SPEC v0.21: D-120 Nginx Proxy Manager instead of Caddy and cert-sync; D-121 Android host property

SPEC §10.5 rewritten (10.5.1 Nginx Proxy Manager, 10.5.2 Custom Nginx Configuration, 10.5.3 Fallback:
NPM down); Caddy and cert-sync text replaced in §1, §6.1, §7, §8.5, §8.7, §8.9, §10 intro, §10.1 to §10.4,
§10.7, §10.10, §11.2 to §11.4, OQ-3, OQ-6, E-7 and §14.7; the Kvitt host name is KVITT_PUBLIC_HOST.
D-121: kvitt.publicHost, RELEASE_SERVER_URL and @ReleaseServerUrl (AC-2.6, AC-2.7, §10.6, §13.3 S-1).
Decisions D-120 and D-121 with pointers in D-023, D-103 to D-106 and D-117; HANDOFF; prompts/README;
a superseded-in-part note on PLAN-M2c-pi; dated D-120 notes under M-2, L-1, L-2, L-8 and I-5.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01TQd3pEP9UeN2uifbHknQ7B
This commit is contained in:
2026-10-19 20:04:09 +02:00
parent d090c6c850
commit 4bcff2cf89
6 changed files with 191 additions and 147 deletions
+34 -30
View File
@@ -3,9 +3,13 @@
Living document, rewritten in place (never a second copy). A new orchestrator session reads CLAUDE.md, then this.
## 1. Where things stand
- `main` at `96e4cbd` (the APK download rate limit, D-119, landed after the SPEC v0.19 docs commit `3ef2c15`) plus the SPEC v0.20 docs commit from `agent-docs-v020` once it lands. `origin/main` (Gitea) at `96e4cbd`: everything up to the rate limit is pushed.
- `main` at `d090c6c` (the Android host property, D-121, on top of `9231a1d`, the end of the NPM rework, D-120) plus the SPEC v0.21 docs commit from `agent-docs-v021` once it lands. `origin/main` (Gitea) at `d090c6c`: everything is pushed (`9231a1d` triggered the Actions run of §3 item 1; `d090c6c` touches only `android/` and triggers none).
- History rewrite and first push, 2026-10-18 (owner authorised): superseded PNG versions were dropped from the whole history (no LFS), the current goldens and UX mock-ups restored in `18c527f`; pack 45 MiB; `origin/main` pushed at `18c527f`. A bundle of the pre-rewrite history (`kvitt-pre-rewrite.bundle`) is kept outside the repository, in the orchestrator's scratchpad. Every commit from M2b on got a new hash; the oxipng pass (`7c453d8` before the rewrite) no longer exists as a commit (it only touched PNGs). SPEC, `docs/decisions.md`, this file and `prompts/README.md` cite the rewritten hashes since v0.18 (`7c453d8` stays as the oxipng pass's name; `docs/validation/ac-coverage.md` keeps its pre-rewrite hashes verbatim); since v0.19 the plans cite rewritten hashes too (eight pre-rewrite hashes in `PLAN-*.md` replaced by the commit with the same subject); `docs/spikes/pi-image-arm64.md` and the READMEs had none stale. For any other old hash, find the commit by its subject with `git log --oneline --grep`.
- History (rewritten hashes): `5e1be38` plans and prompts, `c885aba` M2c-dev, `76ee957` SPEC v0.3, `c277ee3` M2b, `06f48ef` M2a, `6dce4c7` SPEC v0.4, `7869ec4` M3, `8f0e156` M5, `853c038` SPEC v0.5, `15056dc` M4, `d9dd584` SPEC v0.6, `50339ae` M9, `86e08a1` SPEC v0.7, `aa765bb` M12a, `d39eb2c` SPEC v0.8, `b09f322` M12c, `6eaccec` M12b, `0006bd7` UX Kalas categories and tokens, `6da6b1b` UX sprint U2, `e676812` SPEC v0.9, `557af7e` M12d, `7730a31` M6, `9ee67d4` SPEC v0.10, `a41f240` and `6453932` app wiring before M8, `1ea752c` M7a, `ea7d494` M7e, `7380e50` M7d, `0e1d99d` M7b, `0b140aa` M7c, `c4ab35f` M7f, `b24c8b5` M6b, `f4dc10c` SPEC v0.11, `3dda414` M7g, `cb832a7` and `3d9eea8` M8, `6159e18` SPEC v0.12, `8c60715` M12e, `b7b7483` M12e launcher icon, `51d6ff8` SPEC v0.13, `a312758` Vico removal, `f56dcbe` FCM `fid`, `3e6c337` Pi image arm64 check, `e2dba84` SPEC v0.14, `328c466` M8b, `6eb3f85` `FcmMessageBody.fid`, `bfde12c` SPEC v0.15, `75b549f` M8c goldens, `a0c25b0` M8c licences row, `c30a49e` R8 release build, `bacbd18` M8d, `bfc3acb` SPEC v0.16, `2476667` M2c-pi bundle, `a59abeb` M8e, `b4d9467` Dockerfile `CARGO_BUILD_JOBS`, `956908a` M8f, `1328afe` chips cleanup, `400eac9` SPEC v0.17, `18c527f` goldens restored after the rewrite; then `2fac1ec` to `e70fe0d` infra overhaul, `8660b7f` server gap pass, `7fb86f5` infra security, `dc7a296` to `a0259ab` M12f, `8103dae` to `b985ce9` server overhaul, `48fe119` SPEC v0.18, `e4e6e5d` to `c582363` Android overhaul, `3ef2c15` SPEC v0.19, `96e4cbd` APK download rate limit (D-119).
- History (rewritten hashes): `5e1be38` plans and prompts, `c885aba` M2c-dev, `76ee957` SPEC v0.3, `c277ee3` M2b, `06f48ef` M2a, `6dce4c7` SPEC v0.4, `7869ec4` M3, `8f0e156` M5, `853c038` SPEC v0.5, `15056dc` M4, `d9dd584` SPEC v0.6, `50339ae` M9, `86e08a1` SPEC v0.7, `aa765bb` M12a, `d39eb2c` SPEC v0.8, `b09f322` M12c, `6eaccec` M12b, `0006bd7` UX Kalas categories and tokens, `6da6b1b` UX sprint U2, `e676812` SPEC v0.9, `557af7e` M12d, `7730a31` M6, `9ee67d4` SPEC v0.10, `a41f240` and `6453932` app wiring before M8, `1ea752c` M7a, `ea7d494` M7e, `7380e50` M7d, `0e1d99d` M7b, `0b140aa` M7c, `c4ab35f` M7f, `b24c8b5` M6b, `f4dc10c` SPEC v0.11, `3dda414` M7g, `cb832a7` and `3d9eea8` M8, `6159e18` SPEC v0.12, `8c60715` M12e, `b7b7483` M12e launcher icon, `51d6ff8` SPEC v0.13, `a312758` Vico removal, `f56dcbe` FCM `fid`, `3e6c337` Pi image arm64 check, `e2dba84` SPEC v0.14, `328c466` M8b, `6eb3f85` `FcmMessageBody.fid`, `bfde12c` SPEC v0.15, `75b549f` M8c goldens, `a0c25b0` M8c licences row, `c30a49e` R8 release build, `bacbd18` M8d, `bfc3acb` SPEC v0.16, `2476667` M2c-pi bundle, `a59abeb` M8e, `b4d9467` Dockerfile `CARGO_BUILD_JOBS`, `956908a` M8f, `1328afe` chips cleanup, `400eac9` SPEC v0.17, `18c527f` goldens restored after the rewrite; then `2fac1ec` to `e70fe0d` infra overhaul, `8660b7f` server gap pass, `7fb86f5` infra security, `dc7a296` to `a0259ab` M12f, `8103dae` to `b985ce9` server overhaul, `48fe119` SPEC v0.18, `e4e6e5d` to `c582363` Android overhaul, `3ef2c15` SPEC v0.19, `96e4cbd` APK download rate limit (D-119), `481889a` SPEC v0.20, `f1abb0d` `cargo audit` in the server gate, `4f69ae0` Gitea login in the run line, `d05f13a` Pi stack behind NPM, `62f5279` and `9231a1d` infra README for NPM (D-120), `d090c6c` Android host property (D-121).
- Landed since SPEC v0.20 (recorded in v0.21):
- `cargo audit` in the server gate (`f1abb0d`, no decision): the gate is `cargo fmt --check && cargo clippy --all-targets -- -D warnings && cargo audit && cargo test`; the pre-commit hook runs `cargo audit --no-fetch` only when `server/Cargo.lock` or a server `Cargo.toml` is staged. First run: 0 advisories over 319 crates.
- NPM rework (`4f69ae0`, `d05f13a`, `62f5279`, `9231a1d`; D-120): TLS terminates in the owner's Nginx Proxy Manager on the Home Assistant Pi; Caddy, cert-sync, the Caddyfile, the HA SSH key and the router forward to the Kvitt Pi are gone. Kvitt gets its own host name `KVITT_PUBLIC_HOST` (owner-chosen; a DuckDNS name is being set up), because `icehack.asuscomm.com` is NPM's proxy host for Home Assistant. Router external 8443 → HA Pi 443. The server publishes 8080 on the Kvitt Pi's LAN, `KVITT_TRUSTED_PROXY=${HA_HOST}` (NPM's IPv4, checked by `install.sh`). SPEC §10.5 rewritten (NPM proxy host, the Custom Nginx Configuration with D-117's headers and the `/i/` log exclusion, fallback: no LAN path for release builds). Rehearsed on the Mac: `nginx -t` against NPM v2.16.0 templates, 21 checks of the server behind nginx passed. The Gitea login interpolates the secret in the `run` line (the env form came through empty on act_runner 0.6.1).
- Android host property (`d090c6c`, D-121): `kvitt.publicHost` (Gradle property, else `android/local.properties`, else `kvitt.example.invalid`), `BuildConfig.PUBLIC_HOST`, the manifest placeholder `publicHost`, `RELEASE_SERVER_URL`, `@ReleaseServerUrl` in `:feature:onboarding`. 944 Android tests.
- Landed since SPEC v0.19 (recorded in v0.20):
- APK download rate limit (`96e4cbd`, D-119, security review L-3): E-6 `GET /apk/files/{file_name}` at most 10 requests per minute per client IP in its own window, not counted against the 60 of the other unauthenticated endpoints; a 404 counts; 429 `rate_limited` with `Retry-After`; E-5 and E-4 stay on the 60; unit and integration tests.
- Landed since SPEC v0.18 (recorded in v0.19):
@@ -15,50 +19,50 @@ Living document, rewritten in place (never a second copy). A new orchestrator se
- Server gap pass (`8660b7f`, D-113): `kvitt-server invite-partner` (0 ok, 2 household full, 3 no admin), so a scripted partner needs no second admin device; `push sent` info line; CLI tests (`invite-partner`, `migrate`, `healthcheck`, SIGTERM drain); `tools/contracts-check.py` 134 checks.
- Infra security (`7fb86f5`, D-117): Caddy log redaction of `/i/<code>` and query strings, CSP and three more response headers, Caddy non-root (uid 10002, read-only, caps dropped), `/certs` 0750, dev stack mounts only the FCM key, `/opt/kvitt/deployed-digests` with a WARNING on change, `kvitt_cicd` with `write:package` only, the plain root key (old variant B) not acceptable.
- M12f (`dc7a296`, `5b2ed37`, `2bd6b58`, `c366fbd`, `a0259ab`; D-109 to D-112, D-116): S-16 lost version read-only, S-22 "<version> (schema <n>)" from `/health`, `SettlementRepository` returns null instead of throwing, E-17 retry after each successful sync, invite links dropped while paired, private notifications on the lock screen, `localeFilters` en/sv, `:data` tests (429, D-040, WorkManager via `work-testing`), the home/account screenshot harness fix (stale `sv_dark` goldens re-recorded), 19 unused string keys removed (`strings.csv` 562 rows). Gate at landing: 941 tests. Release APK 10.0 MB.
- Validation reports (SPEC §11.4): `docs/validation/ac-coverage.md` (taken at `956908a`, `ed1b130` before the rewrite): 270 rows, 260 Done, 5 Partial, 1 Missing, 4 device-only; feature complete in code, not yet against §4's acceptance rule; its S-16 and S-22 mismatches are fixed by M12f. `docs/validation/security-review.md` (taken at `18c527f`): Critical 0, High 1, Medium 3, Low 10, Info 8; L-6 kept (D-116), L-7 fixed (D-116), L-1, L-2, L-5, L-9, M-2 fixed (D-117), L-8 fixed (`e70fe0d`); H-1 (`725901c`, D-115), M-3 (`baa5806`) and L-4 (`b985ce9`, D-114) fixed by the server overhaul; L-10 left as the spec says, decision pending (§4). Owner answers of 2026-10-19: M-1 no (backups stay unencrypted, rclone keeps the `drive` scope; accepted risk in SPEC §8.1 and §8.7), L-3 accepted and fixed (`96e4cbd`, D-119), the 16 MP and 512-character limits (D-114) confirmed; M-2's remaining question (the Home Assistant SSH key form) is moot with the NPM rework (D-120) and closed. H-1 must be fixed before port 8443 is opened.
- Validation reports (SPEC §11.4): `docs/validation/ac-coverage.md` (taken at `956908a`, `ed1b130` before the rewrite): 270 rows, 260 Done, 5 Partial, 1 Missing, 4 device-only; feature complete in code, not yet against §4's acceptance rule; its S-16 and S-22 mismatches are fixed by M12f. `docs/validation/security-review.md` (taken at `18c527f`): Critical 0, High 1, Medium 3, Low 10, Info 8; L-6 kept (D-116), L-7 fixed (D-116), L-1, L-2, L-5, L-9, M-2 fixed (D-117), L-8 fixed (`e70fe0d`); H-1 (`725901c`, D-115), M-3 (`baa5806`) and L-4 (`b985ce9`, D-114) fixed by the server overhaul; L-10 left as the spec says, decision pending (§4). Owner answers of 2026-10-19: M-1 no (backups stay unencrypted, rclone keeps the `drive` scope; accepted risk in SPEC §8.1 and §8.7), L-3 accepted and fixed (`96e4cbd`, D-119), the 16 MP and 512-character limits (D-114) confirmed; M-2's remaining question (the Home Assistant SSH key form) is moot with the NPM rework (D-120) and closed; D-120 also moves L-1 and the headers to NPM, makes L-2 moot, reverses the L-8 fix and replaces I-5's check with a conntrack check (dated notes in the review, SPEC §11.4). H-1 must be fixed before port 8443 is opened.
- Server overhaul (landed on `main` `8103dae` to `b985ce9`): shared helpers and one test harness, clippy pedantic wins; H-1 rate limiter (`725901c`, D-115), pictures at most 16 MP and 64 MB decoded (`baa5806`, D-114), `fcm_token` at most 512 characters (`b985ce9`, D-114); shared row inserts, invite creation and CLI helpers, `tests/common/` harness, synchronous `tasks::start`; 268 server tests; all crates at their latest versions. Recorded in SPEC v0.18.
- In flight:
- NPM rework (D-120 reserved, agent running): the owner runs Nginx Proxy Manager on the Home Assistant Pi, and NPM renews the Let's Encrypt certificate for `icehack.asuscomm.com`. Caddy and cert-sync are replaced by an NPM proxy host that forwards to the Kvitt Pi's server on LAN port 8080, so TLS terminates in NPM, the Kvitt Pi's compose stack loses `caddy` and `cert-sync`, and the `infra/pi/` bundle and its owner checklist change. Security review M-2 (the HA SSH key form for cert-sync) becomes moot and is closed. SPEC §10.5 and the other Caddy and cert-sync text are rewritten when D-120 lands (a docs pass records it; next free decision D-121). Open assumptions are §3 item 3.
- This docs pass (`agent-docs-v020`: SPEC v0.20, D-119, HANDOFF, `prompts/README.md`).
- `cargo-audit` in the server gate: the orchestrator is installing `cargo-audit` on the Mac; an agent adds it to the gate after the install.
- This docs pass (`agent-docs-v021`: SPEC v0.21, D-120, D-121, HANDOFF, `prompts/README.md`, the PLAN-M2c-pi note, dated notes in the security review).
- Dev server rebuild (orchestrator/agent).
- D-121 follow-ups: `:data` `DataModule.DEFAULT_SERVER_URL` is still the old literal `https://icehack.asuscomm.com:8443` (S-22's fallback) and moves to `AppBuildInfo.releaseServerUrl`; `infra/publish-apk.sh` passes `-Pkvitt.publicHost` from `KVITT_PUBLIC_HOST` (follow-up agent running).
- Gitea Actions: the first run (2026-10-19) got as far as the login (container, `apk add`, checkout worked) and failed with "password is empty" (env-based form); the run triggered by the `9231a1d` push tests the fixed login step (owner, §3 item 1).
- Earlier outcomes (details in SPEC §11.2, §11.3 and `docs/decisions.md`): M8 end-to-end test passed on the owner's S26 against the dev server on 2026-10-18 (pairing, scripted partner, offline expense synced, push by `fid` delivered); M8b to M8f UI fixes; the M2c-pi bundle `infra/pi/` (D-103 to D-106, owner checklist in `infra/README.md` "Pi: what the owner does"); release builds R8-shrunk and signed (alias `kvitt`, RSA 4096, v2).
- Compatibility: an APK from M12e on cannot pair or list devices against a pre-M12e server, so the server is rebuilt and redeployed before a new APK is installed (SPEC §14.8). The dev server already runs schema 3.
- Remote: `origin` on Gitea, SSH key works; pushed at `96e4cbd`. The owner created the Gitea secret `kvitt_cicd` and the Drive token (`infra/secrets/rclone.conf`, `[gdrive]`; folder id in `.env` as `RCLONE_FOLDER_ID`); the cert-sync key `infra/secrets/ha_ssh_key` exists (all git-ignored). The first push triggered the Gitea Actions workflow; its result is the owner's §3 item 1.
- Both gates exist: Android `tools/gradle-serial.sh check` (from `android/`), server `cargo fmt --check && cargo clippy --all-targets -- -D warnings && cargo test` (from `server/`). Both green on `main` as far as reported.
- Remote: `origin` on Gitea, SSH key works; pushed at `d090c6c`. The owner created the Gitea secret `kvitt_cicd` and the Drive token (`infra/secrets/rclone.conf`, `[gdrive]`; folder id in `.env` as `RCLONE_FOLDER_ID`). Unused on the Mac since D-120: `infra/secrets/ha_ssh_key` and `ha_ssh_key.pub` (git-ignored; may be deleted).
- Both gates exist: Android `tools/gradle-serial.sh check` (from `android/`), server `cargo fmt --check && cargo clippy --all-targets -- -D warnings && cargo audit && cargo test` (from `server/`, `cargo audit` since `f1abb0d`). Both green on `main` as far as reported (Android 944 tests at `d090c6c`).
- Importer: the dry run of 2026-10-18 reconciled on both real files (Vardag 1746 rows, Hemma 37 rows, 0 unmapped). The CSV files stay in `export/` (git-ignored).
- Development stack (M2c-dev): running on the Mac at `http://192.168.1.210:8080`, database at schema 3, data kept; backups `infra/dev/backups/kvitt-20261018T093143Z.sqlite.gz` and `kvitt-20261018T130443Z.sqlite.gz` (git-ignored). On this Podman setup run `infra/dev/up.sh --down`, then `infra/dev/up.sh`, to swap in a rebuilt image. Since `7fb86f5` `up.sh` copies only `fcm-service-account.json` into `infra/dev/secrets/` (git-ignored). The owner's phone is paired with it; a scripted partner exists (tokens held by the orchestrator outside the repo). Firebase project `kvitt-1337` with apps `se.hagstedt.kvitt` and `se.hagstedt.kvitt.debug`; service account `kvitt-fcm`.
- Installed on the owner's phone (SM-S942B): the debug app `se.hagstedt.kvitt.debug`, paired against the dev server, with the build whose checks are §3 item 2; the release app `se.hagstedt.kvitt`, installed 2026-10-19, not yet paired (its checks are §3 item 2 as well).
- Installed on the owner's phone (SM-S942B): the debug app `se.hagstedt.kvitt.debug`, paired against the dev server; the release app `se.hagstedt.kvitt`, installed 2026-10-19: About and the release app opening were confirmed; release pairing is deferred to production (release builds require `https`, AC-2.8).
- Disk: the MacBook had about 72 GB free during the v0.20 docs pass (2026-10-19); run `tools/clean-builds.sh` when it drops under about 5 GB.
## 2. Open work (one line each, with who owns it: agent or owner)
- Land `agent-docs-v020` (orchestrator): SPEC v0.20, D-119, HANDOFF, `prompts/README.md`.
- Land `agent-docs-v021` (orchestrator): SPEC v0.21, D-120, D-121, HANDOFF, `prompts/README.md`, PLAN-M2c-pi note, security-review notes.
- D-121 follow-ups (in flight): `:data` `DataModule.DEFAULT_SERVER_URL` → `AppBuildInfo.releaseServerUrl`; `infra/publish-apk.sh` passes `-Pkvitt.publicHost`.
- Literals left by D-120 and D-121 (agent, no plan yet): the server's `DEFAULT_PUBLIC_URL` (`https://icehack.asuscomm.com:8443`) and `DEFAULT_TRUSTED_PROXY` (`caddy`) in `config.rs`, `openapi.rs`'s `PUBLIC_SERVER_URL`, `contracts/openapi.yaml`'s server URL and the E-7 and E-12 examples, `server/README.md`, the fakes' `AboutState` URLs; harmless (compose always sets the variables) but they name Home Assistant's host; a contract change needs the §14.8 rule. The pinned-root config's domain (D-082) follows only if that fallback is prepared.
- `install.sh`'s last line says "Next: README step 7 (NPM proxy host)"; the NPM step is 6 (agent, one-line fix in `infra/pi/install.sh`).
- Possible firewall rule on the Kvitt Pi limiting access to port 8080 to the HA Pi (`HA_HOST`); not done, not decided (owner).
- `PLAN-M2c-pi.md` is superseded where it names Caddy and cert-sync (note at its top); SPEC §10 and `infra/README.md` are current.
- Restore drill AC-18.2 (`docs/validation/restore-drill.md`, M10; the one Missing row of the AC-coverage audit).
- AC-18.1, AC-20.1 and AC-16.3 shown on the Pi and the phones (M2c-pi run, M10).
- A `KvittWorkerFactory.createWorker` test (Android agent, no plan yet).
- Security review L-6: an S-1 design state if the owner wants a confirm step for an unusual server (OQ-6, default not in v1; owner, §4).
- Security review L-10: `forbidden` or `not_found` for the other user's device id (owner decision, §4).
- Tooling: `cargo-audit` in the server gate (agent, after the orchestrator's install on the Mac, in flight).
- State-class naming: `*Ui`, `*UiState` and `*State` are mixed across features; needs a decision if a convention is wanted (owner; no plan).
- Unverified risk: the workflow pushes a git tag; whether that re-triggers the path-filtered workflow is not known (watch the first runs).
- Land the NPM rework (D-120) when its agent reports, then a docs pass records D-120 and rewrites SPEC §10.5 and the Caddy and cert-sync text (orchestrator).
- The owner's answers to the phone checks of §3 item 2 (debug app checks; release app pairing with a device invite the orchestrator provides).
- Owner runs the Pi bundle on the Kvitt Pi: deferred by the owner on 2026-10-19 ("when it makes sense to move over to the Pi"); the bundle and its checklist are rewritten by D-120 first; record the router-forward answer in `infra/README.md` then (the cert-sync key-form question is closed).
- Plans (SPEC §11.2 v0.20), in dependency order; prompt `prompts/<id>.md`:
- M1, M2a, M2b, M2c-dev, M3, M4, M5, M6, M6b, M7a to M7g, M8 (phone test passed), M8b to M8f, M9, M12a to M12f, U1, U2, R8 release build, M2c-pi bundle, server gap pass, infra overhaul, infra security, the server overhaul and the Android overhaul: landed.
- Next: a server image with the H-1 fix before the router forward; the NPM rework (D-120) reshapes the Pi bundle; the owner runs it on the Kvitt Pi when he moves over (D-100, deferred); M10 First release after it (keystore per §10.6, D-101); M11 Import history after M9 and M10 (procedure §10.8 "Production run", D-025; each run must reconcile, D-045).
- Unverified risk: the workflow pushes a git tag; whether that re-triggers the path-filtered workflow is not known (watch the next runs).
- Owner runs the Pi bundle on the Kvitt Pi: deferred by the owner on 2026-10-19 ("when it makes sense to move over to the Pi"); the checklist is the D-120 one (nine steps, NPM at step 6, router at step 7).
- Plans (SPEC §11.2 v0.21), in dependency order; prompt `prompts/<id>.md`:
- M1, M2a, M2b, M2c-dev, M3, M4, M5, M6, M6b, M7a to M7g, M8 (phone test passed), M8b to M8f, M9, M12a to M12f, U1, U2, R8 release build, M2c-pi bundle and its NPM rework, server gap pass, infra overhaul, infra security, the server overhaul, the Android overhaul, the APK download rate limit, `cargo audit` and the Android host property: landed.
- Next: the owner runs the bundle on the Kvitt Pi when he moves over (D-100, deferred; a server image with the H-1 fix before the router forward); M10 First release after it (keystore per §10.6, D-101); M11 Import history after M9 and M10 (procedure §10.8 "Production run", D-025; each run must reconcile, D-045).
## 3. HUMAN steps pending (relay in chat)
1. Gitea, the repository, Actions: did the run "Build Kvitt images" triggered by the first push succeed (yes/no), and how long did it take? Parked by the owner ("come back to it"). If the run's tag push failed with `kvitt_cicd` limited to `write:package`: may `write:repository` be added (yes/no)?
2. Phone checks, answers pending (yes/no each):
1. Debug app (`se.hagstedt.kvitt.debug`): screen switches are instant, with no fade or slide; Inställningar, Om Kvitt shows the server version as "… (schema 3)"; the recurring editor shows Konto, Betalt av, Delning as three fixed rows.
2. Release app (`se.hagstedt.kvitt`, installed 2026-10-19): it opens to "Para ihop" without a crash; then pair it with the device invite the orchestrator provides; Home shows "Test"; add one expense and watch it sync.
3. NPM questions (for D-120; the agent proceeds on the assumptions): is `icehack.asuscomm.com` already a proxy host in Nginx Proxy Manager for another service (assumed free)? Does the router forward external 8443 to NPM's 443 or to a dedicated NPM port (assumed 443)?
4. Pi checklist, later: deferred by the owner ("when it makes sense to move over to the Pi"). The steps of `infra/README.md` "Pi: what the owner does" are rewritten by D-120 (no cert-sync key, the router forward goes to NPM) and relayed in chat when the owner moves over.
1. Gitea, the repository, Actions: the run "Build Kvitt images" triggered by the `9231a1d` push: did the step "Log in to Gitea Container Registry" pass (yes/no), and did the whole run succeed (yes/no)? If the step "Tag and Push Version" failed with a permission error: may `write:repository` be added to `kvitt_cicd` (yes/no)?
2. The DuckDNS name for Kvitt: tell the orchestrator once it exists. An agent then puts it into `infra/pi/.env` as `KVITT_PUBLIC_HOST` and into `android/local.properties` as `kvitt.publicHost` on the owner's word (D-120, D-121).
3. Phone checks: done. About ("… (schema 3)") confirmed and the release app opens confirmed (2026-10-19). Release pairing is deferred to production: release builds require `https` (AC-2.8), so the release app pairs only once the Pi runs behind NPM.
4. Pi checklist, later: deferred by the owner ("when it makes sense to move over to the Pi"). The nine steps of `infra/README.md` "Pi: what the owner does" (D-120: NPM proxy host at step 6, router forward 8443 → HA Pi 443 at step 7) are relayed in chat when the owner moves over.
## 4. Owner questions (asked, unanswered)
- §3 item 1: the workflow result and duration (parked), and `write:repository` if the tag push failed.
- §3 item 2: the debug and release app checks.
- §3 item 3: the two NPM questions (assumed: `icehack.asuscomm.com` free in NPM; external 8443 to NPM 443).
- Kvitt's host name (`KVITT_PUBLIC_HOST`, a DuckDNS name being set up; §3 item 2).
- `write:repository` for `kvitt_cicd`, only if the workflow's tag push fails (§3 item 1).
- OQ-6 (SPEC §12): a confirm step on S-1 for an unusual server; default "not in v1".
- Security review L-10: should a member get `not_found` instead of `forbidden` for a device id of the other user (hides whose it is)? As built and as the spec says: `forbidden`.
- State-class naming (`*Ui`, `*UiState`, `*State` mixed): fix one convention, or leave as built?
@@ -67,9 +71,9 @@ Living document, rewritten in place (never a second copy). A new orchestrator se
- Nothing.
## 6. Numbering
- Decisions: D-001 to D-119 used, except D-008 and D-009 (never used); D-120 reserved by the running NPM agent. Next free: **D-121**.
- Spec version: v0.20 written; next **v0.21**.
- Server schema version **3** (migration 0003); next **4**. Room schema version **1** (`android/data/schemas/se.hagstedt.kvitt.data.db.KvittDatabase/1.json`); next **2**. Both unchanged by v0.20.
- Decisions: D-001 to D-121 used, except D-008 and D-009 (never used); nothing reserved. Next free: **D-122**.
- Spec version: v0.21 written; next **v0.22**.
- Server schema version **3** (migration 0003); next **4**. Room schema version **1** (`android/data/schemas/se.hagstedt.kvitt.data.db.KvittDatabase/1.json`); next **2**. Both unchanged by v0.21.
- Sync protocol **1**, push payload **1**.
## 7. How the work is run
+8 -6
View File
@@ -27,7 +27,7 @@ entry and the old one a strike-through plus a pointer. The spec is updated in th
| D-020 | 2026-10-17 | No bundled font: the app uses the Material 3 default typeface. No licence step. | PLAN-M6 open point (b); avoids a font licence clearance. | SPEC §6.2, §11.2, §13.4; PLAN-M2b, PLAN-M6, PLAN-U1, PLAN-U2 |
| D-021 | 2026-10-17 | `sync_mutations` pruning runs daily at 04:00 Europe/Stockholm (DST-aware), deleting rows whose `received_at` is older than 90 days. | PLAN-M4 open point (c); the spec said only "daily". | SPEC §14.1 (rules text, not the DDL), §14.4.9, §11.2; PLAN-M4 |
| D-022 | 2026-10-17 | Development infrastructure (M2c-dev, landed `c885aba`). No new CLI installs: Firebase project and Android apps through the Firebase Management REST API with a `gcloud` access token, no Firebase CLI; `rclone` belongs to M2c-pi. The agent may write git-ignored secrets, `google-services.json` included, under `infra/secrets/`; M8 copies it into `android/app/` with `infra/dev/copy-google-services.sh`. Phone health check against the dev server is a HUMAN step. Gate: `docker compose -f infra/dev/compose.yaml --env-file infra/dev/.env.example config`; `shellcheck $(find infra -name '*.sh')`. As built: dev stack without Caddy or TLS, bind-mounted data folder, the whole `infra/secrets/` mounted read-only at `/secrets` (a missing key file means push disabled), env file `infra/dev/.env.example` (the Pi's stays `infra/env.example` for M2c-pi), `kvitt-server healthcheck` used by compose, backups with `sqlite3 .backup` (§10.9). The server must start when `KVITT_TRUSTED_PROXY` does not resolve. Firebase on `kvitt-1337` with apps `se.hagstedt.kvitt` and `se.hagstedt.kvitt.debug`; service account `kvitt-fcm@kvitt-1337.iam.gserviceaccount.com` with `roles/firebasecloudmessaging.admin` (`roles/firebasemessaging.admin` does not exist). | Owner's tools already on the Mac; M2c-dev report. | SPEC §10 intro, §10.4, §10.7, §11.2, §11.3, §14.7; PLAN-M2c-dev, PLAN-M8 |
| D-023 | 2026-10-17 | M2c-pi: ~~key-based SSH to the Kvitt Pi and the Gitea credentials are HUMAN steps; the allowances to act on the Kvitt Pi, the Home Assistant Pi, Gitea and Google Drive are requested from the owner at the start of that sprint;~~ (revised by D-100: no SSH from the Mac to the Pis and no agent allowances; the owner runs the `infra/pi/` bundle; the Gitea credentials stay a HUMAN step); the Gitea Actions workflow lives at `.gitea/workflows/`, owned by M2c-pi; secret files on the Pi (`ha_ssh_key`, `ha_known_hosts`, ~~`rclone.conf`~~) mode 0600 owned by uid 1000, the uid of the `cert-sync` ~~and `backup`~~ container (revised by D-105: `backup` runs as uid 10001 and `rclone.conf` is owned by 10001:10001); adding `gitea.lan` to `insecure-registries` is its own HUMAN step. | PLAN-M2c-pi open points (a) to (c). | SPEC §10.2, §10.3, §11.2, §11.3, §14.7; PLAN-M2c-pi |
| D-023 | 2026-10-17 | M2c-pi: ~~key-based SSH to the Kvitt Pi and the Gitea credentials are HUMAN steps; the allowances to act on the Kvitt Pi, the Home Assistant Pi, Gitea and Google Drive are requested from the owner at the start of that sprint;~~ (revised by D-100: no SSH from the Mac to the Pis and no agent allowances; the owner runs the `infra/pi/` bundle; the Gitea credentials stay a HUMAN step); the Gitea Actions workflow lives at `.gitea/workflows/`, owned by M2c-pi; secret files on the Pi (~~`ha_ssh_key`, `ha_known_hosts`~~, ~~`rclone.conf`~~) mode 0600 owned by uid 1000, the uid of the ~~`cert-sync`~~ ~~and `backup`~~ container (revised by D-105: `backup` runs as uid 10001 and `rclone.conf` is owned by 10001:10001; revised by D-120: cert-sync and its secrets `ha_ssh_key` and `ha_known_hosts` are removed); adding `gitea.lan` to `insecure-registries` is its own HUMAN step. | PLAN-M2c-pi open points (a) to (c). | SPEC §10.2, §10.3, §11.2, §11.3, §14.7; PLAN-M2c-pi; revised by D-100, D-105, D-120 |
| D-024 | 2026-10-17 | Release keystore: `keytool -genkeypair -v -keystore kvitt-release.jks -alias kvitt -keyalg RSA -keysize 4096 -validity 10000`, created by the owner ~~and kept outside the repository~~ (was `android/keystore/release.jks`; revised by D-101: at the repository root, git-ignored by `*.jks`); `local.properties` holds its absolute path. | §10.6 did not fix the key parameters (PLAN-M10 H-1); a keystore outside every checkout cannot be committed or deleted with a worktree. | SPEC §10.6, §11.2, §11.3; PLAN-M10 |
| D-025 | 2026-10-17 | Production Splitwise import (M11): a backup first; user ids resolved by display name from the database; CSVs copied to the Pi into `/opt/kvitt/import/`, bind-mounted read-only into `server` as `/import`; the import runs as `docker compose run --rm --entrypoint kvitt-import server ...` with the server stopped (`--entrypoint` because the image's entrypoint is `kvitt-server`); CSVs deleted afterwards. The category table stays at `server/crates/kvitt-import/data/splitwise_categories.csv`. | PLAN-M11 open points (a) to (c); PLAN-M9 table location. | SPEC §10.2, §10.8, §11.2, §11.3; PLAN-M2c-pi, PLAN-M9, PLAN-M11 |
| D-026 | 2026-10-17 | A 409 `cursor_ahead` on pull is treated as epoch recovery: cursor 0, epoch cleared, every local expense and settlement marked pending with a new `pending_mutation_id`, cycle restarted. | M5 report (`8f0e156`): a cursor ahead of the server means the server lost rows, as after a restore without `rotate-epoch`. | SPEC §14.4.2 |
@@ -107,10 +107,10 @@ entry and the old one a strike-through plus a pointer. The spec is updated in th
| D-100 | 2026-10-18 | No SSH from the Mac to the Pis. M2c-pi produces a self-contained bundle `infra/pi/` (compose file, Caddyfile, env template, install, deploy, bootstrap and backup scripts, a README checklist) that the owner copies to the Kvitt Pi and runs there. The OQ-3 values (Kvitt Pi address, Home Assistant Pi address, SSH port and user, certificate directory, SSH add-on) become placeholders the owner fills in the bundle's `.env`; OQ-3 is answered by this. Where the spec has a Mac script act on the Pi (`infra/deploy.sh`, `infra/publish-apk.sh`), the step becomes a bundle script the owner runs; M2c-pi fixes the layout inside `infra/pi/`, the script names and how images and APKs reach the Pi. M2c-pi runs as the last step before M10, after the app is proven on the phone against the development server. Agents never need allowances for the Kvitt Pi, the Home Assistant Pi, Gitea or Google Drive. Revises D-023 and the M2c-pi plan; D-001 stands. Fulfilled by D-103 to D-106 (bundle landed `2476667`). | Owner decision 2026-10-18: no agent access to the home servers; the Pi is set up once, by hand, from a checklist. | SPEC §10 intro, §10.1, §10.10, §11.1, §11.2 (M2c-pi, M10), §11.3, §12 (OQ-3); PLAN-M2c-pi |
| D-101 | 2026-10-18 | The release keystore `kvitt-release.jks` (created with the D-024 `keytool` command) lives at the repository root of the main checkout, git-ignored by the `*.jks` pattern of `.gitignore` and never committed; `android/local.properties` references it by absolute path in `kvitt.release.storeFile` with the other three `kvitt.release.*` keys. The owner keeps a copy of the file and its password in their password manager. Revises D-024 ("outside the repository"). | Owner decision 2026-10-18; the `*.jks` ignore rule already keeps the file out of git. | SPEC §10.6, §11.2 (M10), §11.3; PLAN-M10 |
| D-102 | 2026-10-18 | S-7 editor options as fixed rows. Below the category quick row, account, date, payer and split are four full-width rows in a fixed order (Konto, Datum, Betalt av, Delning) instead of a wrapping chip row, so no option moves when a value changes. Rows form a segmented group (outer corners 24, inner 8, gap 3) in surface-container-high (secondary-container while the split section is open), each at least 48 dp and tappable in full. A row has a leading element in a 24 dp box, a label (label-large, muted) in a column as wide as the widest label, the value (body-large, one line, ellipsis) and the trailing affordance at the right edge (expand_more/expand_less, swap_horiz for the payer). At ≥ 1.5× font scale the label sits over the value, which takes up to two lines, and every row keeps the two-line height. TalkBack sentences are unchanged. S-9 and S-19 keep their chips (S-19 follows in M8e, D-107). Also: connected buttons get 8 dp side padding so 'Household' fits half of S-11's filter bar at 1x. Landed by M8d (`bacbd18`): `EditorOptionRows` in `:core:design`, `KvittDimens.connectedButtonSide`. | Owner finding on S-7, 2026-10-18: the chips moved when a value changed. | SPEC §13.3 (S-7), §13.4; `docs/ux/design-spec.md` §3.7, §3.10 |
| D-103 | 2026-10-18 | The M2c-pi bundle `infra/pi/`: `compose.yaml` (was `infra/compose.yaml`), `.env.example` (was `infra/env.example`), `Caddyfile`, and the on-Pi scripts `install.sh` (idempotent: creates `/opt/kvitt/{secrets,import,apk}`, installs the secrets with their modes, adds `gitea.lan` to `insecure-registries`, sets the volume owners, starts the stack), `deploy.sh <tag> [--skip-backup]` (backs up before every deploy, because the Pi cannot tell whether a tag carries a migration; keeps `/opt/kvitt/previous-tag` for rollback), `bootstrap.sh`, `backup-now.sh`, `install-apk.sh`, helpers in `lib.sh`. APK path: `infra/publish-apk.sh` on the Mac writes `infra/pi/out/apk/` (`version_code` = `git rev-list --count HEAD`, `version_name` = `YYYY.MM.DD-<short sha>` in Europe/Stockholm, passed as `-Pkvitt.versionCode` / `-Pkvitt.versionName`; copies `mapping.txt` as `kvitt-<version_code>.mapping.txt`; refuses a dirty tree and an unsigned APK); the owner `scp`s it to `/opt/kvitt/apk/`; `install-apk.sh` copies it into the `data` volume, writing `latest.json` last and keeping the newest 5; mappings stay outside the volume. `.env` keys grow by `KVITT_REGISTRY_USER`, `KVITT_REGISTRY_TOKEN` (read by `install.sh` only) and the informational `KVITT_PI_HOST`, `KVITT_PI_SSH_USER`. The §11.2 gate becomes `docker compose -f infra/pi/compose.yaml config`. Landed `2476667`. Fulfils D-100. | M2c-pi report: D-100 left the layout, script names and the image and APK paths to M2c-pi. | SPEC §10 intro, §10.2, §10.3, §10.6, §10.7, §10.10, §11.2 (M2c-pi), §11.3, §14.7; `infra/README.md` |
| D-104 | 2026-10-18 | Images `${KVITT_REGISTRY}/kvitt-server`, `kvitt-cert-sync`, `kvitt-backup` with `KVITT_REGISTRY=gitea.lan/jacobh2` (was `${KVITT_REGISTRY}/kvitt/<name>`). `.gitea/workflows/build-images.yaml` follows the owner's working example: runner `host` (a Raspberry Pi 3, 1 GB; builds run on the runner itself), Alpine `apk add git docker-cli docker-cli-buildx`, a plain-HTTP buildx builder created once and reused, secret `kvitt_cicd`; triggers on `server/**`, `infra/cert-sync/**`, `infra/backup/**`, `.gitea/workflows/**` and `workflow_dispatch`; tags `v1.MINOR.PATCH` (the next patch after the newest tag) plus `latest`, the git tag pushed afterwards; `--build-arg CARGO_BUILD_JOBS=1` (honoured by `server/Dockerfile` since `b4d9467`; first build duration unknown). `infra/build-images.sh` on the Mac is the fallback (default tag the short SHA; builds locally and prints the push commands when `infra/secrets/registry.env` is absent). Both sidecars schedule with supercronic instead of busybox crond (crond runs no job as non-root, tested); the cert-sync image adds `openssl`. Pinned versions (2026-10-18): `caddy:2.11.7@sha256:685ae3d7c34d9b5f8caeddc5d14051db7bbd88ff1f5a2fa0c16bd8ca328fb7af`, `alpine:3.24.2@sha256:294b683cb724975bec92580e1e685676bd4b50bda910ddb8c51d4cabeaec77e6`, openssh-client-default 10.3_p1, curl 8.22.0, openssl 3.5.9, supercronic 0.2.49; Docker Engine from Docker's Debian arm64 apt repository (docs.docker.com: the Raspberry Pi OS page is 32-bit only), `docker-ce` 5:29.8.2. | M2c-pi report: the owner's example workflow pushes under the owner path with flat image names; busybox crond did not run jobs as a non-root user. | SPEC §10.1, §10.2, §10.3, §12 (OQ-1); `.gitea/workflows/build-images.yaml`, `infra/README.md` "Versions" |
| D-105 | 2026-10-18 | `backup` runs as uid 10001 on the Pi too, so it can open the WAL database (`sqlite3 .backup` needs the `-shm` file the server owns); `/opt/kvitt/secrets/rclone.conf` mode 0600 owned 10001:10001. `cert-sync` stays uid 1000 with `ha_ssh_key`, `ha_known_hosts` 0600 owned by 1000. `backup.sh` uses a temporary copy of the read-only `rclone.conf` so rclone can save refreshed tokens. `install.sh` chowns `/data`, `/data/apk`, `/backups` to 10001 and `/certs` to 1000 once. Reconciles SPEC §10.2; revises D-023 for `rclone.conf`. | M2c-pi report: uid 1000 could not back up the live database without loosening the server's files. | SPEC §10.2, §10.9; `infra/README.md` "Backup uid" |
| D-106 | 2026-10-18 | `sync.sh` always requests the `tar -C <dir> -cf - fullchain.pem privkey.pem` stream, so both `authorized_keys` variants (A restricted, B plain) work unchanged; it refuses a key not matching its certificate and a certificate expiring before the installed one; an unreachable Caddy admin API at first start is not a failure (Caddy loads the files when it starts). `backup.sh` uploads every pending `kvitt-*` / `pictures-*` file including leftovers of failed runs, confirms each by name and size with `rclone lsf`, deletes the local copies only after all are confirmed, and writes `last-success` only after that. | M2c-pi report: the add-on's support for `authorized_keys` options is unknown; rehearsal of renew, refuse and failed-upload cases on the Mac. | SPEC §10.5.1, §10.9 |
| D-103 | 2026-10-18 | The M2c-pi bundle `infra/pi/`: `compose.yaml` (was `infra/compose.yaml`), `.env.example` (was `infra/env.example`), ~~`Caddyfile`~~ (removed by D-120), and the on-Pi scripts `install.sh` (idempotent: creates `/opt/kvitt/{secrets,import,apk}`, installs the secrets with their modes, adds `gitea.lan` to `insecure-registries`, sets the volume owners, starts the stack), `deploy.sh <tag> [--skip-backup]` (backs up before every deploy, because the Pi cannot tell whether a tag carries a migration; keeps `/opt/kvitt/previous-tag` for rollback), `bootstrap.sh`, `backup-now.sh`, `install-apk.sh`, helpers in `lib.sh`. APK path: `infra/publish-apk.sh` on the Mac writes `infra/pi/out/apk/` (`version_code` = `git rev-list --count HEAD`, `version_name` = `YYYY.MM.DD-<short sha>` in Europe/Stockholm, passed as `-Pkvitt.versionCode` / `-Pkvitt.versionName`; copies `mapping.txt` as `kvitt-<version_code>.mapping.txt`; refuses a dirty tree and an unsigned APK); the owner `scp`s it to `/opt/kvitt/apk/`; `install-apk.sh` copies it into the `data` volume, writing `latest.json` last and keeping the newest 5; mappings stay outside the volume. `.env` keys grow by `KVITT_REGISTRY_USER`, `KVITT_REGISTRY_TOKEN` (read by `install.sh` only) and the informational `KVITT_PI_HOST`, `KVITT_PI_SSH_USER`. The §11.2 gate becomes `docker compose -f infra/pi/compose.yaml config`. Landed `2476667`. Fulfils D-100. | M2c-pi report: D-100 left the layout, script names and the image and APK paths to M2c-pi. | SPEC §10 intro, §10.2, §10.3, §10.6, §10.7, §10.10, §11.2 (M2c-pi), §11.3, §14.7; `infra/README.md` |
| D-104 | 2026-10-18 | Images `${KVITT_REGISTRY}/kvitt-server`, ~~`kvitt-cert-sync`~~ (removed by D-120), `kvitt-backup` with `KVITT_REGISTRY=gitea.lan/jacobh2` (was `${KVITT_REGISTRY}/kvitt/<name>`). `.gitea/workflows/build-images.yaml` follows the owner's working example: runner `host` (a Raspberry Pi 3, 1 GB; builds run on the runner itself), Alpine `apk add git docker-cli docker-cli-buildx`, a plain-HTTP buildx builder created once and reused, secret `kvitt_cicd`; triggers on `server/**`, ~~`infra/cert-sync/**`~~ (dropped by D-120), `infra/backup/**`, `.gitea/workflows/**` and `workflow_dispatch`; tags `v1.MINOR.PATCH` (the next patch after the newest tag) plus `latest`, the git tag pushed afterwards; `--build-arg CARGO_BUILD_JOBS=1` (honoured by `server/Dockerfile` since `b4d9467`; first build duration unknown). `infra/build-images.sh` on the Mac is the fallback (default tag the short SHA; builds locally and prints the push commands when `infra/secrets/registry.env` is absent). Both sidecars schedule with supercronic instead of busybox crond (crond runs no job as non-root, tested); ~~the cert-sync image adds `openssl`~~. Pinned versions (2026-10-18): ~~`caddy:2.11.7@sha256:685ae3d7c34d9b5f8caeddc5d14051db7bbd88ff1f5a2fa0c16bd8ca328fb7af`~~, `alpine:3.24.2@sha256:294b683cb724975bec92580e1e685676bd4b50bda910ddb8c51d4cabeaec77e6`, ~~openssh-client-default 10.3_p1, curl 8.22.0, openssl 3.5.9~~, supercronic 0.2.49 (revised by D-120: no Caddy pin and no cert-sync image or packages; only the backup sidecar uses supercronic); Docker Engine from Docker's Debian arm64 apt repository (docs.docker.com: the Raspberry Pi OS page is 32-bit only), `docker-ce` 5:29.8.2. | M2c-pi report: the owner's example workflow pushes under the owner path with flat image names; busybox crond did not run jobs as a non-root user. | SPEC §10.1, §10.2, §10.3, §12 (OQ-1); `.gitea/workflows/build-images.yaml`, `infra/README.md` "Versions" |
| D-105 | 2026-10-18 | `backup` runs as uid 10001 on the Pi too, so it can open the WAL database (`sqlite3 .backup` needs the `-shm` file the server owns); `/opt/kvitt/secrets/rclone.conf` mode 0600 owned 10001:10001. ~~`cert-sync` stays uid 1000 with `ha_ssh_key`, `ha_known_hosts` 0600 owned by 1000.~~ (cert-sync removed by D-120) `backup.sh` uses a temporary copy of the read-only `rclone.conf` so rclone can save refreshed tokens. `install.sh` chowns `/data`, `/data/apk`, `/backups` to 10001 ~~and `/certs` to 1000~~ once. Reconciles SPEC §10.2; revises D-023 for `rclone.conf`. | M2c-pi report: uid 1000 could not back up the live database without loosening the server's files. | SPEC §10.2, §10.9; `infra/README.md` "Backup uid" |
| D-106 | 2026-10-18 | ~~`sync.sh` always requests the `tar -C <dir> -cf - fullchain.pem privkey.pem` stream, so both `authorized_keys` variants (A restricted, B plain) work unchanged; it refuses a key not matching its certificate and a certificate expiring before the installed one; an unreachable Caddy admin API at first start is not a failure (Caddy loads the files when it starts).~~ (revised by D-120: cert-sync removed) `backup.sh` uploads every pending `kvitt-*` / `pictures-*` file including leftovers of failed runs, confirms each by name and size with `rclone lsf`, deletes the local copies only after all are confirmed, and writes `last-success` only after that. | M2c-pi report: the add-on's support for `authorized_keys` options is unknown; rehearsal of renew, refuse and failed-upload cases on the Mac. | SPEC §10.5.1, §10.9 |
| D-107 | 2026-10-18 | S-19's options as fixed rows, extending D-102. The recurring template editor replaces its wrapping chips row (Konto · Betalt av · Delning) with `EditorOptionRows`: three full-width rows in the fixed order Konto, Betalt av, Delning, with no Datum row. They look and behave as in S-7: label column, value on one line with an ellipsis, label over value with a two-line value at ≥ 1.5× font scale, and Delning in secondary-container while the split section is open. A row's place and size never depend on its value. The schedule fields stay in the "Schema" section. TalkBack sentences are unchanged. A Robolectric layout test asserts identical row bounds for short and long values at 100 % and 200 %. The chip composables stay in `:core:design` for S-9 (`EditorChipRow`, `DateChip`); `AccountChip`, `PayerChip`, `SplitChip` were removed in `1328afe`. Landed by M8e (`a59abeb`). | Owner finding (D-102): chips moved when a value changed; S-19 had the same chips row. | SPEC §13.3 (S-19); `docs/ux/design-spec.md` §3.7, S-19 |
| D-108 | 2026-10-18 | Navigation is instant. Every destination change (pushed screens, bottom-destination switches, S-7 and S-9 and the other editors, onboarding and the session gate) has no enter, exit, pop or size transition (`EnterTransition.None` / `ExitTransition.None`, no `SizeTransform`, set once on the app's `NavHost`, `NavTransitions` in `:app`). Replaces the Navigation bullet of design-spec §2.11. Motion inside a screen (ripples, state layers, expanding sections, FAB extend/collapse, bottom-sheet open, the seesaw and bunting) is unchanged. Not a contract change. Landed by M8f (`956908a`). | Owner finding on the S26, 2026-10-18: the default 700 ms fades made the app feel slow. | SPEC §13.2; `docs/ux/design-spec.md` §2.11 |
| D-109 | 2026-10-18 | S-16 shows a rejected change's lost version read-only. `Rejection` gains `lost: RejectedVersion?`, with `RejectedVersion(row: LedgerRow, accountName: String?, payer: User?, mode: SplitMode?, shares: List<Share>, notes: String, deleted: Boolean)` decoded in `:data` from `sync_rejections.rejected_payload` (the entity as pushed; `row` is the ledger row it would have been, `pending` false; for a settlement `accountName`, `payer` and `mode` are null and `shares` empty). `lost` is null when the payload cannot be decoded or its category or a user is not on the phone. "Öppna" on a rejection opens a dialog "Din version" (`home_sync_lost_title`, body `home_sync_lost_body`) with the read-only ledger row, account, payer, split and notes, the badge "Du tog bort den här" (`home_sync_lost_deleted`) for a refused deletion, and "Visa nuvarande version" (`home_sync_lost_show_current`), which opens S-8 or S-10; without `lost` it opens the current version as before. The `:core:design` ledger rows take a nullable `onClick` (null = read-only). Landed by M12f (`dc7a296`). UI contract change. | AC-coverage audit: S-16's "open a rejected change (read-only)" opened the current editable row, not the lost change. | SPEC §13.3 (S-16), §13.4, §14.2.4, §14.4.6; `docs/ux/design-spec.md` S-16 |
@@ -121,6 +121,8 @@ entry and the old one a strike-through plus a pointer. The spec is updated in th
| D-114 | 2026-10-18 | Upload and token limits. A picture upload (E-9, AC-4.4) must also be at most 16 000 000 pixels (width × height, for example 4000 × 4000) and decode to at most 64 MB (16 MP × 4 bytes, so a 16-bit image is refused above 8 MP), checked from the image header before decoding with the decoder limits set as well (`MAX_PIXELS`, `MAX_DECODE_ALLOC` in `src/api/pictures.rs`), refused with `image_invalid`; the 8000 px per side and 5 MiB limits stay. `fcm_token` at E-17 is 1 to 512 characters after trimming (`FCM_TOKEN_MAX`); longer is `invalid_request` with field `fcm_token`, like the empty string (D-034); a Firebase Installation ID has 22 characters. Landed by the server overhaul (`baa5806`, `b985ce9`). | Security review M-3: 8000 × 8000 decodes to 256 MB RGBA, above the 160 MiB container. L-4: a device could store a token up to the 1 MiB body limit, sent to FCM in every push. | SPEC §4 (AC-4.4), §9.6, §14.3.4 (E-17) |
| D-115 | 2026-10-18 | Rate-limit token acceptance, revising D-032: a bearer token is accepted only when a response to it is 2xx; a request with a token not yet accepted counts against its client IP's unauthenticated limit before it runs and is refused with `rate_limited` once the IP is over; a 2xx gives that count back; 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 counts against the IP exactly once. Landed by the server overhaul (`725901c`), with unit tests per rule and an integration test of random tokens without the protocol header (60 × 426, then 429, the paired device unaffected). | Security review H-1: a random token without `X-Kvitt-Protocol` (426) became accepted for good, bypassed the IP limit and was never pruned, so a stranger could grow the in-memory map without bound. | SPEC §8.6 |
| D-116 | 2026-10-18 | Notifications are posted with `VISIBILITY_PRIVATE` and a public version carrying the generic "Kvitt" / "Open Kvitt to see it" (`notif_generic_title`, `notif_generic_body`), so a secure lock screen shows no description, amount, account or name (new AC-13.9; `NotificationPrivacyTest`). L-6 (a link can prefill any server on S-1) stays as built: S-1 opens its server row expanded when the link's server is not the release default, release builds require `https://`, and a paired phone ignores links (D-084); a confirmation step would need a design state and is not in v1 (OQ-6). Landed by M12f (`2bd6b58`). | Security review L-7 and L-6. | SPEC §4 (AC-2.6, AC-2.8, AC-13.9), §12 (OQ-6) |
| D-117 | 2026-10-18 | Infra security fixes, landed `7fb86f5`. (1) Caddy's access log and error log (`log default`) filter `request>uri` with ``regexp `^(/i/)[^?]+\|([?]).+` `${1}${2}REDACTED` `` (`wrap json`), so invite codes and the `?from=` name never reach a log; `Authorization` stays redacted by default. (2) Deferred response headers `Content-Security-Policy "default-src 'none'; style-src 'unsafe-inline'; base-uri 'none'; form-action 'none'; frame-ancestors 'none'"`, `X-Content-Type-Options nosniff`, `X-Frame-Options DENY`, `Referrer-Policy no-referrer`. (3) The `caddy` container runs as `10002:10002` with `group_add` 1000, `read_only`, `cap_drop: [ALL]` plus `cap_add: [NET_BIND_SERVICE]` (the image's binary has that file capability and cannot be exec'd without it), `no-new-privileges`, tmpfs `/config` and `/tmp`; `install.sh` sets `/certs` 0750 1000:1000 and `caddy_data` to 10002 (extends D-105). (4) The development stack mounts `infra/dev/secrets/` (git-ignored), into which `up.sh` copies only `fcm-service-account.json`. (5) `deploy.sh` and `install.sh` print each pulled image's digest and append `<UTC time> <KVITT_TAG> <image reference> <digest>` to `/opt/kvitt/deployed-digests`, with a WARNING when a recorded reference changed digest; the workflow prints the pushed digests. (6) `kvitt_cicd` needs only `write:package` (the tag push uses the checkout credentials). (7) The plain root key (variant B) for cert-sync is no longer acceptable: variant A (restricted) or a dedicated non-root user. (8) One-off check after the router forward that Caddy's `remote_ip` shows public addresses. | Security review at `18c527f`: L-1, L-2, L-5, L-9, M-2, I-5. | SPEC §8.7, §8.9, §10 intro, §10.2, §10.3, §10.5.1, §10.5.2, §10.10; `infra/pi/`, `infra/dev/`, `.gitea/workflows/build-images.yaml`, `infra/README.md` |
| D-117 | 2026-10-18 | Infra security fixes, landed `7fb86f5`. (1) ~~Caddy's access log and error log (`log default`) filter `request>uri` with ``regexp `^(/i/)[^?]+\|([?]).+` `${1}${2}REDACTED` `` (`wrap json`), so invite codes and the `?from=` name never reach a log; `Authorization` stays redacted by default.~~ (revised by D-120: NPM's `location ^~ /i/` without access or error log; NPM logs no request headers) (2) ~~Deferred~~ Response headers `Content-Security-Policy "default-src 'none'; style-src 'unsafe-inline'; base-uri 'none'; form-action 'none'; frame-ancestors 'none'"`, `X-Content-Type-Options nosniff`, `X-Frame-Options DENY`, `Referrer-Policy no-referrer` (revised by D-120: set at NPM's server level with the server's copies hidden). (3) ~~The `caddy` container runs as `10002:10002` with `group_add` 1000, `read_only`, `cap_drop: [ALL]` plus `cap_add: [NET_BIND_SERVICE]` (the image's binary has that file capability and cannot be exec'd without it), `no-new-privileges`, tmpfs `/config` and `/tmp`; `install.sh` sets `/certs` 0750 1000:1000 and `caddy_data` to 10002 (extends D-105).~~ (removed by D-120) (4) The development stack mounts `infra/dev/secrets/` (git-ignored), into which `up.sh` copies only `fcm-service-account.json`. (5) `deploy.sh` and `install.sh` print each pulled image's digest and append `<UTC time> <KVITT_TAG> <image reference> <digest>` to `/opt/kvitt/deployed-digests`, with a WARNING when a recorded reference changed digest; the workflow prints the pushed digests. (6) `kvitt_cicd` needs only `write:package` (the tag push uses the checkout credentials). (7) ~~The plain root key (variant B) for cert-sync is no longer acceptable: variant A (restricted) or a dedicated non-root user.~~ (moot since D-120: no Home Assistant SSH key) (8) ~~One-off check after the router forward that Caddy's `remote_ip` shows public addresses.~~ (revised by D-120: the conntrack check of SPEC §10.10) | Security review at `18c527f`: L-1, L-2, L-5, L-9, M-2, I-5. | SPEC §8.7, §8.9, §10 intro, §10.2, §10.3, §10.5.1, §10.5.2, §10.10; `infra/pi/`, `infra/dev/`, `.gitea/workflows/build-images.yaml`, `infra/README.md` |
| D-118 | 2026-10-18 | Shared test support lives in `:core:design`'s test fixtures (`src/testFixtures`, package `se.hagstedt.kvitt.design.testing`: `Variant`, `TestClock`, `TestEnvironment`, `ScreenSnapshots` with capture modes `Root`/`Screen`/`ScreenWhenIdle`, `MainDispatcherRule`, `withMain`), consumed as `testImplementation(testFixtures(project(":core:design")))`; not a module, so the §6.2 graph is unchanged; the architecture check rejects test fixtures outside test configurations. The convention plugin `kvitt.android.screenshots` wires Roborazzi and the test libraries; feature build scripts carry no screenshot block. Shared failure texts (`format/Failures.kt`: `FAILURE_NETWORK`, `commonFailureText`), limits (`format/Limits.kt`: `BP_TOTAL`, `DISPLAY_NAME_MAX`, `DESCRIPTION_MAX`, `NOTES_MAX`, `FIRST_DATE`, `DAYS_AHEAD_MAX`) and `savedSplitLabel` live in `:core:design`. `KvittMotion` holds only the seesaw spring; `KvittIcon.Description` and unused dimens/shapes removed; catalog entries `roborazzi-junit-rule`, `roborazzi-compose`, `androidx-hilt-navigation-compose`, `androidx-lifecycle-viewmodel-compose` removed; one `CopyCategoryKeywords` task shared by `:data` and `:core:fakes`. 941 tests, no golden changed, release APK 10.46 MB. Landed by the Android overhaul (`e4e6e5d` to `c582363`). | Eight modules each had a copy of the same screenshot harness, and M12f had to fix the same capture bug in two of them; test fixtures rather than a `:core:testing` module let `:core:design`'s own tests use them, add no module to the §6.2 graph and are treated by lint as test sources. The failure-text map, the limits (`BP_TOTAL` in five copies) and the saved-split label were duplicated across features; the removed tokens, icon and catalog entries were never referenced. | SPEC §6.2 (`:core:design` row, Screenshot tests), §11.1, §11.3; `docs/ux/design-spec.md` §2.11 and the icon table; `android/README.md` |
| D-119 | 2026-10-19 | APK download rate limit. E-6 `GET /apk/files/{file_name}` is limited to 10 requests per minute per client IP in its own fixed window (`APK_DOWNLOAD_LIMIT` in `src/api/rate_limit.rs`), separate from and not counted against the 60 per minute of the other unauthenticated endpoints; the client IP as in D-032 (peer, or the last `X-Forwarded-For` entry from the trusted proxy); refusals are 429 `rate_limited` with `Retry-After`; every request counts, a 404 included; in memory, pruned like the other windows. E-5 `GET /apk/latest` and E-4 `GET /i/{code}` stay on the 60 per minute. Landed `96e4cbd` (unit test and an integration test in `tests/core_api/rate_limit.rs`). Owner: "APK download limit assumed is OK". | Security review L-3: each E-6 answer can be a ~10 MB APK, so 60 per minute per IP let one client pull about 600 MB a minute from the Pi; E-5 and E-4 answers are small and the phone checks E-5 at most every 6 hours (AC-16.2). | SPEC §8.6, §10.6, §14.3.4 (E-6), §11.4 |
| D-120 | 2026-10-19 | TLS terminates in the owner's Nginx Proxy Manager (NPM) on the Home Assistant Pi instead of Caddy and cert-sync, which are removed (`infra/cert-sync/**`, `infra/pi/Caddyfile`, images, volumes `certs` and `caddy_data`, secrets `ha_ssh_key` and `ha_known_hosts`, `.env` keys `HA_PORT`, `HA_USER`, `HA_CERT_DIR`). (1) Kvitt gets its own public host name `KVITT_PUBLIC_HOST` (owner-chosen, recorded in HANDOFF; a DuckDNS name is being set up), because `icehack.asuscomm.com` is NPM's proxy host for Home Assistant and NPM selects by name. `KVITT_PUBLIC_URL=https://${KVITT_PUBLIC_HOST}:8443`. A path prefix on another name is not supported. (2) The router forwards external TCP 8443 to the Home Assistant Pi's 443 (NPM); nothing is forwarded to the Kvitt Pi. NPM issues the name's Let's Encrypt certificate via HTTP-01 on port 80. (3) The Kvitt Pi runs `server` (publishes 8080 on all LAN interfaces, hardening unchanged) and `backup` (unchanged, D-105). NPM forwards plain HTTP over the LAN (accepted by the owner). (4) `KVITT_TRUSTED_PROXY=${HA_HOST}`, NPM's LAN IPv4 address, checked by `install.sh`; the server is unchanged and takes the last `X-Forwarded-For` entry from that peer. (5) The NPM proxy-host settings and the Custom Nginx Configuration of §10.5.2 carry over D-117's headers and log redaction: headers at server level with the server's copies hidden; its own `location /` without `add_header` (NPM's default adds `X-Served-By`, which makes nginx drop server-level headers); `location ^~ /i/` with `access_log off` and `error_log /dev/null crit` because NPM's access log writes `$request_uri` with the query string; `client_max_body_size 6m`; `proxy_read_timeout 60s`; HSTS on (NPM's 2 years). (6) Health checks: `http://localhost:8080/health` on the Pi, then through NPM (warning only); `install.sh` removes the old stack's leftovers. (7) Gitea login: the secret is interpolated in the `run` line (`printf '%s' "${{ secrets.kvitt_cicd }}" \| docker login …`), the owner's proven form, because the env-based form came through empty on act_runner 0.6.1 with a user-level secret; this reverses L-8's 'never expanded inside a shell line'; act_runner masks the value. Revises D-103 (bundle files), D-104 (images, Caddy pin, cert-sync packages, trigger path), D-106 (cert-sync part), D-117 parts 1, 2, 3, 7, 8, D-023's cert-sync secrets. Verified on the Mac 2026-10-19: nginx -t against NPM v2.16.0 templates; rehearsal server-behind-nginx 21 checks passed (headers once each, HSTS, http→https 301, 413 at 7 MB, rate limit per forwarded client, forged X-Forwarded-For contained, invite code/name/`/i/` absent from nginx logs including on 502). | Owner 2026-10-19: NPM already manages Let's Encrypt; the HA name is taken; 8443 → NPM 443 confirmed. | SPEC §8.5, §8.7, §8.9, §10 intro, §10.1, §10.2, §10.3, §10.4, §10.5, §10.7, §10.10, §11.2, §14.7 |
| D-121 | 2026-10-19 | The Android app's public host is a build property. `kvitt.publicHost` is read from the Gradle property (`-Pkvitt.publicHost=…`), else the key `kvitt.publicHost` in `android/local.properties` (owner's, git-ignored), else the placeholder `kvitt.example.invalid` (RFC 2606; never resolves, so a release built without it cannot reach a real host). The build refuses a value that is not a host name. `:app` exposes it as `BuildConfig.PUBLIC_HOST`, as the manifest placeholder `publicHost` (the AC-2.6 `https` intent filter, port 8443, `/i/`, `autoVerify="false"`) and as `RELEASE_SERVER_URL` = `https://<host>:8443` (`link/ReleaseServer.kt`). `:feature:onboarding` receives that URL through the Hilt qualifier `@ReleaseServerUrl` for S-1's prefill and its non-default-server check. Invite-link parsing still accepts a server named by the link (AC-2.6). Tests and goldens use fixed fixture hosts. `infra/publish-apk.sh` passes `-Pkvitt.publicHost` from `KVITT_PUBLIC_HOST` (follow-up agent running). The pinned-root config's domain (D-082) is still the old literal and must follow the property if that fallback is ever prepared. Landed `d090c6c`. | D-120 gave Kvitt its own host name; it was hard-coded in three places. | SPEC AC-2.6, AC-2.7, §8.5, §10.5.3, §10.6, §13.3 S-1, §12 OQ-6; `android/README.md` |
+10
View File
@@ -32,6 +32,8 @@ Location: `infra/cert-sync/sync.sh:36-42`; SPEC §10.5.1 ("variant B: added as a
Issue: cert-sync holds `/secrets/ha_ssh_key`. With variant B there is no `restrict,command=` on the HA side, so whoever holds the key (a compromised cert-sync container, or the key copied around on the Mac and the Pi) gets an unrestricted root shell on the Home Assistant Pi, which also owns ports 80/443 and the house's automation. The script's own fixed `tar` command does not restrict the server side.
Smallest fix (infra + owner): make variant A the only accepted variant (`restrict,command="tar -C <dir> -cf - fullchain.pem privkey.pem"`, plus `from="<Kvitt Pi LAN address>"`), and if the add-on cannot do that, use a dedicated non-root user that can read `/ssl` only. Record the variant in `infra/README.md` as §10.5.1 already asks.
> **Note 2026-10-19 (D-120, SPEC v0.21):** moot. Caddy and cert-sync are removed; TLS terminates in the owner's Nginx Proxy Manager on the Home Assistant Pi, and no SSH key for the Home Assistant Pi exists in the stack any more (`install.sh` deletes `ha_ssh_key` and `ha_known_hosts` left on the Kvitt Pi). Closed.
**M-3. Picture decoding can exceed the server's memory limit (8000 × 8000 px vs 160 MiB).**
Location: `server/crates/kvitt-server/src/api/pictures.rs:39` (`MAX_DIMENSION`), `:66-92` (`process`), `infra/pi/compose.yaml` (`mem_limit: 160m`); limit from SPEC AC-4.4.
Issue: the dimension check happens before decoding (good), but an 8000 × 8000 image is allowed: decoding to RGBA is 256 MB, plus `to_rgba8`/`to_rgb8` copies, on a container limited to 160 MiB. A 5 MiB PNG or WebP easily encodes that size (a flat colour compresses to a few KB). Only an authenticated device can call `PUT /me/picture`, so in this threat model the attacker is a lost phone with a live token (until revoked) or a bug in the app, not a stranger; the effect is a crash and restart of the server (and of the sync for both phones) on each upload.
@@ -44,11 +46,15 @@ Location: `infra/pi/Caddyfile:11-14`; SPEC §8.7 (the Rust server logs the route
Issue: Caddy logs `request.uri` in full: `/i/<CODE>?from=<name>` and the `?from=` value, together with the client IP and User-Agent, into the json-file logs on the Pi (10 MB × 3). Codes live 72 h and are single-use, so the risk is low (an attacker with Pi log access is out of scope), but it is personal data and a secret in a log that §8.7 says must not hold either.
Fix (infra): add `log_skip` (Caddy ≥ 2.8) for the `/i/*` matcher, or `log { format filter { request>uri delete } ... }` — one line in the Caddyfile; or drop the access log entirely (the server already logs every request).
> **Note 2026-10-19 (D-120, SPEC v0.21):** Caddy is gone; the redaction moved to NPM's Custom Nginx Configuration (SPEC §10.5.2): `location ^~ /i/` has `access_log off` and `error_log /dev/null crit`, so neither the code nor the `?from=` name reaches NPM's logs, a 502 included (rehearsed on the Mac). Other paths are logged with their query strings, which hold no personal content; NPM logs no request headers.
**L-2. The Caddy container is not hardened like the other three.**
Location: `infra/pi/compose.yaml` (`caddy` service: no `user`, no `cap_drop`, no `read_only`); SPEC §8.9 hardens only the server.
Issue: Caddy runs as root with every default capability and a writable root filesystem, and its admin API is bound to `0.0.0.0:2019` for every container on the `kvitt` network (only cert-sync needs it). This is the process that terminates TLS from the internet.
Fix (infra): `cap_drop: [ALL]`, `cap_add: [NET_BIND_SERVICE]` (or none: 8443 is unprivileged), `read_only: true` with `tmpfs: [/tmp]` and `caddy_data`/a `caddy_config` volume writable; keep `no-new-privileges`. Admin on `0.0.0.0` is required for cert-sync; acceptable on the internal network.
> **Note 2026-10-19 (D-120, SPEC v0.21):** moot. There is no Caddy container; the TLS-terminating process is the owner's existing NPM install on the Home Assistant Pi, outside this repository. On the Kvitt Pi only `server` publishes a port (8080 on the LAN), with the hardening of SPEC §8.9.
**L-3. APK download is unauthenticated and 11 MB per request at 60 requests per minute per IP.**
Location: `server/crates/kvitt-server/src/api/apk.rs:76-101`, `src/api/rate_limit.rs:56-63`; SPEC §8.6, §10.6.
Issue: the only large unauthenticated response. One IP may pull about 660 MB/min, several IPs more, saturating the home uplink (and the Pi's SD card) for as long as they like. The file is public by design (the landing page links it).
@@ -79,6 +85,8 @@ Location: `.gitea/workflows/build-images.yaml` ("Log in to Gitea Container Regis
Issue: `${{ }}` is substituted into the script text before the shell runs it: a token containing a quote or `$` breaks the line (and could inject), and the value appears in the step's process arguments on the runner. The token also needs `write:repository` (for the tag push) on top of `write:package`.
Fix (infra): pass it as `env: { REGISTRY_TOKEN: ${{ secrets.kvitt_cicd }} }` and use `printf '%s' "$REGISTRY_TOKEN" | docker login …`; keep the scopes minimal (the tag push can use the checkout token if Gitea provides one).
> **Note 2026-10-19 (D-120, SPEC v0.21):** the fix of `e70fe0d` is reversed (`4f69ae0`). The env-based form came through empty on act_runner 0.6.1 with a user-level secret ("password is empty" on the first run), so the login step interpolates the secret in the `run` line again, as `printf '%s' "${{ secrets.kvitt_cicd }}" | docker login … --password-stdin` (the owner's proven form); act_runner masks the value in the log. The quoting risk remains for a token containing `"` or `$`; Gitea tokens are hex. The scope stays `write:package` only.
**L-9. Images are pulled by tag over plain HTTP from the LAN registry.**
Location: `infra/pi/compose.yaml` (`${KVITT_REGISTRY}/kvitt-server:${KVITT_TAG}`), `infra/pi/install.sh` (`insecure-registries`), `infra/pi/deploy.sh`; SPEC §10.3, D-001/D-104 (LAN is trusted).
Issue: anyone on the home Wi-Fi who can answer for `gitea.lan` (or the Gitea Pi itself, if compromised) can hand the Kvitt Pi a tampered server image that sees every token hash and all data. Accepted by the spec's trust in the LAN; noted because the server is the one component that holds everything.
@@ -101,6 +109,8 @@ Fix (server): return 404 in both cases, or leave as is.
**I-5.** Client-IP correctness to verify after deploy: the rate limiter trusts the last `X-Forwarded-For` entry from Caddy (`rate_limit.rs:203-227`). If Docker's userland proxy ever fronts port 8443 (IPv6 without `ip6tables`, or `userland-proxy` quirks), Caddy sees the bridge gateway for every client and all internet clients share one 60/min bucket and one 10-failed-accepts lockout. Check once in Caddy's log that `request.remote_ip` is a public address.
> **Note 2026-10-19 (D-120, SPEC v0.21):** the trusted proxy is now NPM (`KVITT_TRUSTED_PROXY` = `HA_HOST`, its LAN IPv4 address) and the server publishes 8080 on the LAN; the one-off check is a `conntrack -L -p tcp --orig-port-dst 8080` on the Kvitt Pi showing `HA_HOST` as the source and no `172.*` address in the reply half (SPEC §10.10), instead of Caddy's `remote_ip`.
**I-6.** `bootstrap` and `pair-device` print the invite code to stdout (`src/cli/bootstrap.rs:41-62`). Through `docker compose run --rm` the line also sits in that one-off container's json log until the container is removed. Expected; the owner reads it and it expires in 72 h.
**I-7.** The Ktor/OkHttp client keeps default redirect handling (`data/api/KvittApi.kt:246-248`). Neither the server nor Caddy redirects, and TLS excludes a third party, so the bearer cannot be bounced to another host. Fine.