Kvitt is served on the default HTTPS port through Nginx Proxy Manager, so the
OpenAPI server is https://{host} (contracts/openapi.yaml and openapi.rs
PUBLIC_SERVER_URL), examples E-7 public_url and E-12 link are
https://kvitt.example.invalid without a port, tools/contracts-check.py expects
the portless header, and the server's tests, config message and README name
no port.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TQd3pEP9UeN2uifbHknQ7B
KVITT_PUBLIC_URL is required: every command refuses to start without it (compose
always sets it). KVITT_TRUSTED_PROXY defaults to none, which the rate limiter never
resolves. contracts/openapi.yaml and the utoipa document: servers[].url
https://{host}:8443 with variable host (default kvitt.example.invalid). E-7 and E-12
examples on kvitt.example.invalid.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TQd3pEP9UeN2uifbHknQ7B
E-6 (/apk/files/{file_name}) gets its own per-IP bucket of 10 requests per
minute (security review L-3), keyed like the 60/min unauthenticated limit
(peer or trusted-proxy X-Forwarded-For, D-032) and answered 429
rate_limited with Retry-After. E-6 no longer counts against the 60; E-5 and
E-4 stay on the 60/min limit.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TQd3pEP9UeN2uifbHknQ7B
An authenticated device could store a token of up to the 1 MiB body limit, sent to FCM in every push for it. Now
(proposed D-114) more than 512 characters after trimming is 400 invalid_request with field fcm_token, like the
empty string (D-034); E-17's only documented error stays the only one. Test at 512 and 513.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TQd3pEP9UeN2uifbHknQ7B
AC-4.4 allowed 8000 x 8000 pixels: a few kilobytes of flat PNG or WebP decode to 256 MB RGBA, more than the server
container's 160 MiB. Now (proposed D-114) the header must also give at most 16 000 000 pixels and at most 64 MB
decoded (16 MP x 4 bytes; a 16-bit image is refused above 8 MP), checked before decoding, with the image crate's
decoder limits set as well; refused with the existing image_invalid. The decoded image is dropped before the square
is converted, and the square converts in place (into_rgba8/into_rgb8), so the peak is one decoded image plus one
square. Integration test at and over the pixel cap; unit test of the size rules.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TQd3pEP9UeN2uifbHknQ7B
Before, any response other than 401 (426 without X-Kvitt-Protocol, 405, ...) marked a random bearer token accepted
for good: it never counted against its IP and was never pruned, so a stranger could bypass the 60/min IP limit and
grow the in-memory map without bound. Now (proposed D-115, SPEC §8.6): a request with a token that is not accepted
counts against its IP before it runs and is refused once the IP is over; a 2xx accepts the token and gives that
count back (a newly paired device costs its IP nothing, as before); a 401 for an accepted token counts against the
IP and withdraws the acceptance; token entries are pruned once their window expires, accepted or not. Every 401
still counts against the IP exactly once. Unit tests for each rule and an integration test for random tokens
without the protocol header (60 x 426, then 429, the paired device unaffected).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TQd3pEP9UeN2uifbHknQ7B
Taken from a -W clippy::pedantic scan, no allow added: map_or/map_or_else, method references instead of closures,
clone_from, write! instead of push_str(format!), borrowed Option<&str>, push_of without a needless Option, the
envelope's entity moved instead of cloned, the scheduler's plan() no longer a needless async fn (its doc now says
what it does on failure), sort_unstable, digit separators and inline format args in tests. Not taken: missing
# Errors/# Panics docs, must_use, doc backticks, assert!(x.is_empty()) in tests, too_many_lines.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TQd3pEP9UeN2uifbHknQ7B
core_api/common.rs becomes tests/common/mod.rs and sync_api/m4.rs tests/common/sync_fixtures.rs; each binary
includes them through #[path] (the binaries of SPEC §14.7 stay as they are). TempDir's Default moves into the
harness instead of three copies; the skeleton helpers take ids, tokens and TempDir from it. Milestone wording in
test docs and the server README replaced by what the code does; README seams and test layout updated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TQd3pEP9UeN2uifbHknQ7B
- kvitt-server invite-partner: a partner invite by the admin as E-12 does (cancels the pending one, 72 h, link with
?from=<admin>), exit 2 household full, 3 no admin; README lines in server/ and infra/.
- Tests: invite-partner, migrate (schema at SCHEMA_VERSION, idempotent), healthcheck (200, 503, nothing listening),
serve SIGTERM drain (in-flight request finishes, new connections refused, exit 0; gives up after SHUTDOWN_DRAIN).
- Push sender: one info line "push sent" per message with push_type, devices_sent, devices_targeted and 8-character
device-id tails (SPEC §8.7 aggregates); per-device line stays at debug.
- tools/contracts-check.py: stale TODO(spec) replaced by the §9.11 CS-5 citation; shapes updated for invite_kind
(D-086), app_version_name (D-090) and fid (D-096), 134 checks pass.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TQd3pEP9UeN2uifbHknQ7B
Server: E-3 invite_kind; E-12 link ?from=<admin name>, E-4 passes it into kvitt://invite;
X-Kvitt-App-Version-Name stored in devices.app_version_name (migration 0003, schema 3);
OpenAPI and examples follow. Phone: :core:model, :core:fakes, :data, features and :app
updated together; goldens re-recorded for S-20 (version name) and new S-1/S-22 states.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TQd3pEP9UeN2uifbHknQ7B
- D-055 build-time copy of the keyword table, D-056 parseCategoryKeywords,
D-057 enum declaration orders, D-058 E-7 example shows server_version
15 and 16 and its test compares them again.
- D-051 fulfilled: §9.7 hex values from docs/ux/design-spec.md §2.3;
design-spec.md binding, strings.csv the source of string resources.
- D-059: ExpenseEditorOptions.frequentCategoryIds and
CategoryRepository.deletable; new milestone M12d with PLAN-M12d and
prompts/M12d.md; M7a-f depend on M6 and M12d.
- suggestCategory readings (null when other is archived), §11.2 landed
marks and the 2026-10-18 dry run, dev stack on schema 2, PLAN-M12b
SQLite 3.51.x, design-spec §6 notes in PLAN-M6/M7c/M7f; HANDOFF rewritten.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TQd3pEP9UeN2uifbHknQ7B
- migrations/0002_categories_v2.sql, byte-identical to contracts/schema/migrations/0002_categories_v2.sql
(SPEC §14.1.1), with an identity test; SCHEMA_VERSION = 2 (E-1 reports it, the importer's check follows).
- Seed test on a fresh database (the fourteen rows of AC-7.2, server_version 11 to 14) and an upgrade test of a
v1 database in use (0001 applied through the migrator, then users, a custom, a renamed and an archived default,
an account, a recurring template, expenses): every earlier row unchanged, the four rows above the old
sequence, foreign_key_check empty, schema_version 2, a pull from the old cursor returns exactly the four rows.
- DefaultCategoryKey with the fourteen keys (wire enum and contracts/openapi.yaml); icons 33 and colours 14
(§9.7) in validation.
- The two category examples M12a held back (E-26 response-default, sync/pull-page-categories-v2) and its
contracts-check patch for them and the enum.
- Importer: splitwise_categories.csv per the §10.8 v0.8 table (D-053), DEFAULT_KEYS of fourteen, synthetic
fixtures for the finer mapping (Insurance/Försäkringar to insurance among them), a schema-1 refusal test.
- Test harnesses seed users at server_version 15/16 (after the fourteen seeds); pull counts 10 -> 14.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TQd3pEP9UeN2uifbHknQ7B
jacobh2