoints reject writes without one (400). Replaying the same key+body returns the original response with the Idempotent-Replayed: true header; the same key with a different body returns 409 IDEMPOTENCY_KEY_REUSE (24h window). Safe retry loop: keep the key, keep the body.
Test keys are simulation-only. With a gnubok_sk_test_* key, reads return real company data (responses carry X-Gnubok-Mode: test) and every write is forced into dry-run; writes that cannot be simulated return 403 TEST_KEY_WRITE_BLOCKED. Nothing a test key does ever persists: it is ?dry_run=true baked into the credential. Full end-to-end write tests therefore need a live key against a company you own.
Atomic writes. A mutation either commits fully or errors with no side effects. There is no partial state to clean up after an error response (bulk-create endpoints that do partial success say so explicitly).
Audit inline. Successful financial writes include meta.audit (voucher number, audit-trail URL, immutability timestamp). No follow-up read needed to confirm what was booked.
Expansion. Some list/detail endpoints take ?expand=a,b (documented per endpoint). If an expansion fails the response still succeeds and names the failed parts in meta.partial_expansions; check it before trusting expanded fields.
Report query parameters. A report read never answers an unfiltered report to a dimension filter it cannot apply: dim_no, dim_code, dimensions, project or cost_center on a report that does not document them is 400 VALIDATION_ERROR. Any other query parameter a report does not document is not applied, and the response names it in the X-Ignored-Query-Params header: check that header when a filter seems to have had no effect.
Async operations. Long-running actions (fiscal-period lock/close/year-end, imports) return 202 with an operation id; poll GET /api/v1/operations/{id} until status is succeeded/failed. The response shape is identical whether the work ran inline or queued.
Versioning. Dated versions (current: see meta.api_version). Responses carry the Gnubok-Version header; request pinning via a Gnubok-Version request header is reserved for a future breaking change and is not read today. Additive changes ship without a version bump; see https://app.gnubok.se/docs/api/versioning.
Index badges. Every operation line below carries machine-readable annotations from the spec: scope: (required key scope), risk: (low/medium/ high; confirm with a human before unprompted high-risk calls), idempotent (safe to retry), dry-run (previewable), reversible (a single follow-up call can undo it, e.g. invoice credit).
Endpoint index
API version 2026-05-12, 294 operations. Paths are shown without their /api/v1 prefix (full base URL: https://app.gnubok.se/api/v1).
Core (11)
Full detail: references/core.md
Journal entries (18)
Full detail: references/journal-entries.md
Periods and registers (49)
This is the start of the file. Read the full SKILL.md on GitHub.
Security
- Security scanPassed, 95 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/accounted-api/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/erp-mafia/accounted/accounted-api)