c0ecf2fa3bebd46bdfd0169efd73b89653d1dfed
2 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
 Jakob Wennberg
|
18cbc4c30a |
fix(security): audit remediation 2026-09-01: api_keys identity, viewer gates, OAuth binding, XSS, MFA gate (#2155)
* fix(security): bind api_keys to the caller, lock hash-as-bearer RPCs and provider token tables Security audit 2026-09-01, critical items. - api_keys INSERT requires user_id = auth.uid() again (an admin could forge a key for any co-member and act as them in every company they belong to); SELECT is own-keys-or-admin; a BEFORE trigger freezes the identity and credential columns against user-session UPDATEs. - rotate_mcp_refresh_token and validate_and_increment_api_key become service_role only: they match rows by a presented SHA-256, so a hash readable by co-members was a bearer credential. - validate_and_increment_api_key fails closed when the key's user is no longer a member of the key's company. - provider_consent_tokens and provider_otc: the DELETE policies collapsed to "caller has any team row" (correlated subquery on a non-existent team_members.company_id). All member policies dropped; service_role only, matching every existing code path. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * fix(security): role gates, ownership guards and posting integrity in the database Security audit 2026-09-01, high items at the database layer. - One table-level guard, enforce_company_writer_role(), blocks the read-only viewer role on 55 company-scoped tables including through the 15 membership-only SECURITY DEFINER writers. Keyed on the JWT role claim so it fires inside definer bodies; no-op for service_role and trigger cascades. - company_members user_id/company_id immutable from user sessions; invitations can never grant owner; team_members gains a transition guard (admins keep non-owner role moves); companies team_id and archiving are owner-only and team attachment needs team membership. - Direct statements (current_user = authenticated) can no longer insert posted headers, add lines under posted verifikat, or post a draft with a voucher number the sequence never issued. Sanctioned RPCs run as the definer and are untouched; the engine's own draft-then-post shapes still pass. - create_document_version refuses viewers and foreign storage paths; validate_version_chain needs membership and loses anon EXECUTE; match_documents / match_booking_templates lose anon; cron maintenance RPCs become service_role only; the production-only seed_asset_categories is dropped. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * build: pin tsx as an exact devDependency instead of fetching it with npx at build time prebuild ran "npx tsx" with no lockfile entry, so every Vercel, Docker and CI build downloaded tsx@latest and its transitive tree from the registry with no integrity check, inside the build environment. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * fix(security): refuse the viewer role on API-key and MCP write paths The v1 wrapper and the MCP company routing checked company membership but never role, and both run as service role, so a read-only viewer holding an API key could post vouchers and change settings through the API. Mutating methods and non-read scopes now return 403 ROLE_READ_ONLY for viewers on v1; MCP write tools refuse viewers the same way. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * fix(security): stop serving uploaded SVG, XML and HTML as executable content on the app origin Uploads persisted the browser-declared mime type and the inline proxy served it verbatim, sandboxing only text/html; the storage proxy forwarded the uploader's Content-Type. Any writer, or any Peppol sender, could plant a scripted SVG or XHTML that executed on app.gnubok.se. - inline route: allow-list of natively safe types (PDF, raster images) served as before; everything else gets the opaque sandbox CSP. - storage proxy: octet-stream + attachment + sandbox unless the DB mime for the key is on the allow-list. - document-service: the stored mime is the magic-byte validated type. - logo upload: magic-byte validation, SVG refused. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * fix(security): byrå brand logo upload decides the type by magic bytes and drops SVG Same pattern as the company logo route: the logos bucket is public, so a scripted SVG (or anything declared as an image) must never land there. The upload pickers stop advertising SVG. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * fix(security): bind Enable Banking, Stripe and WooCommerce callbacks to the initiating user The callbacks resolved the pending row by oauth_state alone, so a victim who completed an attacker-initiated consent had their bank account, merchant account or store attached to the attacker's company. requireFlowInitiator() now requires the cookie session of the user who started the flow: no session redirects to login with the callback URL preserved, a different user is refused and nothing is exchanged. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * fix(security): guard tenant-controlled outbound fetches and surface the disabled rate limiter WooCommerce and Shopify syncs fetched a member-editable store URL with plain fetch() and redirect following under the service role, and the invoice PDF renderer fetched company_settings.logo_url unguarded. All three go through a new safeFetch() (public-IP validation via url-guard, https only, redirect: 'manual', body size cap) and re-normalise the stored host at use time. checkRateLimit() keeps failing open on hosted but logs one error per process when Upstash is not configured and exports isRateLimiterConfigured(). Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * fix(security): decide the API MFA gate from server-authenticated factors, not the session cookie getAuthenticatorAssuranceLevel() without arguments derives nextLevel from session.user.factors, which comes from the unsigned sb-*-auth-token cookie. Deleting factors from the cookie made an enrolled account look like it had nothing to step up to, on every /api route and in requireAuth. Both gates now read factors from the getUser() result or listFactors() and the level from the verified JWT claim, and fail closed on errors. Page-branch gate hardened the same way. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * fix(security): bind Fortnox/Visma, Gmail and Skatteverket callbacks to the initiating user The arcim-migration callback exchanged the provider code onto whatever consent the one-time state named, with no check of who completed the flow and no org-number comparison, so a phished Fortnox admin handed their ledger to the attacker's company. provider_otc now records the initiating user (migration 20260902100000); the callback requires that session and, after the exchange, refuses a provider company whose org number differs from the consent's company. The Gmail and Skatteverket callbacks enforce the same initiator check. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * fix(security): BankID signup confirms the email before linking the identity Signup created an email-confirmed, MFA-exempt account for any address the caller typed and returned a magic link, so an attacker could pre-register a victim's email and keep a permanent BankID login into the account the victim later adopted. The user is now created unconfirmed, the identity carries email_verified_at NULL (migration 20260902101000), bankid_linked is not set until the mailed confirmation is clicked, and BankID login of a pending identity is refused with the confirmation re-sent. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * fix(security): bind MCP OAuth redirect URIs to the consenting user and cap scopes A user-registered redirect URI was allowlisted globally, the consent page named no client, and all scopes were pre-checked, so one phishing link handed an attacker a full-scope key for the victim's company. Registered URIs now resolve only for the registrant or a colleague sharing a company; the consent page shows the client identity and redirect host; non-built-in clients default to read-only pre-checks; scopes are capped by the user's role (viewer: read only) at consent and at /token. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * fix(auth): client follow-ups for BankID confirmation, callback mismatch copy and decision log - register client handles the new confirmation_sent response from BankID signup with the existing inbox screen instead of calling verifyOtp. - BankID login surfaces the email_unconfirmed explanation. - WooCommerce settings map woocommerce_error=wrong_user to its own copy. - Logo help text no longer advertises SVG. - DECISIONS.md records the audit remediation choices. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * fix(mcp-oauth): literal SoD columns in the api_keys insert so the phantom-column scanner resolves them Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> * test(logo): type the upload fixtures as Uint8Array<ArrayBuffer> so they are valid BlobParts Fixes the typecheck ratchet on PR #2155 and ratchets the baseline down by the one legacy error the change removed. Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com> --------- Co-authored-by: Jakob Wennberg <311770904+jakobwennberg-oss@users.noreply.github.com> Co-authored-by: Claude Fable 5.1 <noreply@anthropic.com> |
||
|
Jakob Wennberg
|
9fe37b85b5 |
feat(agents): per-key approval authority, the amount an agent may post unattended (#2079)
* feat(agents): per-key approval authority, the amount an agent may post unattended An API key gets an optional ceiling in SEK. Above it the agent may still stage the work, it just may not finish it alone: a human approves the same verifikat in the app. Default is NULL, so every existing key keeps its behaviour and turning this on is entirely opt-in. Enforced at the two places an API key reaches the ledger, and at both the refusal happens BEFORE the point of no return: - MCP: in commitPendingOperation, before the atomic claim, so the operation stays 'pending'. Behind the claim it would be caught by the generic handler, marked terminal 'rejected', and the staged verifikat would be gone. - REST: in journal-entries.commit, before commitEntry, so the draft stays a draft and the voucher sequence never advances (BFL 5 kap. 7 §). The dry run refuses too, rather than promising a voucher number the key cannot deliver. Not enforced inside commit_journal_entry: a RAISE there is swallowed by engine.ts into a retryable 500, and it would cost a DROP+CREATE on the function that issues every voucher number. Operations whose amount is only known during dispatch (batch allocation, bulk booking, the settlement link paths) fail OPEN behind an explicit allowlist. Pricing them ahead of dispatch would be a guess, and a wrong guess silently breaks batch allocation the day someone sets a limit. The allowlist is derived from what production actually stores: create_voucher carries total_debit on 1389 of 1389 rows, categorize_transaction carries amount on 2002 of 2003, create_supplier_invoice_from_inbox carries total on 208 of 228. This is a blast-radius cap, not a security boundary. A per-entry ceiling is defeated by splitting one entry into several, and an LLM will find that, so UNATTENDED_COMMIT_LIMIT_EXCEEDED forbids splitting first: one affärshändelse is one verifikat (BFL 5 kap. 6 §). A cumulative rolling-window limit is the primitive that actually bounds exposure and is left to a separate change. The guard is written NULL-first everywhere. An absent, unparseable or non-positive ceiling always means unlimited, never "block everything". Agents read their own ceiling from gnubok_get_agent_briefing instead of discovering it by burning a staged verifikat on a 403. Changing a ceiling is auditable: it now renders in behandlingshistorik (BFL 5 kap. 11 §). The audit trigger already fired on the column, but the report dropped the event because the field was not in its diff map. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * chore(skill): regenerate accounted-api skill for the new commit pitfall apiskill:check is a ratchet: the generated reference must match the endpoint registry. Never hand-edited. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * test(agents): pin the DB default itself, and declare the briefing field required Two review findings, both real: - the default test stored an explicit NULL, so it stayed green even if the column default changed to a positive ceiling: the one change that would silently start blocking every existing key. It now omits the column. - gnubok_get_agent_briefing documents unattended_commit_limit as always present and emits it unconditionally, so it belongs in the output schema's required list. Declined the NOT VALID constraint suggestion, with the reason recorded in the migration: api_keys is 388 rows / 768 kB in production. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * docs(agents): name the TOCTOU window in the REST ceiling check A security scan flagged that the line sum is read before commitEntry, so a concurrent write to the draft's lines can post over the ceiling. Real, and accepted: closing it means enforcing inside commit_journal_entry, where a RAISE becomes a retryable 500 and destroys the staged operation on the MCP path. Recorded in the code rather than left implicit, so nobody later mistakes this for a hard control. A per-entry ceiling is already defeated by splitting, which needs no race; the cumulative rolling-window limit is the primitive that bounds exposure. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * fix(agents): price the settlement and batch paths that were bypassing the ceiling A security scan flagged that known money-posting operations fail open, and it was right. The first cut priced only create_voucher, categorize_transaction and create_supplier_invoice_from_inbox, on the belief that the batch and settlement paths computed their totals only inside SQL at dispatch. Production says otherwise: the staged preview already carries the amount, because it is the number a human is shown when approving the operation. Over the last 120 days each of these is present and numeric on 100% of that type's staged rows: link_transaction_journal_entry transaction_amount 1369 rows bulk_book_transactions tx_sum 273 rows link_supplier_invoice_voucher payment_amount 55 rows match_batch_allocate total_allocated 24 rows mark_invoice_paid total 3 rows So a key with a ceiling could post any amount through the four largest settlement paths. Now priced, and the ceiling applies. Only reconciliation_match stays unpriced: it carries pair_count, which is a COUNT. Pricing off that would compare pairs against kronor, which is worse than not enforcing. link_document_to_voucher and attach_document_to_transaction move no money at all; the transaction_amount they carry is context, not a posting. Genuinely unpriceable types still fail OPEN. This control can only ever narrow what a key does, and a wrong guess at an amount blocks a legitimate commit, so guessing high would leave an agent unable to work. Adds a test that walks the whole allowlist, so a typo'd field name cannot silently make a type unpriceable again: that is exactly the hole this closes. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * fix(mcp): drop the ceiling from the agent briefing, the payload budget has no room The tools/list context-budget bench sits at 65 000 tokens and main now leaves roughly 20 tokens of headroom. An always-present field on the briefing's output schema costs about 85, so this addition alone pushed the bench red. The bench's own note is explicit that the answer is to demote a tool rather than raise the ceiling, so raising it here would be the wrong trade for a nice-to-have. Nothing is lost that matters: the operation is never destroyed when it is refused, so discovering the ceiling from UNATTENDED_COMMIT_LIMIT_EXCEEDED costs one round trip and no work. That error already carries both attempted and limit, and GET /api/settings/api-keys returns the value. Re-exposing it on the briefing is worth doing once there is budget to spend. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> * docs(api): spell affärshändelse correctly in the commit pitfall Fixed in the route's registerEndpoint pitfalls, which is the source; the skill reference is regenerated from it and never hand-edited. 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> |